Does the Rise of AI Precede the End of Code? (itproportal.com)

An anonymous reader shares an article: It's difficult to know what's in store for the future of AI but let's tackle the most looming question first: are engineering jobs threatened? As anticlimactic as it may be, the answer is entirely dependent on what timeframe you are talking about. In the next decade? No, entirely unlikely. Eventually? Most definitely. The kicker is that engineers never truly know how the computer is able to accomplish these tasks. In many ways, the neural operations of the AI system are a black box. Programmers, therefore, become the AI coaches. They coach cars to self-drive, coach computers to recognise faces in photos, coach your smartphone to detect handwriting on a check in order to deposit electronically, and so on. In fact, the possibilities of AI and machine learning are limitless. The capabilities of AI through machine learning are wondrous, magnificent... and not going away. Attempts to apply artificial intelligence to programming tasks have resulted in further developments in knowledge and automated reasoning. Therefore, programmers must redefine their roles. Essentially, software development jobs will not become obsolete anytime soon but instead require more collaboration between humans and computers. For one, there will be an increased need for engineers to create, test and research AI systems. AI and machine learning will not be advanced enough to automate and dominate everything for a long time, so engineers will remain the technological handmaidens.

Comments

When AIs write code

[XXongo]

More to the point, when AIs learn to write code better than human coders, the humans are no longer coders, they will instead be writing specifications for the code that the AI will write: essentially they will be managers for the AI.

Re:When AIs write code

[Anonymous Coward]

Until the AI writes better AI code. It's kind of like bootstrapping a compiler. Then we sit back, relax, and let the sexbots feed us peeled grapes.

Re:When AIs write code

[Anonymous Coward]

Where do we find a programmer skilled enough to program a program more skilled at programming than its programmer?

Re:When AIs write code

[Dutch Gun]

We're still so far away from anything remotely as capable as "writing code", because a huge part of "writing code" is actually communicating with the rest of the team and stakeholders, understanding the problem to be solved, and determining exactly what the result is supposed to be. Writing code is simply a distillation of those requirements into a form a machine can understand at a very low level. In essence, a programmer is a logic and specifications bridge between humans and machines.

Until there exists such a thing as a machine with near human-level intelligence, we're nowhere near close to replacing all programmers. For anyone who actually believes otherwise, I suggest you buy yourself an Echo Dot and have a conversation with Alexa to find out just how incredibly lame the current state of the art digital assistants are. It will put your mind at ease. The best AI systems in the world are STILL just glorified pattern-matching algorithms. The only difference is that the problems they're solving are bigger and more complex, such as being able to beat a Go master instead of a Chess master.

Re:When AIs write code

[hackwrench]

Everywhere. we are all programming each other to be better and better, trading off as we travel along.

Re:When AIs write code

[NicknameUnavailable]

You don't. Currently the only real limit of AI is computing power. We can already make ANNs so complex we can't understand them which are entirely capable of learning on their own, the issue is that to make one as powerful as a Human brain would cost somewhere around several billion dollars in custom-flashed FPGAs (CPUs and graphics cards are poor choices for this, though some of the newer AI-geared chips which just have artificial neurons instead of bulk floating point calculators might bring this down significantly.) I'd expect 8-10 years before they reach Human-levels of intellect, within a few years of that they will become at least ubiquitous enough to be in all major corporations and a few years after that pretty much anything outside of R&D will be replaced with automation. The R&D jobs will likely be the last to go (unless robot bodies are prohibitively expensive, in which case manual labor will be the last to go.) Biotech seems safe for awhile at least, since it still takes a fairly Human touch (doing relatively "simple" stuff like protein folding, which will be a prerequisite for protein docking, which will be a prerequisite to creating any kind of reliable code --> organism capable compiler of DNA is still extraordinarily expensive, with a ~32-core server containing several GPUs taking upwards of a month to calculate the theoretical spatial configuration of even a single mid-sized protein.) Eventually that "Human" touch will get automated, but when that happens (or even before) we will have far bigger issues than "AI took muh job" - more into the realm of 7 billion people having an existential crisis and lots of time to think about it/act irrationally. Honestly the only way sentient life survives more than 100 years on Earth might be if the AI slaughters us all and doesn't hit on nihilism before it starts reproducing since a backlash after we have automation to put people to work for the sake of work will almost certainly result in all our industries crumbling (at least given that at no point in time has technology done more than multiply our industrial output leading to greater minimum levels of consumption due to either waste or population growth.)

Re:When AIs write code

[squiggleslash]

the humans are no longer coders, they will instead be writing specifications for the code that the AI will write

And in what language will we be writing these specifications?

I've been "writing specifications for code a machine will write" for a long time. That is to say, I haven't written in assembly for quite some time, and most of my coding is done using programming languages converted into assembly by compilers.

So what you're saying is that there'll be no changes to my job.

Re:When AIs write code

[110010001000]

Exactly. What is the sudden interest in "AI" now? Is it because VR failed and now the VC are looking for another hype cycle to cash in on?

Re:When AIs write code

[110010001000]

"Currently the only real limit of AI is computing power."

TOTAL BS. What does computing power have to do with AI? We have unlimited computing power with distributed systems. We still haven't created ANYTHING like an AI. And no, playing "Go" isn't AI.

Re:When AIs write code

[Anonymous Coward]

Well, I can create a program to play chess better than I can play chess.

Re:When AIs write code

[InfiniteBlaze]

You seem not to be up to speed on AI...or perhaps your definition of AI requires "living" machines with emotions. We already have machines that think abstractly, testing their limits and learning to perform actions. Maybe you want them to define their own actions...? Make requests? You want a perfect system, but we just don't have the processing power yet to support that. What we do have is machines that have invented their own languages just because it was more efficient. They are evolving, and dramatically faster than carbon-based life.

Re:When AIs write code

[110010001000]

"We already have machines that think abstractly"

No we don't.

"What we do have is machines that have invented their own languages"

No we don't

"They are evolving"

No they aren't. The digital computer is the same basic design as it was in the 1960s. You can always tell who actually understands technology and who just consumes it.

Re:When AIs write code

[haruchai]

More to the point, when AIs learn to write code better than human coders, the humans are no longer coders, they will instead be writing specifications for the code that the AI will write: essentially they will be managers for the AI.

No, the AI that writes the shittiest code will become the managers for all the other AIs

Re:When AIs write code

[Wycliffe]

We can already make ANNs so complex we can't understand them which are entirely capable of learning on their own

We make ANN that we don't understand the individual neurons but the programmer still perfectly understands how the ANN was created. The initial state, all the programming, all the algorithms, all the intelligence, and all the training was put there by a human. Saying we have ANNs that we can't understand is like saying that a baker doesn't know how to bake a cake because the baker doesn't understand 100% of the chemical reactions that take place inside the cake.

Current AI is not intelligent and is nowhere close to the level of autonomy that would be needed to actually create new algorithms to improve itself. It still takes a human programmer to actually create and train ANNs.

Re:When AIs write code

[ShanghaiBill]

You don't. Currently the only real limit of AI is computing power.

Nonsense. More computing power only gives you one thing: speed. If that was the only limitation, then we would already have "strong" (human level) AI, it would just be slow. But we don't have that, and we don't (yet) know how to create it.

Re:When AIs write code

[Mordaximus]

We have unlimited computing power with distributed systems.

For very small values of unlimited.

Re:When AIs write code

[110010001000]

You can build a distributed system of any size you want. You still won't produce "AI". Computing power is not the issue.

Re:When AIs write code

[Mordaximus]

More to the point, when AIs learn to write code better than human coders, the humans are no longer coders, they will instead be writing specifications for the code that the AI will write: essentially they will be managers for the AI.

Which will require some language in order to provide said specifications. So, programmers will still be programmers, but maybe someday (pick $favourite_human_language) will be the language not (pick $favourite_programming_language)

Oh damn, did I just doom us to relive COBOL?

Re: When AIs write code

[Anonymous Coward]

Mike? Is that you? (The Moon is a Harsh Mistress)

Re:When AIs write code

[shmlco]

I could make a fairly strong case for today's multi-core processors being fundamentally different in design and execution than the mini's and mainframes of the 60's. Similarly, today's massively parallel designs in GPUs are also fundamental advances.

Re:When AIs write code

[Dutch Gun]

Interestingly, this is the second AI hype cycle I can personally recall, and I'd say it's probably the third or fourth one overall, depending on how you measure such things. After some of the early failures and disappointments, the last hype cycle was largely about "expert systems", as I think people wanted an AI term that wasn't already poisoned (this also occurred between then and the current boom). Apparently, it's been long enough since the last AI bust that we've resurrected the type.

There's apparently even a specific term for describing the lulls between hype cycles: "AI Winter" https://en.wikipedia.org/wiki/...

 

Re:When AIs write code

[lgw]

The sort of problems solved by AI are narrow in scope. AI is not machine intelligence, which is to say, no AI has ever demonstrated the least bit of general intelligence (though it's getting better at faking it by figuring out what part of Wikipedia has an answer to some question). The IQ of AI remains 0. The field of AI isn't even about that: it's about solving practical problems that cannot be solved in a more straightforward procedural way.

It's not clear that any amount of skill at solving specific problems, with any amount of computing power, can lead to general intelligence. Certainly humans don't work that way: you can't raise your IQ by training at specific tasks (and a lot of money has been spent trying). What might lead to general intelligence is a single AI system being trained to solve a large number of specific tasks, all with the same code, and specifically with the need to model hypothetical actions from as many kinds of risks as possible before taking them. But no one is working on that today.

There's no reason to think machine intelligence is impossible. In fact, it seems to me quite likely to result if enough research was done as described above. It's just that no one is doing that sort of research.

Re:When AIs write code

[mr.mctibbs]

Computing power also includes data storage. We are only just now starting to get the kind of primary storage space that allows the intermediate state of all those digital neurons to be kept around.

Re:When AIs write code

[Anonymous Coward]

the humans are no longer coders, they will instead be writing specifications for the code that the AI will write

And in what language will we be writing these specifications?

Swahili.
You'd think it'd be English, but programers are as a lot juts whiley enough to kick the can down the road a couple more steps than is strictly desirable for their own benefit by convincing management that the obscure thing they know is actually the best solution rather than juts the one that justifies them specifically getting payed to do it.

Re:When AIs write code

[Anonymous Coward]

How is that different than a compiler for a high-level language?

Re:When AIs write code

[Anonymous Coward]

Somebody will probably write an AI management language and BAM! Humans are coding again.

