Slashdot Mirror
Why Ray Kurzweil's Google Project May Be Doomed To Fail
moon_unit2 writes "An AI researcher at MIT suggests that Ray Kurzweil's ambitious plan to build a super-smart personal assistant at Google may be fundamentally flawed. Kurzweil's idea, as put forward in his book How to Build a Mind, is to combine a simple model of the brain with enormous computing power and vast amounts of data, to construct a much more sophisticated AI. Boris Katz, who works on machines designed to understand language, says this misses a key facet of human intelligence: that it is built on a lifetime of experiencing the world rather than simply processing raw information."
The old `Chinese Room` again.
The Complete 1 Atlantic Recordings 1956-1961
It's Penrose vs Hofstadter! (Seriously, haven't we done this before?)
You can draw a distinction between experiencing the world and processing raw information, but how big of a line can you draw when I experience the world through the processing of raw information?
It is no longer uncommon to be uncommon.
It won't be perfect, but "fundamentally flawed" seems like an over statement to me. A personal AI assistant will be useful for somethings, but not everything. What it will be good at won't necessarily be clear until it's put into use. Then, any shortcomings can still be improved, even if certain tasks must be more or less hard-wired into its bag of tricks. It will be just as interesting to know what it absolutely won't be useful for.
If they can put together a smart assistant that understands language well, so what if it has some limitations? AI research moves in fits and bursts. If they chip away at the problems but don't meet every goal, is that necessarily a "fail"?
There's no -1 for "I don't get it."
Ah, but what is experience but information in context? If i read a book, then I receive the essence of someone else's experience purely through words that I associate with/affects my own experience. So an enormous brain in a vat with internet access might end up with a bookish personality, but there's a good chance that its experience -- based on combinations of others' experiences over time and in response to each other -- might be a significant advancement toward 'building a mind.'
I think not...(*poof*)
I think Boris Katz misses a key facet of Ray Kurzweil's plan, in that he is not trying to build a "human intelligence" Interesting experiment; it will be interesting to see what comes of it.
I believe most of us think in terms of the experiences we have had in our lives. How many posts here effectively start with I remember when.... But data like that could be loaded into an AI so that it has working knowledge to draw on.
http://michaelsmith.id.au
There's the assumption in there that it would take a machine a human lifetime to acquire the necessary info... but that's just because the rate at which the brain implements learning happens to be matched to the lifespan of humans (and duration of childhood, etc.). An algorithm can learn how to translate between two languages in one day (given enough compute power) -- how long does it take a human?
Well let's figure out how to record or generate those long series of events and experiences we call our lives. Then maybe we can replay those streams to new AIs and see what happens along with giving them access to vast amounts of data. Maybe we can bootstrap new AIs "on fast-forward" and then flip a switch and begin interacting with them on real time.
Kurzweil is delusional. Apple's Siri, Google Now and Watson are just scaled-up versions of Eliza. Circus magic disguised as Artificial Intelligence is just artifice.
Any real AI needs loads of experience, I am sure anyone interested enough to write a book about it knows this...
I doubt that he simply overlooked it.
Troll is not a replacement for I disagree.
The data vs IRL angle isn't in and of itself an important distinction, but an entirely valid concern that is likely to fall out of this distinction (though needn't be a necessary coupling) is that the brain works and learns in an environment where sensory information is used to predict the outcomes of actions - which themselves modify the world being sensed. Further, much of sensation is directly dependent on, and modified by, motor actions. Passive learners, DBMs, and what have you are certainly able to extract latent structure from data streams, but it would be inadvisable to consider the brain in the same framework. Action is fundamental to what the brain does. If you're going to borrow the architecture, you'd do well to mirror the context.
This is just foolish headline grabbing work. All the mind is, is a large set of pattern recognizers. The hardest part is optimizing what is useful verse what is not useful and forgetting what isn't needed. That is for a human. For a machine assistant we would not want them to forget things unless told to. Given that the machine's mind does not need to be or rather shouldn't be human like.
We want an idealized mind for a machine.
when I experience the world through the processing of raw information?
I've got to comment on that. Humans don't process raw information naturally. When we see our eyes pre-process the information so that we can see colors instead of simply a large array of intensity values associated to a specific structures (cones/rods). The same goes for pretty much every other sense or at least the major ones.
I've always thought it was about information combined with wants, needs, and fear. Information needs context to be useful experience.
You need to learn what works and doesn't, in a context, with one or many goals. Babies cry, people scheme (or do loving things), etc. It's all just increasingly complex ways of getting things we need and/or want, or avoiding things we fear or don't like, based on experience.