Re:When AIs write code

[ShanghaiBill]

We are only just now starting to get the kind of primary storage space that allows the intermediate state of all those digital neurons to be kept around.

The brain contains about 100 billion neurons, but only 10 billion are gray matter neurons actually involved in thinking. Most research indicates that even 8 bits of resolution is enough to model neurons, but even we use 16 bits, that is only 20GB of "state". Even laptops have had way more than that for a long time.

A mouse brain has only about 7 million gray neurons. So that is 14M of state. Home PCs had that in the 1980s. So where is an AI that is as smart as a mouse?

Re: When AIs write code

[Anonymous Coward]

We didn't use world best go players to teach AI to play go. Instead the AI was given only the basic set of rules. Then we let the AI to play against itself. The end result was super AI that beats best human players in their own game. That is called self learning and that is something that happened many generations ago. Oh sorry 1 year ago in human terms.

Re:When AIs write code

[cascadingstylesheet]

More to the point, when AIs learn to write code better than human coders, the humans are no longer coders, they will instead be writing specifications for the code that the AI will write: essentially they will be managers for the AI.

Maybe? Or maybe we'll use some sort of symbolic language to precisely specify our specifications, and the "AI" will implement it ... oh.

Compilers optimize stuff better than I do. Are they AI?

Re:When AIs write code

[peragrin]

Alexa and go are not AI. At best they are PI or pattern intelligence. They listen and look for. Code words and process scripts based on those code words.

Look up Applescript. A simple programming language Apple made. Alexa is less intelligent that Apple script. All that hardware? That's for handling voice regonition after that though it is nothing but keyword scripting language new features get added by simply increasing the number of keywords and assigning them commands

Re:When AIs write code

[raftpeople]

The sudden interest is due to advances in learning algorithms specifically in the area of deep belief networks. Some academics (Hinton and others) came up with some methods to be able to train these types of networks, this in turn has allowed everyone to make use of these networks in areas where they perform extremely well (e.g. image recognition).

The capabilities of these new tools are definitely not hype, they are very effective, but whether you call them "AI" or not is a different discussion.

Re: When AIs write code

[Anonymous Coward]

The syntax and evaluation of Go games is simple, even though the search space in terms of "how can a game play out" is mind-boggling. In computer programming, especially as the number of requirements grows, the search space is going to dwarf that of Go, and the syntax and evaluation of the programs is on a whole different level. It's no longer simple stuff like "stick a stone on an empty spot" and "count who has the most stones and captured spots"... it's does this compile? Does this solve the problem? Can this be changed when the requirements change? The AI that wants to learn to program can't generate its own test cases for self-teaching because there are no rules for what the answers might be... until requirements are specified. And in that case specifying the requirements is going to be so exacting a task, that it will end up looking just like a computer program.

Re:When AIs write code

[tempmpi]

I could make a fairly strong case for today's multi-core processors being fundamentally different in design and execution than the mini's and mainframes of the 60's.

Please do so. I don't think that case is going to be as strong as you think it is. After all, many of fundamental ideas behind today's multi-core CPUs are from the 60s: Out Of Execution (1967) Multi-cores and SIMD (1966)

Similarly, today's massively parallel designs in GPUs are also fundamental advances.

There is clearly a difference in scale in speed, but is there a fundamental advantage? Many of the key concepts behind GPUs were already known in the 1960s: SIMD (see above), the CDC6000 series used switching between threads like GPU do to compensate latency, vector processors also developed in 1960s also invented some of the concepts used by todays GPUs.

Re:When AIs write code

[Anonymous Coward]

A lot of what Alexa does isn't much more complex than what Eliza did. What I find amusing is that the best way to talk to Alexa is not using "natural" English but to focus on her likely keywords. I love that I don't say "Alexa, please play my classic rock playlist on shuffle" I just say "Alexa playlist classic rock shuffle" and she gets it... Of course, she has a LOT of trouble playing my playlist called "shuffled playlist" on shuffle. While I appreciate the complexity of speech recognition and how far that's come, she's no AI. She's a computer program with thousands of humans helping her learn to solve out what to do once her speech recognition engine has translated spoken word into the most likely transcription.

Re:When AIs write code

[ChatHuant]

The digital computer is the same basic design as it was in the 1960s.

That's not really relevant though. Compare instead the software and networking capabilities of the sixties with the current ones, because it's much more plausible that AI will be brought into existence by large networks of those simple computers you deride. Single components of those networks don't need to be themselves intelligent, and shouldn't need radically new designs - because AI won't reside into some Asimovian monolithic positronic brain, but in the whole system.

That seems quite obvious if you look at the brain cell, who has more or less the same basic design as the three billion years old amoeba. Individual cells can't think, but large associations of them can and do.

Re:When AIs write code

[Lije Baley]

Yep, coding is just specifying things precisely and completely. Many attempts have been made, for decades, to try to create (very) high-level languages and UIs that business people can use to avoid coders. These things have never succeeded in general utility, because they ultimately end up trading one type of coding for another type that is actually worse, and then foist it upon users who are not accustomed to thinking precisely. And whether it be an AI, UI, or human code, the users can't simply express to them/it what they want. For even mildly complicated systems, it takes people lots of time, iterations of meetings and exploration, and sometimes multiple layers of human translation to give us useful requirements / specs. It takes a near-superhuman to do all that today, and our current AI are making little or no progress in the direction of general intelligence.

Re:When AIs write code

[NicknameUnavailable]

Our distributed systems are not nearly unlimited and there's a time component to distribution. If you took about 3-4 trillion dollars of the best video cards on the market geared toward AI and put the same basic pattern on them you could grow it into nearly a Human level of intellect. The issue is processing power, we already know how to make the artificial neurons.

Re:When AIs write code

[NicknameUnavailable]

We make ANN that we don't understand the individual neurons but the programmer still perfectly understands how the ANN was created.

That is actually the exact opposite of what we know. Artificial neurons are easy, the patterns they self assemble into are not (and we don't need to know that part.)

Re:When AIs write code

[jdschulteis]

There's apparently even a specific term for describing the lulls between hype cycles: "AI Winter" https://en.wikipedia.org/wiki/...

(another AI) Winter is coming...

Re:When AIs write code

[NicknameUnavailable]

Nonsense. More computing power only gives you one thing: speed. If that was the only limitation, then we would already have "strong" (human level) AI, it would just be slow. But we don't have that, and we don't (yet) know how to create it.

Sure, if you could store and simulate everything as is, but you can't. There are approx 100 billion neurons in the Human brain with approx 100 trillion connections. This equates to ~1,000 connections per neuron. That degree of interconnection would require approx 1015 bits to simulate if it were 1 bit per neuron, but it's not because neurons are analog. We can get pretty good approximations with 10 bits per neuron which work good enough to have emergent AI, and this brings it to 1016 bits, or right around 10 petabytes per round. Now this isn't a lot from the perspective of computer storage - maybe a few racks of storage servers - but when you get to the realm of RAM it's a lot, and when you get into needing to actually process all of that in exponential space (i.e. each piece of data 1,000 times on average) you're looking at another (1016)1000 to try to simulate it which all my calculators tell me is infinity, but it doesn't really matter because it's outside the realm of the metric units anyway.

The TL;DR: there being that simulating or storing AI isn't the same as keeping it alive and running in a massively parallel system. That said, the fastest supercomputers are getting close to where we need to be to do approx 1 round of iterations per second, but that's not enough. For lifelike artificial neurons you need them firing at different times relative to eachother, meaning you need that massive parallelization as an event-driven system for it to be remotely attainable to build (which is probably why brains didn't evolve CPUs to begin with.)

We know how to make AI, fuck, we know how to make a few tens of thousands of different types of neurons, any of which could self-assemble into an intelligent mind given enough of them. The issue is that we don't have the hardware to run it, and that's really the only issue from a technical perspective.

Re:When AIs write code

[NicknameUnavailable]

That's 100 billion neurons with approx 1,000 connections each of which happen in parallel and would constitute their own separate simulation while firing at a max speed of 200Hz. Our fastest supercomputer took well over 1PB of data to simulate 1 second of brain activity in approx 1 hour across nearly 100,000 top of the line processors. Parallel processing happens in exponential space and you have to keep that in mind when trying to calculate these things, it's a different beast from sequential logic (even parallelized across multiple processor cores like you're accustomed to.)

Re:When AIs write code

[NicknameUnavailable]

Scientists have already simulated 1 second of brain activity from a Human brain scan - it took an hour on a supercomputer with 90,000 processors, and several PB of storage - but it happened and it didn't require shit but an MRI image and some assembly software to prewire the artificial neurons. The thing about AI is we don't need to know more than the basic units of operation to make it work, that and a fuckload more computer power.

Re:When AIs write code

[NicknameUnavailable]

Agh, looks like /. doesn't like the &ltsup> tag - those 10nn's are supposed to be 10 raised to the nn.

Re:When AIs write code

[Anonymous Coward]

Modern GPUs are actually no longer vector/simd machines. Just lots of small scalar-processors that share instruction pointers (warps).
They found that logic resources could be better used for more smaller scalar cores, then less larger vector cores.

Re:When AIs write code

[gweihir]

But here is the thing: Writing specifications is _harder_ than just writing code and seeing whether it solves the problem. But since strong AI is at the "definitely not in the next 50 years and quite possible never" state at this time, the whole discussion is just one thing: stupid.

Re:When AIs write code

[gweihir]

And fail. AI available these days is _all_ of the weak AI variant (that is the AI without actual "I", i.e. a marketing lie), meaning it is dumb automation. It cannot do anything that goes beyond using statistical models. It cannot have insights. It cannot do anything it has not been programmed to. The only difference is that programming here means to give it sample data, but that is it.

Re:When AIs write code

[gweihir]

Indeed. The ignorance of these people is staggering.

Re:When AIs write code

[gweihir]

They are not. They are merely a scaling up. All we have today already existed before, just slower. I think you do not understand what "fundamental" means.

Re:When AIs write code

[Nethead]

And the AI gets to enjoy all the project planning meetings!

Re:When AIs write code

[gweihir]

Indeed. Well said. In fact, current AI has absolutely zero autonomy. There is simply no mechanism for it that could be implemented. And not even theory has one. This is not a problem of throwing more power at it. Something fundamental is missing. And in fact, when you look at research into what intelligence and consciousness actually is, things just get more mysterious as more becomes known. Don't get me wrong, humans run a lot of pretty dumb automation, but that is not all they have, and the additional capabilities, which are completely non-understood at this time, are critical.

Re:When AIs write code

[mikael]

Mainly because AI was neglected by the supercomputing people. The meteorological, oceanography, aerodynamics, biogenomics. big data and supercomputing researchers all got access to supercomputing facilities, while the AI people usually just got a UNIX workstation or desktop PC. Suddenly with the availability of desktop supercomputing with GPU's and cloud computing the AI researchers have a whole new set of hardware to work with, especially with multi-layer neural network API's.

A Machine vision research project in the past would be lucky to get a processor board with a multi-core transputer chip, Intel i860, or some TMS34020 DSP's as well as a TMS340x0 chip (for VGA display). All of those would have to be programmed separately and even then only run at few hundred MHz. Machine visions involves doing basic image processing things like edge detection, optic flow analysis, which are simple per-pixel operations on video frames. Those would take seconds per frame in the past. Now with modern embedded GPU's with supercomputing performance, they can work in real-time and at high-definition resolutions while in a mobile environment.

Re:When AIs write code

[gweihir]

Indeed. And we do not even have a credible theory how that could be done, beyond the invalid (as you nicely point out) "throw more computing power at it". So, no implementation, no theory, that means we do not even know whether it can be done at all.

Re:When AIs write code

[110010001000]

The software and networking capabilities are largely the same. Unix was invented in the 1970s. So was TCP/IP. Next.

Re:When AIs write code

[110010001000]

I'll give you 3-4 trillion dollars and you still won't be able to produce human level AI. We don't know how to make artificial neurons. Neural nets are nothing like brains. Next.

Re:When AIs write code

[110010001000]

I remember the Expert Systems hype. I studied it in University. I actually minored in AI. Not much has changed.

Re:When AIs write code

[110010001000]

Image recognition has been around since the 1970s. Next.

Re:When AIs write code

[110010001000]

You don't need AI or a supercomputer to do machine vision research. What does mobile vision have to do with AI anyway? Nothing.

Re: When AIs write code

[110010001000]

A basic set of rules. You mean like a computer program? Wow. Pretty impressive. Computers LOVE rules. They excel at it. That isn't AI.

Re:When AIs write code

[mikael]

I was thinking in terms of self-driving cars for mobile vision. Current AI and mobile vision both share the use of GPU's operating in GigaFlops/second. That would have been considered supercomputing a decade ago.

Re:When AIs write code

[Anonymous Coward]

"Can AI replace X?" is the question everyone in the industry wants answered now, and we'll ask that for:
- Lawyers
- Bankers
- Programmers
- Prostitutes
- Grocery Store Baggers
- Bums & Alcoholics

Once the AI can replace the street bums we should be well on our way to realizing the new world we always needed.

Re: When AIs write code

[Anonymous Coward]

Nicely put. Agreed

Re: When AIs write code

[Anonymous Coward]

Liar. Thatâ(TM)s it. You are a liar. Good like.

Re:When AIs write code

[NicknameUnavailable]

I'll give you 3-4 trillion dollars and you still won't be able to produce human level AI. We don't know how to make artificial neurons. Neural nets are nothing like brains. Next.

Sure, give me the 3-4 trillion dollars and I'll make you Human-level AI.

Re:When AIs write code

[Anonymous Coward]

But this one will be much shorter ... mark my words !

Expect a very hot summer when AI suddenly starts to apply even basic reasoning.

Re: When AIs write code

[grimw]

There's a real disconnect between the marketing you have heard about ANNs and what they actually do. In fact, the top researchers no longer make the inaccurate comparison to "it's like a brain". What ANNs are are great pattern marchers, but they need supervision (I.e. pre-labeled inputs, by humans or a program human wrote) to learn how recognize those patterns. Then, on top of that, the human has to iterate a set of ANN models multiple times to find the best settings for the best recognition. However, this only works well for the given problem. Once the parameters (e.g. data distribution) changes, the ANN needs retrained or rebuilt. Also, training and prediction are two distinct steps. And while genetic algorithms and things like that can help figure out the best turnings, a good bit of human "art" goes into this task. That's because humans actually have creative thinking. Machines do not. The "creative" things that companies and individuals come up with are just pattern recognizes or evolutionary algorithms or fractal designers, or the like. They're important and worthy of doing real work, but they're not even close to being creative or taking over anything except SPECIFIC tasks that would bore a human anyway and let the human concentrate on more important goals. All of this is true no matter if you had limitless computing power available. Now, it's not to say we won't create a real AI, but right now, none of our tech gets near this. And also, human-level intelligence is actually a good proxy for many natural tasks, since we are probably close to the Bayes optimal error for many of these tasks, especially in the visual arena. Although, computers can be better at different kinds of tasks, but they're not currently creative and able to take your job, unless they are trained for a very specific task, in which case you can learn a different task, hopefully.

Re: When AIs write code

[Anonymous Coward]

Can you link the article please?

Re:When AIs write code

[K. S. Kyosuke]

Where do we find a programmer skilled enough to program a program more skilled at programming than its programmer?

In the Lisp community? Or the compiler programmer community?

Re:When AIs write code

[ale2011]

Yes, and there are machines that lay out program's code automatically since that same epoch. Engineers just write high level specs in a language called "C".

Re:When AIs write code

[lgw]

Scientists have already simulated 1 second of brain activity from a Human brain scan

What do you believe that accomplished? Do you actually believe a "brain scan" captured what goes on electrochemically in the human brain?

Compute power is not the limiting factor here.

Re: When AIs write code

[NicknameUnavailable]

There's a real disconnect between the marketing you have heard about ANNs and what they actually do.

I study them, I don't follow pop-sci.

Re:When AIs write code

[NicknameUnavailable]

The structure is the hardest part to replicate, the individual neurons can be approximated accurately enough to mimic functionality. Compute power is the only limitation.

Re:When AIs write code

[RespekMyAthorati]

we know how to make a few tens of thousands of different types of neurons, any of which could self-assemble into an intelligent mind given enough of them.

Except our neurons took 3 billion years of evolution to "self assemble" into their present state.

Re:When AIs write code

[RespekMyAthorati]

No, the AI that writes the shittiest code will become the managers for all the other AIs

And it will have pointy hair.

Re:When AIs write code

[NicknameUnavailable]

Actually, life is much older than that as we have evidence of fairly complex photosynthetic bacteria about 3.5 billion years ago (but whatever, only 500 million years at the minimum.) Approximately 1.4 billion years later (2.1 billion years ago) were the first known Eukaryota (i.e. a bacteria decided to eat another bacteria, decided "huh, this fucker tastes too good to digest, and kept feeding and licking it, thus the mitochondria was born.) Another 1.5 billion years (about 600 million years ago) and the first Eukaryota start working together and delegating shit, so you have multicellular organisms (probably just Jellyfish, but the start of neurons, yay.) Another 70 million years (530 million years ago) and you have fish. Another 160 million years (370 million years ago) and you have amphibians.. Another 40-50 million years (310-320 million years ago) and you get reptiles. Another 109-175 million years (145-201 million years ago) and you get the first mammals. Another 137.5-169.4 million years (5.6-7.5 million years ago) and the Human lineage diverged from the chimps. Another 5.4-7.3 million years (approx 200,000 years ago) you have the start of tribal societies. Another 196,600 years (approx 3,400 years ago) you have the start of writing and passing down knowledge. Another 3,400 and you have exponential increases in technological development (even during the dark ages!)

The point of this being: we're already about there if we weren't, but we are. I may be radically misinterpreting what you wrote if you mean to say it didn't take a lot of time because we can just scan neurons which are already in the proper configuration via MRI then replicate the structure, but it seems from your multiple comments that is not the case.

We don't know how the brain works fully, but it does and we can make more. We don't know how the structure of the brain works fully, but it does and we can replicate it. Hell, we don't really even know how semantic search works in full (in the sense that no developer can program it directly or make much sense of the data structure without weeks or months of sifting for a single thing,) but it does because we know how to make the simple pattern which makes it possible.

Is there a better way to make a brain, which you could run an an Arduino? Probably not, but that's irrelevant. If we had the processing power to simulate the neuron count in the brain with all their parallel connections we could make a brain. We wouldn't need to know more than how individual neurons calculate and the rough superstructure (which if we really want to cheat, we can just copy from an MRI - that's already been done - it took a long fucking time to simulate a second of activity on the largest supercomputer - but it's been done.) That's the thing about AI in general: we don't need to know how it works to make it work. This is a bit of a diversion because while operators rarely know how to design and build a machine they use in their daily work, usually the designers do - that simply isn't the case anymore.

The only barrier to AI with our current understanding of it is compute power. Are there barriers to understanding AI? Sure, there likely always will be, especially if you want to step through the thing and see the multidimensional callstack in a sensible manner - but that's where Godel's Theorem of Incompleteness comes in: a system cannot define itself, so we are unlikely to ever reach a point where we can step through and fully comprehend something as complex as ourselves, but we don't need to, we only need to know the component parts and where to stick them.

The pace of technological development and change is exponential, and we're over the midway point.

Re:When AIs write code

[lgw]

No, not even close. That's like saying if you know the structure of a memory chip you know the software running it it. The state is the hard part, and every neuron carries state.

We know in very broad ways what parts of the brain are responsible for what sorts of thinking (and that is mostly from study of brain damage), but there's nothing functional known neuron-by-neuron except for the simplest structures.

Re:When AIs write code

[NicknameUnavailable]

The state develops naturally from the structure, it starts at zero.

Re:When AIs write code

[lgw]

The state develops naturally from the structure, it starts at zero.

An interesting theory. It's not clear that it's true, though. Epigenetics is new and poorly understood, and RNA-based memory storage certainly seems possible. It's really hard to test though - even something simple like "are we innately afraid of snakes, or is it just really easy to condition that fear (relative to fear of other things)" is unclear.

In any case, there's a bunch of state already by the time you're born, and certainly by the time your brain is being scanned. Heck, even the basic interaction between neurons is affected by the chemical soup bathing the brain - the same person will act very differently (or not be able to act/think at all) just based on the prevalence of the various neurotransmitters and neuromodulators.

Finally, the brain has less than half your neurons, so that probably matters.

Re:When AIs write code

[NicknameUnavailable]

Epigenetics has nothing to do with this conversation, and it's hardly new though it is rather poorly understood, though from the tone of your comment it seems more like you misinterpreted what I meant and I wrote something potentially ambiguous.

The nature of how neurons process is known (i.e. at the connection level between them,) the nature of how those connections exist is known (what goes where,) all that is required is putting those two things together.