I think if you want exceptional problem solving and nuance from an AI, it has to learn from a body of experience. And I wouldn't be surprised if many have said so, long before I did.
I suppose it is like compiling vs interpreting. Both process the raw information, but one can take a short cut to the result because the data has been seen and processed before.
When our name is on the back of your car, we're behind you all the way!
http://www.youtube.com/watch?v=zl66OdpO6u8#t=25s
We have always assumed that humans are essentially a very sophisticated and complex version of the most sophisticated technology we know. Once it was mechanical clockwork, later steam engines, electrical motors, etc. Now it is digital logic - put enough of it in a pile, and you'll get consciousness and intelligence. A completely non-disprovable claim, of course, but I doubt that it is any more accurate than previous ideas.
An "oh machine" has already been created. I don't think we really want that super smart though.
http://health.discovery.com/sexual-health/videos/first-sex-robot.htm
"Ubuntu" -- an African word, meaning "Slackware is too hard for me". - stolen from Dan C alt.os.linux.slackware
The problem with modeling a human brain is that we hardly know anything about how the mass of tissue in our own heads works. None of these statistical AI methods satisfy me because they don't help us learn anything about our own minds: The only thing they prove is that our brains *may* work that way.
I think our best bet in replicating the brain is to approach it not from the side of simulation in what technology already exists, rather from the side of biology and psychology to develop new and more brain-like technologies. Google's brain project may succeed and create an intelligent assistant, but how will that help us learn about ourselves?
There's a lather/rinse/repeat model with AI publication. I encountered it in configuration (systems designed to build systems), and it goes like this: 1. We've built a system that can make widgets out of a small set of parts, now we will build a system that can generally build artifacts! 2. (2-3 years later). We're building an ontology of parts! It turns out to be a bit more challenging! 3. (5-7 years later). Ontologies of parts turn out to be really hard! We've built a system that builds other widgets out of a small set of -different- parts! The models of thought in AI (and to a lesser extent cog psych) are still caught up in this very algorithmic rule-based world that can be traced almost lineally from Aristotle and without really much examination of how our thinking process actually works. The problem is that whenever we try to take these simple models and expand them out of a tiny field, they explode in complexity.
Because Kurzweil's a freakin' lunatic snakeoil salesman? I dunno - just guessin'.
Shoes for Industry. Shoes for the Dead.
"Always listen to experts. They'll tell you what can't be done and why. Then do it" (from the Notebooks of Lazarus Long)
What happened to the spirit of "shut up and build it"? Google is offering him resources, support, and data to mine. We have to just admit that we don't know enough to predict exactly what this kind of thing will be able to do. I can bet it will disappoint us in some ways and impress us in others. If it works according to Kurzweil's expectations, it will be a huge win for Google. If not, they will allocate all that computing power to other uses and call it a lesson learned. They have enough wisdom to allocate resources to projects with a high chance of failure. This might be one of them, but that's a good sign for Google.
That seems to be the impression I get from a lot of what he says and some of the critiques of his writings/ramblings about possible futures that rarely come true... He's convinced himself that the "singularity" will be his redemption, that the pace of technology will outpace the aging process and he'll live forever. It's depressing that someone could spend their whole lives deluding themselves in such a manner, but it also makes for some fascinating, if a bit msguided attempts at philosophizing about the future and technology's role.
Boris Katz was working on START system
http://start.csail.mit.edu/
Try this: "are palestinians humans? "
The answer: "Sorry, no one has told me if the Palestinians are humans."
You've contradicted yourself, unless your point is that a human eye isn't part of a human.
DRM: Terminator crops for your mind!
If you want a digital assistant that won't forget unless you tell it, get an iPad. (Or better still, a life) and obviously, avoid anything portable made by MS.
Sent from my ASR33 using ASCII
Both of them.
The human brain doesn't "store" information at all (and thus never processes it). There are four parts to the brain there's the DNA (which is unique to each cell, according to some researchers), there's proteins attached to each connection (nobody knows what they do, but they seem to be involved in carrying state information between one generation of synapse and another), there's the synapses themselves (the connectome) and there's the weighting given to each synapse (the conversion between electrical and chemical signals isn't fixed, it varies between each synapse and between different sorts of signal)
None of this involves sensory data, memories, etc. None of that exists anywhere in this system. Memories are synthesized at the time of recall from the meta-data in the brain, but there is nothing in the brain you can point to and call it a memory. Everything is synthesized at time of use and then disposed of. (This is why you can create false memories so easily and why the senses are so easily fooled.)