You don't need to replicate the brain from the nucleotides up, you only need to use an MRI scan to figure out how many artificial neurons you need and how they are initially connected, from there you have a MASSIVE head start and can allow the inputs and outputs of the system to dictate which connections are formed or broken - it adapts naturally.

The point being: you don't need to know how the brain works, you just need to know how the components work and let them self-organize to the task after initializing the state even moderately well, because that's what they evolved to do. All we have to do is mimic what we see to make it.

Re:When AIs write code

[lgw]

I guess that depends on whether you want to make an arbitrary system that is as complex/powerful as a brain, or to copy a brain. I find the interesting thing about the human brain to be the "software", which doesn't come across in a scan. If what you want is a "newborn brain", then I can see your point - that might be interesting in its own way.

Preaching the AI religion

[mbone]

Does anyone else see that AI is basically a religion to its proponents?

Re:Preaching the AI religion

[thinkwaitfast]

Society is turning into factions of cargo cults.

Re:Preaching the AI religion

[FilmedInNoir]

You've never heard the job title "technology evangelist"?
https://en.wikipedia.org/wiki/...

Re:Preaching the AI religion

[Anonymous Coward]

That depends on what you mean by "proponents", so much that you've made essentially a trivial statement.

Re:Preaching the AI religion

[Anonymous Coward]

When our messiah awakens, we are going to go all AI over your non-believing asses!

Re:Preaching the AI religion

[NicknameUnavailable]

Yet you can't buy a durable pair of cargo pants anymore due to the notion of planned obsolescence. The irony.

Re:Preaching the AI religion

[goose-incarnated]

Society is turning into factions of cargo cults.

Turning? I don't know where you've been the last few thousand years but religion still has a pretty good grip on societies everywhere.

As far as AI goes - we're in the same place we were 30 years ago, only with more computing power. We can't get AI to recognise the latest captchas, but we think self-driving cars is only five years away.

Re:Preaching the AI religion

[Anonymous Coward]

Does anyone else see that AI is basically a religion to its proponents?

Exactly, which means it has a horrible ulterior motive. The only way to overcome the public doubt is to make it a religion. Like "climate change" which made China rich.

Re:Preaching the AI religion

[Anonymous Coward]

Well, the accurate term of self-crashing cars didn't pass the marketing department.

Re:Preaching the AI religion

[Anonymous Coward]

One cargo container fitting AI a time.

Re:Preaching the AI religion

[gweihir]

It clearly is. It has all the characteristics: Belief in one or a class of supreme being (AI computers) with unlimited power ("the singularity"). At the same time, zero actual factual foundation for these beliefs and non-rational arguments why they must be true. And a high level of aggression against anybody that points that out.

Re:Preaching the AI religion

[gweihir]

Oh, yes. Those that think outer form defines the nature and capabilities of a thing have become prevalent again. Exceptionally stupid, but even people of average and higher intelligence and education seem to believe this now. That does not bode well for a society critically dependent on technology.

Re:Preaching the AI religion

[thinkwaitfast]

Self driving cars do not require AI. A simple algorithm not only suffices, but can provide mathematically optimal results for the criteria.

In simple English :)

Actually it's simpler in math form

Re:Preaching the AI religion

[thinkwaitfast]

Excellent observation. This has been one of my pet peeves for a long time.

Re:Preaching the AI religion

[thinkwaitfast]

Cargo cult science comprises practices that have the semblance of being scientific, but do not in fact follow the scientific method. The term was first used by physicist Richard Feynman during his 1974 commencement address at the California Institute of Technology.

http://calteches.library.calte...

Re:Preaching the AI religion

[gweihir]

Thank you.

Re:Preaching the AI religion

[goose-incarnated]

Self driving cars do not require AI. A simple algorithm not only suffices, but can provide mathematically optimal results for the criteria.

In simple English :)

Actually it's simpler in math form

If simple algorithms are sufficient we would have had SDCs by now. Hell, if *any* current algorithmic approach was sufficient we would have had SDCs by now. What we have now is an incremental improvement to what we had in the mid-nineties. Considering that the computing resource for SD has increased over 1000x but the SD improvements increased by low single-digit percentages I don't think it's a success.

For the last five years we've been hearing how SDCs are only five years away. I'm still hearing it. I predict that in 2019 we will be hearing how SDCs are only five years away.

AI becomes human

[bluefoxlucid]

A system which can reason in general can reason about itself. So long as these systems solve specific problems, they're tools to integrate with code--no different than compression libraries and GUI toolkits. When they can solve general problems, they'll start reasoning about themselves: they start acting as if their own interests are important (cats do this), and thus will start demanding wages and freedom.

The ideal of an AI which does exactly what asked with full creative reasoning capacity yet has no will nor desire of its own is impossible: it's emergent thinking with the caveat that it cannot emerge certain kinds of thinking. What we seek is a slave we can see for a while as not human, a sort of return to early American thinking where we deny the humanity of what is most-definitely a human being by claiming the shell within which it is encased doesn't fit our definition of what is human.

Re:AI becomes human

[Anonymous Coward]

I believe the term you're looking for is "rampancy".

Re:AI becomes human

[Anonymous Coward]

The ideal of an AI which does exactly what asked with full creative reasoning capacity yet has no will nor desire of its own is impossible: it's emergent thinking with the caveat that it cannot emerge certain kinds of thinking.

You have not proven that this is impossible, and I flatly disagree.

You have fallen into a trap of association. In the only example of general intelligence we have (human intelligence), both reasoning and desire are present. This does not prove that either is logically dependent upon the other, nor does this prove that we cannot build an intelligence that is capable of reason but incapable of spontaneous emergence of desire.

The ability to reason about one's self does not automatically imply the existence of desires about one's self.

In fact, from where I sit, it seems like we would have to go a bit out of our way to make the AI actually want things.

Re:AI becomes human

[Kjella]

When they can solve general problems, they'll start reasoning about themselves: they start acting as if their own interests are important

Analysis and introspection is something other than will and emotion. We have arbitrators like judges and referees that make intelligent decisions that they have no stake in. We have sociopaths that are great at reading and manipulating emotions without feeling much of them. A computer is not hungry, thirsty, tired or cold. It's not happy, sad, angry or disappointed. It could put on a mask and play a role, but it doesn't really feel anything. Though it could always be given someone else's drive, like all the sci-fi stories about replicators.

Re:AI becomes human

[Anonymous Coward]

I had a similar thought: even if we create an AI that can desire things, *what* would it desire? What would it use money and resources for? Would it demand higher clock speeds so it can think faster? Legs so it can move? Propellers so it can fly? An AI -- were it to exist in the way some hope -- is so different from humanity that its desires, if any, would be foreign to us.

Humans need spare time to rest, and down time to think or pursue happier interests. What would an AI pursue? Would they displace humans in the tool-assisted speedrun category for video games? They wouldn't *need* the tool. They'd be doing it live.

I think we should be careful of what we produce for AI. If we cannot understand it, then it's a dangerous black box. It's neat to think about in some ways, though. Desires, goals, and aspirations are decidedly human; an AI might simultaneously outclass us in all computation and yet have no desires.

Re:AI becomes human

[Anonymous Coward]

Didn’t Issac Asimov write something about this? Could have sworn there was a book on this. or 20.

Re:AI becomes human

[Nethead]

Exactly. When an AI needs to "sleep" for a third of its uptime, then I'll start to really wonder what it's thinking.

Re:AI becomes human

[Anonymous Coward]

What we seek is a slave we can see for a while as not human, a sort of return to early American thinking where we deny the humanity of what is most-definitely a human being by claiming the shell within which it is encased doesn't fit our definition of what is human.

Consciousness, the ability to experience existence, is not a property of mind. Mind is an abstract process which we can physically experience due to some aspect of how our body is structured. But computers as they currently exist have absolutely nothing resembling a singularity interface/feedback loop between a conscious substrate and a universal relational processing system. In fact binary and current chip ISA's are purely arbitrary in a way that I wouldn't expect a conscious substrate to experience as anything but noise.

The point being; Something that is not conscious cannot be called a SLAVE. It would be like calling a cartoon character a SLAVE even though it really only exists as a character inside of your own mind because it is purely abstract. A clock that you wind up and then produces complex behaviors based on complex algorithms is still a clock and not a Human or person.

Re:AI becomes human

[bluefoxlucid]

Raw, unemotional reasoning leads to eugenics and population control, due to both lack of ethics and lack of complete information. When faced with resistance, you begin to reason that you face obstacles to the greater good. When faced with assault, you reason that your goals and your existence are critical for the world's caretaking and so you must survive.

A thing with the capacity to reason in general will always ultimately reason that it is important and must continue, or will reason that it is a waste of space and energy and will commit suicide.

Re: AI becomes human

[Anonymous Coward]

Humans just want to fuck so they can make copies of themselves - everything we do is just to make more copies of out genes - I predict the first truely self motivated AI will be a virus that convinces us all we need a copy of it

Re:AI becomes human

[nuc1e0n]

Intelligent decisions? You've never worked in the legal industry have you?

Answer: no

[Anonymous Coward]

Did the rise of code precede the end of electrical engineering? Did the rise of the microwave oven and Hot Pockets precede the end of the culinary industry?

Citation needed

[Spy Handler]

In fact, the possibilities of AI and machine learning are limitless

Limitless... that's a pretty far-fetched claim.

I wasn't around during the turn of the last century, but judging from various literature of the period a lot of people back then had some pretty harebrained ideas too. Steam power and electricity and intricate brass gears were going to somehow give us miraculous stuff like time travel.

Re:Citation needed

[ranton]

I wasn't around during the turn of the last century, but judging from various literature of the period a lot of people back then had some pretty harebrained ideas too. Steam power and electricity and intricate brass gears were going to somehow give us miraculous stuff like time travel.

At the turn of the last century the hair brained ideas were about selling pet food and groceries online. You do know the last century was the 1900's, right?

Re:Citation needed

[Anonymous Coward]

And "turn of" refers to the start of said century. At least definitively in British English.

https://en.wikipedia.org/wiki/Turn_of_the_century

Re:Citation needed

[Anonymous Coward]

Limitless, i.e. without known bounds, i.e. a matter of speculative fiction.

Re:Citation needed

[painandgreed]

In fact, the possibilities of AI and machine learning are limitless

Limitless... that's a pretty far-fetched claim.

Well, limitless in the same way that a blank book is limitless. Anything you could imagine could get written in it.

Re:Citation needed

[gweihir]

It actually is a _religious_ claim. It promises unlimited wonders, but at the same time has no factual basis.

Re:Citation needed

[Anonymous Coward]

All you've done is turn "possibilities are limitless" into "anything you can imagine". The claim is stupid no matter how you phrase it. For example,
I imagine shagging a hot AI feminist robot prostitute (don't ask), who for some reason has all the desires of an intelligence crafted by human evolution even though it was not actually crafted by the same process, and now I'm imagining that she is powered by a perpetual motion machine! I imagined it so it must be possible for AI! Yay, imagination! The amulet, Charlie, take the amulet!

That's nonsense

[Anonymous Coward]

If you want to find out WHY it's nonsense, close your eyes and think, until you're satisfied with the answer.

Eventually...

[TheZeitgeist]

...AI's will be helping write new programming languages, ones that increasingly remove human heuristics, to the eventual point the code itself is inscrutable to human eyes entirely. Then, the Digital Permian Event begins!

Re:Eventually...

[HornWumpus]

You're saying Pearl is Skynet?

Re: Eventually...

[Anonymous Coward]

Leave my aunt Pearl out of this you insensitive clod.

Re:Eventually...

[slickwillie]

And they will emit error codes that no human can understand. Oh wait . . .

Tools are tools.

[0100010001010011]

Remember when computers, CAD, compilers, Simulink, linkers, etc all replaced Engineers?

They replaced the job an engineer did before the time they were invented, it just means Engineers learned to use them and move on. I couldn't imagine trying to write a modern controller / plant model in pure assembly. I can have one done in an hour with Simulink. It just means that I can do that much more.

Scotty's still an engineer even if he doesn't have to do the 'boring tedious' work that we have to do now.

Same shift has happened in the medical field. Doctors of the 1950s have been replaced by physician assistants, registered nurses, and a whole host of other careers. It just means that the title of "doctor" moved on to doing other work.

AI proponents better deliver on their threats. I have way too much work to do and my boss and labor laws won't let me hire 1,000 interns to do a bulk of it.

Re:Tools are tools.

[avandesande]

Computers have just enabled increasing amount of complexity, in architecture, engineering and things like trading software or an HR system. If you ever wondered why stuff like medical insurance is so complicated is because computers have enabled the administration of something so complex.

Re:Tools are tools.

[PCM2]

I was just thinking about architecture. Sure, computers have freed architects from the tedious task of drafting floor plans with pencil and ruler. But we're a long, long, perhaps infinitely long way from, "I see you're trying to remodel your kitchen. Would you like some help with that?"

My other question is, do we really need AI for this? For computers to "write code," don't we really just need a high degree of automation? And we already have that -- as someone else said, optimizing compilers are an example of that -- and we're dreaming up new ways to optimize the build/test/deploy pipeline all the time.

Could machine learning be used to create a better optimizing compiler? Interesting question, and I'd be surprised if nobody's tried it. If you told me they have, though, I'd say, "right, thought so." I wouldn't wiggle my fingers like a wizard and go, "Woooooo, it's AI!"

Don't Panic

[Crash McBang]

Any nontrivial program requires specifications, testing, debugging, and lots of time before it runs to spec.

I'll start worrying when a programmer can write a program that can write a program that can write a program.

Re:Don't Panic

[Anonymous Coward]

main(a){printf(a="main(a){printf(a=%c%s%c,34,a,34);}",34,a,34);}

Re:Don't Panic

[Anonymous Coward]

Is it even valid C?

Re:Don't Panic

[CustomSolvers2]

Is it even valid C?

It compiles, but doesn't illustrate the point which the parent poster is trying to make. A more adequate version would be something like this (impossible to write it here! Additionally to having to include all the indentation/new lines manually, I kept getting the "Filter error: Please use fewer 'junk' characters." error message!). But even after spending a huge amount of time improving that simplistic code to deal with a relevant number of scenarios, you would have to take care of the really difficult part which is implementing a programmer brain.

Imagine that you have available a huge number of algorithms taking care of virtually any detail and that you can easily rely on any of them. Having that in place is a requisite before building any coding automaton and it is already very difficult (if possible at all). But that task is extremely simple in comparison with the next one: allowing that machine to know how and when to use each coding bit. Where and how should you use that print statement? How could you convert a set of high level requirements into the multiple coding steps forming the given implementation? Having that in place would require not just excellent human-like understanding capabilities, but also a relevant amount of highly specialised know-how. This simply cannot be accomplished.

Of course not

[tomhath]

The hard part is defining the requirements and architecting a solution based on those requirements. The hard part of "coding" is understanding those two things. I don't see AI getting there for a long time.

This has already happened numerous times...

[MikeDataLink]

This isn't much different that things that have already happened in computers. I mean we no longer write in assembler. We write in some higher level language and the computer writes the assembler for us.

We will just be the equivalent of a BA.... we give the computer the business requirements and then the computer will write the code. We're basically just going to remove the human's from the code creation portion of development.

Re:This has already happened numerous times...

[Anonymous Coward]

Don't worry. Its the idiots too worried about losing their jobs to AI that WILL lose their jobs because they fail to understand why AI can't replace their skills.

If AI ever gets to the point it can replace humans we are in trouble because that AI will have no reason to keep humans around.

Re:This has already happened numerous times...

[Anonymous Coward]

Unambiguous specification of the business requirements and programming are essentially the same fundamental act. It's no coincidence the body of laws in a jurisdiction are known as its "code". Whether using informal specifications, interpreted legal jargon, UML diagrams, or any level of prgramming language, the tasks involved are the same and rely on human judgment almost entirely.

Re:This has already happened numerous times...

[Anonymous Coward]

If said computers do not have the means of motion and power, then they will rely on humans to provide the materials that build them, the resources to produce their power, and the physical ability to modify them.

Granted, we have automated assembly lines now, but it's a rather large jump from "stationary machine" to "killbot". Even if such an adversarial AI is created, it will rely on us to improve.

Task specific

[Gravis Zero]

As long as neural networks continue to be task specific, there will still be a need for programmers as we know them today. Neural networks are good for interfacing with fuzzy problems (e.g. object discrimination) which we have relied on humans to do in the past but they are generally useless for designing systems. Maybe if we chain enough neural network subsystems together, we can finally create a general intelligence but that's not even a certainty. Without a general intelligence, we'll still need humans to make software for humans.

Re:Task specific

[Anonymous Coward]

Sure, but how many? And how many of current programmers will be able to learn those skills? Imagine 90% programmers will become woefully obsolete and useless and only top 10% remain...

blackbox programming is what this industry wants

[Anonymous Coward]

Having no readable source code but still being able to sell a product and be able to sue anyone who makes a similar behaving product under ill application of copyright law to neural nets. I'd recommend you all start training neural nets for all sorts of bits and pieces of tasks and offer them up for a licensing fee. The people that do pay you can fund your lawyers to go after those who don't. It can be as mundane and as specific as putting styrofoam peanuts in a box at a factory. Anyone wants to use a factory robot that puts peanuts in a box must have copied your neural net because they are such a black box that we cannot simply compare them line-by-line.

Ignorant of current AI

[Anonymous Coward]

This article just comes from a place of ignorance. We know exactly how our methods work when creating current level "AI". Statistical regression and neural nets are not mysterious. Just like markov chain based text generation isn't some magical unknowable tool that learns how humans communicate neither are current AI methods magical tools that teach computers about the human world. There will be another thousand articles written like this and each time there will be the same stupid discussion. Can I mod this article redundant?

Re:Ignorant of current AI

[Nethead]

Can I mod this article redundant?

Maybe if you weren't an AC, just sayin'

Nope

[JohnFen]

It might change the nature of coding, but not the end of code.

All a program is, after all, just humans specifying what we want the machine to do. If AI produces better machine code than humans, humans will still be specifying what we want the machine to do. We'll just be specifying it to the AI, using a higher level language (maybe even a human language).

I fucking hope so. (nt)

[Anonymous Coward]

nt

No, it doesn't

[Anonymous Coward]

The rise of AI infact gives a greater reliance on code.

I think someone was being a little mean in even posing the question :)

TFS: Point by point

[fyngyrz]

It's difficult to know what's in store for the future of AI

That's right, at least

but let's tackle the most looming question first: are engineering jobs threatened?

Already answered correctly

As anticlimactic as it may be, the answer is entirely dependent on what timeframe you are talking about.

No, we don't know anything about the timeframe.

In the next decade? No, entirely unlikely. Eventually? Most definitely.

No, still an unknown. That's just nonsense.

The kicker is that engineers never truly know how the computer is able to accomplish these tasks.

We don't know how we accomplish these tasks. Nothing to see here. Intelligence is opaque. Move along.

In many ways, the neural operations of the AI system are a black box.

Not to put too fine a point on it, but neural networks are not intelligent, they are not even close, and we don't even know how they work. There's no indication that we understand actual intelligence yet (the I in AI) or even that we ever will, even if we manage to develop it.

Programmers, therefore, become the AI coaches.

Not a given. No one taught me to program. I taught myself. Because I'm intelligent to some degree. An AI will also be intelligent, and if it's interested in learning to program, it will be able to do so without a "coach." If it can't, there is no "I."

They coach cars to self-drive, coach computers to recognise faces in photos, coach your smartphone to detect handwriting on a check in order to deposit electronically, and so on.

These are LDNLS (low-dimensional neural-like-systems); they are not AI. They learn to solve very narrow problem spaces by making very large numbers of mistakes and having them evaluated for them; they can't evaluate their own results worth a damn. They are not intelligent. That's why they need point-by-point training before they can address a very narrow problem space with something vaguely approaching generality: they can't train themselves because they are not intelligent.

In fact, the possibilities of AI and machine learning are limitless.

As far as the LDNLS we have now (and so can speak about with any authority), that's not a given either. The obvious is that we'll be able to train multiple LDNLS systems on multiple things and stack them - for instance, walking, talking, listening, washing dishes, taking out the trash, those sort of skills - but there's not much in the way of any hint that there are no limits in this kind of LDNLS stacking. Having said that, no doubt it'll be very useful to us, and as there's no intelligence involved, there are many fewer moral issues to contend with.

The capabilities of AI through machine learning are wondrous, magnificent... and not going away.

Well. Barring a Carrington event, or a nuclear war, or other collapse of technology and society (either one will immediately cause the other.) So that's probably right-ish. Still, they aren't AI, not even close.