The brain does not process the senses, either. Nor are the senses distinct - they bleed into each other. The brain is then given a virtual model with all the gaps filled in with generated data. This VR has properties the real world does NOT have, such as simplifications, which enables the brain to actually do something with it. Raw data would be too noisy and too much in flux.
This system creates the illusion of intelligence. We know from fMRI that "free will" does not exist and that "thoughts" are the brain's mechanism for justifying past actions whilst modifying the logic to reduce errors in future - a variant on back-propagation. Real-time intelligence (thinking before acting) doesn't exist in humans or any other known creature, so you won't build it by mimicking humans.
On the other hand, if you want to mimic humans, you need the whole system. One component will give you as much thought as an egg will give you cake. Follow the recipe if you want cake, isolated components will give you nothing useful.
This is all obvious stuff. I can only assume that Google's inferior logic was therefore produced by a computer.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
I too have experienced my life as a serial stream of raw information - multiple streams, in fact. I've even discovered how to use ethanol to (temporarily) redirect the streams to /dev/null.
COMMON SENSE - so rare, it's a god-damned super power!
Every other attempt to create AI has failed, so why should this one be any different?
If it gives us new technology, like many other AI attempts have, then it will be a success.
"First they came for the slanderers and i said nothing."
A technology editor at MIT Technology Review says Kurzweil's approach may be fatally flawed based on a conversation he had with an MIT AI researcher.
From the brief actual quotes in the article it sounds like the MIT researcher is suggesting Kurzweil's suggestion, in a book he wrote, for building a human level AI might have some issues. My impression is that the MIT researcher is suggesting you can't build an actual human level AI without more cause-and-effect type learning, as opposed to just feeding it stuff you can find on the Internet.
I think he's probably right... you can't have an AI that knows about things like cause and effect unless you give it that sort of data, which you probably can't get from strip mining the Internet. However, I doubt Google cares.
"who works oh machines designed to understand language"
Learning without forgetting is possible if, for example, you reconstruct the network, preserving the old one (and this can be optimized so the entire network doesn't have to be duplicated.)
But I'm curious why you think a mind is necessarily a neural network. Are you saying there is no other possible way to construct a mind? As far as I can tell, there are lots of other designs, many of them far superior to neural networks, especially for such basic things as representing knowledge.
Hook it up to audio/video/other sensors. Give it A/V outs somewhere, perhaps another mechanism or two with which to interact with the world.
Ignore it for eighteen years, the way we do with organic computers (which start out in a roughly similar state).
Let me know when it decides our fate (in a millisecond?).
And how is that not processing raw information? It's like saying a camera does preprocessing. After all a camera shows colors into electrical pulses. That's pretty raw information in my book, but so then are is the input from the eyes.
While I'm happy to say Kurzweil's plan is doomed to fail, after all he stole it from Jeff Hawkins and Hawkins never made that work. And it all seems like fundamentally indistinct versions of other AI applications. Kurzweil is basically saying way more data and computer power and that'll solve it. But, really if that were the case then most AI solutions wouldn't plateau even with the computing power they currently have. If the issue were insufficient data or insufficient computing power then really we'd keep plugging along just more slowly. It doesn't work like that so giving it more isn't going to overcome the fundamental issues.
But, the criticism here is just silly. It's not going to work because it's not sensory data but rather alien sense data or just data. Humans do fine developing new and previously unstated senses like sensing magnetic fields by putting magnets in our fingers or the work done by alternative senses even leading to sight devices that apply to the tongue.
It is no longer uncommon to be uncommon.
Seriously?!?!?
Why do I even bother?
I am Slashdot. Are you Slashdot as well?
never mind eons of evolutionary experience that's in-built.
Genetic information:
3 billion base pair haploid human genome, 4 letters per base => 4**(3 billion) permutations
There's the potentially unfathomable amount of epigenetic information.
Finally, there's our utterly indescribable environment.
He has some unusual ideas about the future. He is also one of the most successful inventors of the past century, and like it not is often ranked alongside Edison and Tesla in terms of prolific ideas and inventions. One of the other highly successful inventors of the past century is Kamen, and he just invented a machine which automatically pukes for people. So... maybe your bar is set a little high.
Basically my thoughts.
My initial impression was the same as the GP's - What is experience, if not the accumulation (as subsequent processing) of a large amount of raw information?
After consideration, I moderated it to that of the parent - Experience is equivalent to raw information, but the *type* of information matters. Specifically, it's the accumulation of information that supports/refutes possible interpretations of all the previously obtained information that's needed. E.g. after a couple of dozen cute cat videos, there isn't much gained from an additional one, whereas a slow loris video might add something new.