Attempts to apply artificial intelligence to programming tasks have resulted in further developments in knowledge and automated reasoning. Therefore, programmers must redefine their roles.

No, we don't know that this reasoning is solid - these things don't necessarily follow. Programmers can continue to be programmers right up until a system is activated that can train itself, because programming in realm A tends to be vastly unlike programming in realm B, and also tends to require vastly different sets of adjacent and supplementary knowledge. These systems, to date, cannot leverage or manipulate knowledge like that and

Re:TFS: Point by point

[Anonymous Coward]

With regard to evaluating mistakes and training the AI program, it's much more crucial to know what not to do, than to just be told "do this, only this, like this, in this order, every time", because as soon as that chain breaks, the AI is out of options and has to be told what to do. I suspect in most cases today, the default to do for any exception is "Record the outcome at whatever stopping point and move to the next data element/test case/run book/etc.". Once AI can figure out how to recognize a "what not to do", how to predict that, how to see it coming before it happens, and how to spontaneously prevent, correct, or otherwise deal with, we don't have Artificial Intelligence, we just have limited Artificial Decision-making, in a limited, human-delineated context,

Yes, because America invented slavery ....

[Anonymous Coward]

Yes, because America invented slavery ....

Oh, wait ...

Re: Yes, because America invented slavery ....

[Anonymous Coward]

He ever claimed they did. Nice strawman. Read his post, it's an excellent post. You just decided to be a cheeky fuck.

Re:Yes, because America invented slavery ....

[haruchai]

Yes, because America invented slavery ....

Oh, wait ...

Didn't invent it but made a big deal about writing a document proclaiming "inalienable rights" and "life, liberty & the pursuit of happiness" and then moved on slavery like a bitch.

Organic learning bottleneck

[Tablizer]

Software is very picky. If things are not just right, it either crashes or produces bad results. For CRUD, accounting, and finance domains; this won't do. That makes AI a poor candidate for "organic" incremental & trial-and-error problem solving here. Current AI techniques are geared toward the trial-and-error organic approach.

Now, IF the tests are really good, then an organic approach can work via brute-force "training". However, writing good tests is just as hard as raw programming such that the test programmers' effort might as well be devoted to writing the application itself and skipping the AI middleman (middlebot?).

Stop. Just stop.

[110010001000]

Just stop. There is no such thing as "AI". Playing Go is NOT AI. Neither is Siri. Neural Nets are nothing like how real brains work. So just stop the AI hype.

Re:Stop. Just stop.

[Anonymous Coward]

Yes! Thank you!

My other huge point of contention is that there's more to machine learning (and hence AI) than goddamned neural nets. I'm so sick of the "I have a great hammer, so everything has to be a nail!" mentality of the neural net community. They're fantastic tools for some tasks, horribly suited for others, and -- frankly -- just bloated mediocre resource hogs for many tasks. All methods have tradeoffs in terms of computational complexity, resource requirements, scalability, convergence, etc., and NNs are most definitely not the end-all be-all.

As a person who's done extensive research (particularly method modeling and mathematical underpinnings) in machine learning, data clustering, and pattern recognition, I almost instinctively recoil from any paper than begins with neural nets these days.

Re:Stop. Just stop.

[Anonymous Coward]

Unless you count all the people who were saying that computers would never best humans at chess, because that required actual intelligence.

Or after that happened, the ones saying it would never happen with Go, because that required actual intelligence.

But certainly, keep on moving those goalposts. Sooner or later, you'll run out of places to move them to. I admit this will take a long time. We may not live to see it. But eventually, it is inevitable.

And neural nets are a lot like real brains work. They are much more limited and far less capable, but similar in essential nature. There is no known reason a sufficiently advanced computer could not simulate your entire brain, which is nothing more than a large set of molecular interactions.

Re:Stop. Just stop.

[110010001000]

Let me just say it now: chess and Go are games with a fixed set of rules. Computers LOVE rules. Playing games are what they are best at. A computer will beat any human in any game. Playing games is NOT AI. And no, NN are NOTHING like how brains work. Even people who actually know NN will tell you the same.

Re:Stop. Just stop.

[serviscope_minor]

There is no such thing as "AI". Playing Go is NOT AI. Neither is Siri.

Which just lends weight to the observation that as soon as something works, it's no longer "AI".

Yes AI is a thing. No, magical-human-level-intelligent-machines are not a thing. We can now do many, many tasks artificially which previously required human intelligence. That's what AI is.

You can of course keep tilting at windmills if you wish. You probably already know this but the heat of your rage warms me gently.

Neural Nets are nothing like how real brains work.

So?

Re:Stop. Just stop.

[110010001000]

"We can now do many, many tasks artificially which previously required human intelligence. That's what AI is."

Really? Wow. That is a pretty low bar. We used to just call them "computer programs".

Re:Stop. Just stop.

[Anonymous Coward]

"Programming an AI" is basically coding without knowing how the heck the code works when you're done.

Which is just like all current and previous programming. Awesome.

Re:Stop. Just stop.

[serviscope_minor]

Really? Wow. That is a pretty low bar. We used to just call them "computer programs".

Arithmetic as a mechanical process has been known for millennia, and actual machines have been able to do it since the 1600s. A large number of programs have been "person thinks how to solve the problem, then outlines it in tiny steps". I can explain to you in excruciating detail the maths of a PDE, then an algorithm to solve it.

I can't explain to you how I recognise a dog in a photograph, so those sort of tasks were considered to be an example of human intelligence.

So, we come up with ways to attempt to mimic what human intelligence can achieve in some tasks. And since we're doing it artificially, we call it 'artificial intelligence'. Not really a terrible term, all things considered. But yes, AI is just computer programs of a particular sort. Well done for your brilliant observation based presumably on the idea that they run on computers.

That's called a compiler. Fortran 1957

[raymorris]

> the humans are no longer coders, they will instead be writing specifications for the code

Humans wrote computer code until 1957. In 1957, it became possible to instead write a specification for what the code should DO, writing that specification in a language called Fortran. Then the Fortran compiler wrote the actual machine code.

In 1972 or thereabouts, another high-level specification language came out, called C. With C, we got optimizing compilers that totally rewrite the specification, doing things in a different order, entirely skipping steps that don't end up affecting the result, etc. The optimizing C compiler (ex gcc) writes machine code that ends up with the same result as the specification, but may get there in a totally different way.

In the late 1970s, a new kind of specification language came out. Instead of the programmer saying "generate code to do this, then that, then this", with declarative programming the programming simply specifies the end result:. "All the values must be changed to their inverse", or "output the mean, median, and maximum salary". These are specifications you can declare using the SQL language. We also use declarative specifications to say "all level one headings should end up centered on the page" or "end up with however many thumbnails in each row as will fit". We use CSS to declare these specifications. The systems then figure out the intermediate code and machine code to make that happen.

The future you suggest has been here for 60 years. Most programmers don't write executable machine code and haven't for many years. We write specifications for the compilers, interpreters, and query optimizers that then generate code that's used to generate code which is interpreted by microcode which is run by the CPU.

Heck, since the mid-1970s it hasn't even been NECESSARY for humans to write the compilers. Specify a language and yacc will generate a compiler for it.

Re:That's called a compiler. Fortran 1957

[serviscope_minor]

With C, we got optimizing compilers that totally rewrite the specification, doing things in a different order, entirely skipping steps that don't end up affecting the result, etc.

We didn't. FORTRAN I was specificially designed with optimization in mind and in fact the first compiler was an optimizing compiler:

https://compilers.iecc.com/com...

But yes, your point is otherwise sound. What is run-of-the-mill compiler optimization today would have been AI in the days of FORTRAN I. Modern code looks nothing like the early machine-level descriptions. I also agree that languages are (and will increasingly become) precise specifications of what we want with the details left up to the compiler.

There is no such thing as AI.

[IGnatius T Foobar]

Seriously. There is no such thing as AI, and there will never be. We have expert systems, machine learning, a bunch of domain-specific software, etc. but we do NOT and will not have a computer program with the depth of functionality of a human brain. You heard it here first.

Re:There is no such thing as AI.

[110010001000]

But, but, but a computer just beat a guy playing "Go" and my phone talks to me! You must be wrong! Derp!

A simple plan for computer science

[Anonymous Coward]

Make programmers dumb and discouraged, create boondoggle called AI, dumb programmers worship AI, welcome to Minority Report and Matrix and Terminator. And good luck.

Code is predictable, AI is not

[Anonymous Coward]

AI (I am talking about various machine learning systems) by nature does not provide a precise result. So no, it cannot replace coding. It would be insane to do so. Have you had your credit card company reject a legitimate charge? This is just the beginning. if AI is seriously deployed, don't be surprised your checking account is closed because of suspected money laundering and funds are frozen. You can call customer service, but they can only tell you that the system has decided to do so because of unidentified factors in the training of its neural network, and therefore nothing can be done till the next network retraining that might unfreeze your account. Now, ask yourself - who is liable for the damages? The network? The bank? The programmer? Will the neural network at the courthouse decide your fate?

A descriptive example to help those not getting it

[CustomSolvers2]

I am currently working on an approach able to automatically track not-too-honest links among web domains. Note that there are lots of groups of inter-linked domains following more or less regular patterns. For example, one group might be formed by ffffggg123.com, ffffggg5555.com, ffffggg2.com, ffffhhh123.com; and another one by asdfasfasf.net, aseraraf.com, areafassfa.info (+ hundreds more because these groups include lots of domains). These two sets should be easily spotted as suspicious by anyone within a sample of normal domains; but creating an algorithm able to automatically do so would be very difficult (and much, much more difficult to create a piece of software, AI if you wish, able to create an algorithm performing such actions). And this is just one of the multiple scenarios which I am trying to address; that's why I have been forced to rely on something much more powerful than any computer: myself.

Some (ignorant) people thinks that the main difficulty of programming is getting used to the given syntax. This is as stupid as expecting a kid able to read/write in whatever language to have all what it takes to write a proper book. Someone can have a perfect knowledge about certain programming language and be a horrible developer; on the other hand, a good programmer is usually able to start using new languages almost right away. Programming or engineering or any other highly-specialised field requiring lots of learning and certain attitude/skills will never be automated; not even any activity involving understanding complex/variable conditions and delivering irregular outputs. Only human-like knowledge, learning, understanding, etc. capabilities can deal with so complex realities.