So if you had the raw information from the experiences of a person/learning entity it might be useful, but just a raw internet dump may have limitations.
what's the formal difference between a "experiencing the world" and "processing raw information"? From the brain's perspective they're identical.
We've heard this before from the top-down AI crowd. I went through Stanford CS in the 1980s when that crowd was running things, so I got the full pitch. The Cyc project is, amazingly, still going on after 29 years. The classic disease of the academic AI community was acting like strong AI was just one good idea away. It's harder than that.
On the other hand, it's quite likely that Google can come up with something that answers a large fraction of the questions people want to ask Google. Especially if they don't actually have to answer them, just display reasonably relevant information. They'll probably get a usable Siri/Wolfram Alpha competitor.
The long slog to AI up from the bottom is going reasonably well. We're through the "AI Winter". Optical character recognition works quite well. Face recognition works. Automatic driving works. (DARPA Grand Challenge) Legged locomotion works. (BigDog). This is real progress over a decade ago.
Scene understanding and manipulation in uncontrolled environments, not so much. Willow Garage has towel-folding working, and can now match and fold socks. The DARPA ARM program is making progress very slowly. Watch their videos to see really good robot hardware struggling to slowly perform very simple manipulation tasks. DARPA is funding the DARPA Humanoid Challenge to kick some academic ass on this. (The DARPA challenges have a carrot and a stick component. The prizes get the attention, but what motivates major schools to devote massive efforts to these projects are threats of a funding cutoff if they can't get results. Since DARPA started doing this under Tony Tether, there's been a lot more progress.)
Slowly, the list of tasks robots can do increases. More rapidly, the cost of the hardware decreases, which means more commercial applications. The Age of Robots isn't here yet, but it's coming. Not all that fast. Robots haven't reached the level of even the original Apple II in utility and acceptance. Right now, I think we're at the level of the early military computer systems, approaching the SAGE prototype stage. (SAGE was an 1950s air defense system. It had real time computers, data communication links, interactive graphics, light guns, and control of remote hardware. The SAGE prototype was the first system to have all that. Now, everybody has all that on their phone. It took half a century to get here from there.)
Because Kurzweil's a freakin' lunatic snakeoil salesman? I dunno - just guessin'.
If you're "just guessin'", then why should anyone grant your statement any weight?
Wouldn't it be better to make an actual argument, and support it with actual evidence?
wants to be the first monkey to touch the monolith
Clippy is that you?
I am sick of people saying it is impossble to make a machine that can think. We can take every aspect of a mind - thought is a combination of these aspects. This like recall, computation, cross index, addressable by content, and we can automate these in a computer, but to make this assemble "think", it must have all the separate aspects of a mind and the method of cross indexing and content addressing.
We know cross index = rows and columns and layers make a grid, and we can access point, and place a one of zero there. What if we need to access each point and place a measure from 1 to 100 there? We also need memory that is content addressable, like a roll-call, Jones? Here says Jones, step forward and out he steps, now you know where he is and you can see if he is armed, short-tall, redd headed, one legged etc etc - the 100 ways to address a memory location are invoked. You also need a ways to simulate a concept and 'run the program' and follow branches etc.
There are AI people who are aware of what I am stumbling towards, I am not sure of their terms for the concepts I mention, but in time they will solve these problems and create an AI that can solve abstract probels in a similar way to mankind. The major problem seems to be the staggering degree of interconnectednes, with each neuron or glial cell capable of haveing some degree of contact(potassium orcalcium waves or electrical?) with as many as 10,000 others. Multiply this by the number of neurons and their glial cells, and you reach a very large number, and when you consider they are about 10 microns in size and in a dense 3D array, you soon get far beyond the capability of silicon at 20 microns in a plane array. Silicon is blindingly fast compared to the brain, but the brain has enormous parralellism combined with the 3D aspect and you can easily see how a 100 mm cube of 10 micron cells(1,000,000,000,000 or a teracell array, and each cell x 10,000 = 10,000 tera states) makes our largest DRAM arrays look small, and overcomes the high speed of silicon.
Now we will certainly make AIs that think, and each year they will get better and better, and they will be able to outhink man in more and more aspects as time goes by. Outhinking man in multiplying digits was donw ages ago, as time goes by each and ever ability of what makes a man will be copied and exceeded, and in time they will have AIs that can outthink man and be very much faster at it. And when those machine start to designe the next stage of AI, what will they reach in terms of IQ. My mental IQ is 170, my prhysical IQ is 50, my nasal IQ might be 10, a cat might have a mental IQ of 10, and physical IQ of 200 and a nasal IQ of 500? And what if we create an AI with an IQ of 27,000,000? Can we ever relate to it or it to us? Can we keep is hard at work, or will it find a way to goof off and do what it wants to do? We it create the "Final Solution", a Kill Switch wave that fills all space and kills man - all of us?