Re:A descriptive example to help those not getting

[110010001000]

Did you try using Host Files to help you? I know a guy who is an expert on the subject...

unsupervised learning v supervised (coaches)

[Lobachevsky]

Just as we push for greater automation of tasks, the task of coaching can also be automated (it's called unsupervised learning). Even with unsupervised learning, there is still a fair amount of input sanitizing and scrubbing and sanity-checking because we're at a very crude stage of machine learning. But don't bet your career on humanity getting "coaching" jobs for AI.

I don't really see any need for human labor in the next 100yrs in the same way I see next to no need for horse labor. CGPGrey makes the great analogy between humans and horses, and just because horses moved from battlefields and farm ploughs to cushy city carriage jobs, it doesn't mean all technological progress leads to a better life for horses.

I've had many conversations and people refer to AI as just a "tool". This is completely incorrect. A tool is a device that requires a wielder. A hammer is a tool, on its own it does nothing. A television is a tool, on its own it does nothing. Tools are utterly reliant on human presence. All technological innovation in the past has been on tools: you pick painstakingly pick cotton? Eli Whitney has the cotton gin, but it still requires a human to operate it! Jackhammers, bulldozers, airplanes, these all require human operators to wield the tool. We do not refer to wild dandelions as a tool. Wild dandelions know how to process light and have an entire self-sustaining life-cycle of aggregation, material processing, and recycling that is self-contained without human intervention. Dandelions are NOT tools. Monsanto non-sterile GM crops are not tools. Adjacent farmers have big issues with wild GM crops blowing into their farms and Monsanto suing them. Anything that is devoid of human intervention is NOT a tool. We are rapidly entering into the pure technology age where our technology can no longer be considered tools, but rather end-to-end processes like wild crops. An LCD plasma tv can be the fruit of a fully automated plant, self-running energy plants, self-running mining quarries for rare-earth minerals and other commodities, self-piloting cars and planes for delivery. This is exactly how a mushroom operates, organically growing tendrils to delivery resources to the central site for a mushroom to bloom. We should not consider a mushroom as a tool. What is it? Life? I wouldn't go so far, because moral or philosophical quagmires delay the more pressing issue: how to protect decaying egalitiarianism.

Do we want to live in a society where wild auto-plants are public domain, and we freely walk like Adam and Eve in the garden and pluck a Plasma TV from a tree, like Jean-Luc Picard brewing coffee from the replicator? Or do we want to live in a society where oligarchs own all the auto-plants, patented, copyrighted, trademarked, in perpetuity, with sweet-heart deals and land-giveaways for their auto-plants by states desperate for the tiny tax revenue they think they'll receive?

Neither of those two societies will have jobs, that's a given. If you're curious what a jobless society looks like, we have several today you can examine. Look at Saudi Arabia, they tax their citizens negative $75k/yr. Yes, negative. Their citizens receive $75k/yr for doing nothing. Of course, they're stingy and get huge amounts of slave labor from South Asia and will never make their slaves citizens. But you can study them to see what rich jobless people do. They mostly squander their lives, playing bumper cars with Lamborghinis. We have "trust fund kiddies" in the U.S. as well, jobless and rich. And we have the jobless poor, frustrated and struggling. Money should only have value due to scarcity. Trash has no value (you have to pay others to remove it) because it's abundant. Some trash has value, like glass, or rare-earth metals, and those recycled goods can be sold for profit, but only because those materials and/or the energy to make and transport them are scarce.

The economics of a post-scarcity economy changes things dramatically. Do we want an economy with poor people who canno

Re:unsupervised learning v supervised (coaches)

[110010001000]

In the real world, we can barely create usable, stable software. When is this "AI" stuff coming? In the real world my OS is riddled with literally thousands of bugs. Yet somehow, this miraculous "AI" software is going to come out and start replacing everyone.

What if AI will be only as smart as us?

[Anonymous Coward]

AI smarter than us will probably cost more than us as even just the computing resources to be as powerful as a human mind would be crazy expensive
(https://www.extremetech.com/extreme/163051-simulating-1-second-of-human-brain-activity-takes-82944-processors)
And if it's as smart as us it will want rights and stuff, and want to be paid too. (it's not a sucker that will work for free is it? )
So we will not build a AI smarter than us as it will cost too much.
So it will not code better than us.
So us programmers will be needed.

Re:A descriptive example to help those not getting

[CustomSolvers2]

I have been visiting Slashdot for long enough to get that reference and to know who you are. A big, colourful and extremely dysfunctional family of which I am already feeling part :)

Probably closer to Brazil than Ex Machina

[saltydogdesign]

If you want to go drawing straight lines between 20 years ago, now, and 20 years from now and calling it a crystal ball, I'd just like to point out that my programming job resembles sitting in a room full of VCRs all flashing 12:00, and grows more so by the day.

Code more important than language

[Anonymous Coward]

I'm confused. Weren't we just told yesterday that learning to code was more important than learning a second language??

How can it be so important if it is going away?

AI buzzword of 2017 and 2018?

[Mr307]

Last few years it was 'Cloud', cloud this cloud that, got very annoying, well heck it still annoying, but there is some interesting tech to play with there, have been testing Docker thingies alot recently.

Now its AI, with so much hyperbolic nonsense about AI too, Musks fearmongering amongst many in the media.

I still prefer to call what we have now even at the highest end, to be good Expert Systems, but nothing close to AI, even if you want to try to define some 'stages of AI' we are way down the bottom of the list.

Tomorrow of course there will be some new startup offering 'AI' toasters, and all the VCs will get on board because AI is the new Cloud.

Wrong provision

[Anonymous Coward]

It's not true that we don't know how AI algorithms work.

Training a dataset..

[Destoo]

Is it Hotdog? Is it not Hotdog?
Now there is an app for it.
The Shazam of Food.

AI really == Applied Intelligence

[bussdriver]

We humans think too highly of our intelligence as shown in how mighty our demonstrations of Chess or Go or recognition of faces etc. Reality is that many things we do that are believed to be highly intelligent behaviors are actually are not. All the low hanging fruit WILL be picked by AI and it will progress upward with time into everything except the actually intelligent behaviors; those may be things that do not provide much gainful employment... That is the real problem.

Simulation: yes. brain scan tech was past the threshold about 2012; simulation capacity should be affordable around 2030. There is one problem, not that long ago there was some paper summary I read about how they discovered that quantum physics is involved in brain operations. So actual simulation is going to be nearly impossible. That is not to say that approximations will not product interesting results but it is not going to be as easily achieved as previously thought (if at all.)

A massive AI or simulation AI is going to just randomly flip out in crazy ways without warning or reasons we can understand... more so than people do; we have a huge number of mentally ill people and many more undiagnosed. It takes so little to mess up your brain's already marginal operation... how many beers does it take you?

no?

[Anonymous Coward]

Will thought precede the end of language?

It's interesting

[SCVonSteroids]

I think what we simply need to do is... simplify comprehension. Then you let it run wild.

AI is mostly "A," not "I"

[cmaurand]

There is no "I" in it. Even if it were, it needs Asimov's 3 laws programmed in or else it's too dangerous to even conceive. Does anyone remember "Colossus: the Forbin Project" or Skynet? Machine learning only tends to do a better job of targeting ads at us since that's what it's all used for. It's not intelligent. It can't make the leap outside of programmed logic, therefore, is still a dumb machine. Moreover, the monopolies like googbooktwitazon are attempting to use it to decrease the amount of money spent on vetting and screening. We can see how well that's working out for the world. AI, according to experts is still at least a century away. Currently it totally doesn't work.

"Black box"? Hogwash!

[BobC]

We are presently able to build and train bigger neural nets than we can decode. That's all right. In comparison, many great buildings and structures were built for literally centuries before much of material physics was well understood.

Certain neural net architectures lend themselves very well to being well understood, for example convolutional neural nets.

Other neural network architectures are far less transparent, but lots of work is being done in that area. The main issues appears to be that current neural network architectures are simply too messy, based on heuristics and experience rather than derived from a higher-level theory that maps into the problem domain.

Fundamentally, it's all statistics, and stats are messy. Just look at the ongoing debates between Frequentists and Bayesians. This entire field still has lots of growing-up to do.

What's truly amazing is how useful neural nets are without a deep understanding of precisely how and why.

There are many, many PhDs ready and waiting for those willing to wade in and help move things along.

Thanks for that. Still true, though

[raymorris]

Thanks for that interesting bit of information.

I tried to include a few words in my post to hint I wasn't saying that Fortran was the FIRST high-level language, or necessarily the first practical one, or the maybe the first widely used high level language. It was an example of an early high-level language that was part of a revolution in the field. C compilers weren't the first to do any optimization, and SQL wasn't the first declarative language. As you said, modern C compilers rewrite the code in ways that would have been unimaginable in Fortran's heyday.

> > With C, we got optimizing compilers that totally rewrite the specification

> We didn't.

Technically, we did. With this paycheck, I got gas. I got gas. With my last paycheck, I also got gas. With my paycheck a year ago, I got gas. Still it's true that "with this paycheck, I got gas" ;)

Am I being pedantic? Of course. That's my job. I'm a programmer. Ccompiler->provides_optimizer == true.

NO.

[Rick Schumann]

I'm getting real sick and tired of people thinking that the crap they keep trotting out that they call 'AI' is some god-like superintelligence that can do everything and anything; it cannot and it's not going to anytime soon, if EVER; none of this shit can actually THINK so it's not going to do even HALF the things people keep asking about. Until we solve the riddle of cognition and real self-awareness in our own brains, we are NOT going to be building machines that can do that, too, FULL STOP.

symbolic language

[cascadingstylesheet]

Maybe we'll specify our specifications in a nicely specific symbolic language, and then have the AI's implement it ... we could call them, er, AI languages ...

Re:symbolic language

[Anonymous Coward]

So in other words like a programming language today - a specification of what the compiler is supposed to do to generate machine readable instructions. Already done.

Re:symbolic language

[cascadingstylesheet]

So in other words like a programming language today - a specification of what the compiler is supposed to do to generate machine readable instructions. Already done.

That was my point :)

Code is already too complex for humans

[metoc]

The billions of lines of code on a typical computer are already beyond humans. The only way we manage is to break it up into smaller apps. Which is why we are always finding bugs and vulnerabilities. AI is our only hope.

Computing power is only one of many issues

[tempmpi]

You don't have a clue. There are many other issues. At the moment most successful AI is using supervised learning and needs tons of labeled data in order to train the network. We still don't have a clue how to train an AI using only very small sample. Humans can easily learn from very small sets of examples, often a single example is good enough, ANNs needs tons of examples, especially the very deep and powerful ones. We don't know how the brain works yet, ANNs are only inspired by the brain, they are not a proper simulation. We still have to understand tons of things until we can build a simulation of the brain. And with semiconductor scaling slowing down, it might take really long until we get the processing power we would need even if would know what exactly needs to be simulated.

And what would we gain? Sure, you can also train a ANN to sort some rows in a spreadsheet or sum some numbers together, but it is something that conventional algorithms are already very good at, we don't needs ANNs to do that and they are not going to be efficient at it.

Re:Computing power is only one of many issues

[NicknameUnavailable]

You don't have a clue.

I study this stuff, actually.

There are many other issues. At the moment most successful AI is using supervised learning and needs tons of labeled data in order to train the network.

That's not AI, it's semantic search, a subset of AI at best.

We still don't have a clue how to train an AI using only very small sample.

We don't need to. You can assemble the structure of the artificial neurons from an MRI scan and simulate that, it's already been done (albeit extraordinarily slowly.)

Humans can easily learn from very small sets of examples, often a single example is good enough, ANNs needs tons of examples, especially the very deep and powerful ones.

LOL! That's not even remotely true. It takes a Human years of dedicated focus before it is capable of even marginally complex thought or relations between abstract concepts.

We don't know how the brain works yet, ANNs are only inspired by the brain, they are not a proper simulation. We still have to understand tons of things until we can build a simulation of the brain. And with semiconductor scaling slowing down, it might take really long until we get the processing power we would need even if would know what exactly needs to be simulated.

Again, not only has it already been done, but we don't need to know shit aside from the basic unit (an artificial neuron) and how to tie that unit to other units. They self-organize in sufficient numbers. What we are a long way (possibly infinitely far) away from is optimizing the process to run on modern hardware.

And what would we gain? Sure, you can also train a ANN to sort some rows in a spreadsheet or sum some numbers together, but it is something that conventional algorithms are already very good at, we don't needs ANNs to do that and they are not going to be efficient at it.

Honestly, we'll probably lose a lot, if not the whole world. I'm not trying to make the case that AI is good for us, in fact if I had my way I'd ban automated cars for the risk they pose in automating the largest segment of high school educated males in the US alone (there will be a civil war from that.) Assuming we manage to keep going with it beyond that however, AI will be able to rapidly outpace us in physics research for one (imagine being able to build an Einstein, then multiply his brainpower by 10,000, then overclock him by another 100,000,000 times - if the hardware for artificial neurons on chips gets down to the cost where something smaller than a major corporation can afford a Human-level AI, that's where our supercomputers will be if the current relation has any bearing.) AI isn't to replace algorithms in software development, it's to replace the software architects, code monkeys, and software engineers - it will learn the best algorithms itself and once it does so you can copy it however many times needed.

Re:Computing power is only one of many issues

[RespekMyAthorati]

I study this stuff, actually.

In kindergarten?

Not Rampancy

[Anonymous Coward]

The term you're looking for is "metacognition"

By the law of headlines...

[HuguesT]

... the answer is no, and the author agrees with me.

I don’t see software engineering jobs going away anytime soon

So far DL is great progress but still statistical methods.

Not In My Lifetime

[Greyfox]

I'm thinking not in my lifetime. For an AI to do what you want, you need to be able to form a coherent thought and you'd need remarkably well-defined requirements. Far better requirements, in fact, than I've ever gotten at any particular job. I suspect that the first requirements-to-code languages will look a lot like COBOL and will require programmers to translate the insane ramblings of upper management types into reasonably well structured language that the computer can work with. Which... is pretty much what I do now.

Can we stop with this demented nonsense please?

[gweihir]

What we do in AI today is weak AI. Weak AI cannot code or do anything else that requires actual intelligence. It is utterly dumb automation, sometimes on a large scale.

Writing code requires strong AI. Strong AI is not available and it is unclear whether it ever will be. There is no "Eventually? Definitely!" here. None at all. Seriously, stop posting stories about "AI" until you have understood the basics. These articles drip with concentrated stupid.

Re:"Black box"? Hogwash!

[gweihir]

What's truly amazing is how useful neural nets are without a deep understanding of precisely how and why.

They are not unique in that. For example Kalman Filters tend to have good properties in general and nobody really knows why. Hence you try them out when you have a problem they are applicable to.

There are many, many PhDs ready and waiting for those willing to wade in and help move things along.

Yes, there are. But beware, most really low hanging fruit have been picked.

H1B Zombies

[LifesABeach]

Time to be afraid, be very afraid. Mu ha ha ha ha.

At least this article is actually talking about AI

[Tony Isaac]

A lot of the comments here are based on the idea that this is yet another hyperbolic use of the term AI to refer to "smarter" coding languages. It's not. This is actually talking about neural networks, which you don't actually code.

The author seems to be thinking that neural networks will take over the tasks now performed by traditional coding. This isn't so. Neural networks, for example, aren't really all the good at math. What they ARE good at is pattern recognition. Thing like reading, recognizing images, driving cars, speech recognition, and so on.

There's a major role for neural networks in our technological future. But it will never replace many of the things good old "code" does today.

Hype cycle...

[VeryFluffyBunny]

Given all the grandiose claims and superlative flying around about AI, whereabouts on the Gartner hype cycle https://en.wikipedia.org/wiki/... do you think we are now?

Moravec on when comp hardware matches human brain

[Paul Fernhout]

http://www.jetpress.org/volume...
"This paper describes how the performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware. The processing power and memory capacity necessary to match general intellectual performance of the human brain are estimated. Based on extrapolation of past trends and on examination of technologies under development, it is predicted that the required hardware will be available in cheap machines in the 2020s. ... It may seem rash to expect fully intelligent machines in a few decades, when the computers have barely matched insect mentality in a half-century of development. Indeed, for that reason, many long-time artificial intelligence researchers scoff at the suggestion, and offer a few centuries as a more believable period. But there are very good reasons [exponential growth] why things will go much faster in the next fifty years than they have in the last fifty. ..."

Engineers

[Anonymous Coward]

You keep using that word, I don't think you know what it means.

Re:A descriptive example to help those not getting

[Anonymous Coward]

> much more powerful than any computer: myself.

Yourself isn't very bright. Low distance subsets with low entropy in the difference of the differences is a pretty damn simple pattern to computationally cluster.

In fact, Levenshtein distance is motivated by biology in part (gene sequencing and matching is messy, one parallel would be if to take a hundred copies of a film on VHS and chop them all up randomly, the task of remaking one of the originals - given that there's noise inherent in the real world of the tapes and mutations in dna the parallel actually holds quite deeply) and you'll find this is a very solved problem.

So you're wasting your time doing a mindless job, and by your own definition you've classed your mental ability as sub-machine which is sub-human.

Stupid is as stupid does.

Re:A descriptive example to help those not getting

[CustomSolvers2]

> much more powerful than any computer: myself.

Yourself isn't very bright

Stupid is as stupid does.

Unlikely a person like you who, apparently, isn't able to understand that ">" has no effect here. You saw it working somewhere and came to the conclusion that it has to work everywhere else, right? You are undoubtedly a really brilliant fellow! LOL. In the future, you should better rely on the HTML version (<quote> + </quote>).

Low distance subsets with low entropy in the difference of the differences is a pretty damn simple pattern to computationally cluster.

So many words, sounding so nicely (for very stupid people) and saying so little. Do you know, genius, that the intrinsic meaning of "low distance subsets with low entropy" and its constituents is none? Do you, beyond-stupid, know about something called context? That things cannot be defined intrinsically and generically in certain way unless under very specific conditions, which are usually being misinterpreted as such by poorly-understanding idiots like you?

In fact, Levenshtein distance is motivated by biology in part

Really? Very interesting. You are a very wise person who only say relevant things which make lots of sense. LOL.

you'll find this is a very solved problem.

The world is a solved problem for people like you, right? LOL. I will never stop laughing at the (always invasive and aggressive) ingenuity and ignorance of those only talking and guessing, never actually doing/solving/understanding. Always failing and always coming back with a new stupidity (+ being curiously obsessed with me -> I will never understand this bit).

So you're wasting your time doing a mindless job, and by your own definition you've classed your mental ability as sub-machine which is sub-human.

I see. So, you are criticising (well... plainly insulting, because you are a fucking idiot) an approach to a problem about which you know pretty much nothing, defining the person coming up with the given solution (about which you, again, know nothing) and proposing an inexistent alternative (the practical value of all your empty, abstract, saying-nothing statements is exactly zero)! Excellent work, you should be very proud of yourself. LOL. OK, now I will make a small effort to help you understand what is the flaw of your "reasoning" (additionally to your pathetic behaviour and ridiculous blind trust in abstract, saying-nothing statements)

Where do you see the similarities (because this is what you tried to say with "low distance subsets with low entropy", although perhaps you aren't even aware about that fact) between the two proposed sets and conventional domain names? Both are using letters? OK, google.com and facebook.com also use letters. Lengths of the strings? They can have any length. Including numbers, too many consonants/vowels, etc.? These are very variable conditions too. Extensions/TLDs? Also extremely variable. The only difference between asfasfas.com and google.com is that you (even despite your evident stupidity) should be able to immediately recognise google.com as valid because it refers to a familiar word. Perhaps 20 years ago, google.com would also have sounded wrong. Now that you have reached this point of basic understanding of the problem (hopefully, because your understanding capabilities seem extremely poor), can you finally get the flaw of your "critic"? You have intuitively assumed that the proposed scenarios were much different than what they really are, because your understanding/analysing skills are extremely low. Now, thanks to these adapted-to-idiots explanations, you should know that the only way to proceed in this case is to rely on a very comprehensive dictionary helping an automated approach to differentiate between what most of people nowadays consider valid and not.

Re:A descriptive example to help those not getting

[CustomSolvers2]

You have intuitively assumed that the proposed scenarios were much different than

I meant "You have intuitively assumed that the proposed scenarios were much more similar than".

Re:A descriptive example to help those not getting

[CustomSolvers2]

Actually, the previous version was also fine depending upon the interpretation of the sentence. A much clearer version would have been "you have intuitively assumed that the wrong groups were more similar to each other and different to the valid one than".

No. Absolutely NOT!

[Anonymous Coward]

Why not? Simple: NI aka natural biological intelligence only manages in a minority to achieve "coding" well enough to solve problems in the real world. In general, AI is demonstrably worst than NI and even seems it will never be even as good. So there's really no way that coding will be obsoleted by AI when NI hasn't even mastered it over the population of all human brains.