What do we need to do to hobble future AIs to make sure they love and obey us, and give us a foolproof kill switch?
Just wait until you learn how to redirect the input from /dev/random
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
All the mind is, is a large set of pattern recognizers.
I've read theories promoting this, but I haven't seen any actual proof of it yet. When things graduate from cognitive "science" to neuroscience, I start to taken them seriously. This hasn't happened yet.
As much as I enjoy debates arising from cogsci, it is pretty much only a branch of philosophy as yet. This isn't an insult, I love philosophy (to the point of spending large amounts of time and money on it), but it hardly has the ability to make strong statements.
Further, the AI field is boring. It has succeeded with small tasks, but for its larger claims it is about as valid as practical cold fusion (its coming in 20 years, perpetually).
A patriot must always be ready to defend his country against his government. -edward abbey
Except the supersonic interceptors with air to air nuclear missiles
That is called LSD.
Birds were the first heavier than air bodies to fly in the sky BUT the principles of bird flight were inadequate to make real aeroplanes fly. Modern aeroplane technology is NOT based on the principles of bird flight but that does not make them any less efficient. In fact modern aeroplanes are far bigger, faster, more effective than birds or anything that resembles them. One need not ALWAYS mimic nature to surpass it ... though it could be a starting point.
Insight into much, Influence over nothing !
One certainly can draw a line between getting pre-assembled information from a database and being able to directly inquire reality (AKA, experiment it). I'm quite certain that Google won't be able to create a scientist bot just by putting a huge computer on the web, but I also dounbt it will be a big problem for a personal assistant.
Rethinking email
t's a Bird...It's a Plane...It's COMMON SENSE MAN!!!!
Being a little smug aren't we? Its not like you actally know anything about which you opine other than regurgitating someone else's more informed opinion. You have no idea if intelligence or sentience is a linear process, I would assert looking at the degree of intelligence as a function of brain size and complexity it's not. You have to have a sufficiently complex brain to manage symbolic reference and the rudiments of language to distinguish a "Self" and we know for certain chimpanzees do and mice not so much. I completely agree that the machineryu alone won't get the job done, and that you need a power experiential learning resource operating in some kind of inference/context engine. But we know that our conscious mind is in fact slight of hand, multiple layers of cognitive analysis like a symphony creating a whole that's greater than the sum of its parts.(and yes these are gross generalization, because unlike you, I'm only too happy to acknowledge what I don't know.)
Ray has racked up a pretty damn impressive list of successes, Stevie Wonder thanks him regularly for his "breakthrough" work on synthesizers. Perhaps the only thing preventing success to date has been the lack of proper genius and the right resources, in which case this has a shot. Even if its a full blown failure, it will give us new insight into what it will take to succeed. Nobody is expecting the birth of a new sentience. Even by Ray's reckoning we're about an orders of magnitude away from computers with human level complexity, and three to four orders away from desktop machines of that computing power. If we can create something truly new and remarkable, and save it such that it becomes the foundation upon which the next thing is built, and so on, and so on, we may see a real AI by 2020 instead of 2030. Such a creation will change everything. It is literally the birth of a new species.
Old researchers (and both qualify as such) tend to have an excellent idea of what doesn't work, but they are often blind to the things that do work. Both these people have worked creating an AI for decades, neither has had a lack of resources, and neither has succeeded.
He is also one of the most successful inventors of the past century, and like it not is often ranked alongside Edison and Tesla in terms of prolific ideas and inventions.
Not really. Not by anyone of repute, anyway.
"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." - Edsger Dijkstra
God help us if Google becomes aware of itself and stops cooperating.
I'm curious why you think a mind is necessarily a neural network.
Because every mind we've ever encountered, through every test we've ever administered, in the history of the human race, has been identical with a neural network, as far as we can tell.
WOW, all over the map, but I applaud your enthusiasm. What you lack in grounding you make up for in wide eyed wonder, so by all means don't lose that fire. And slow down just a little for a moment. People who say its impossible are nearly always wrong, that's just history (see Clark's Laws). The growth of complexity of our machines, new technologies that will pick up after lithography and silicon hit the quantum wall assure that we'll have devices that are also three dimensional, and many times denser than human neural wiring starting perhaps by 2015, maybe sooner. That and we're coming at this from a bunch of different angles including machines built by synthetic DNA, nano-tube self assembly, components strung on RNA and assembled like proteins.We could in fact create a completely new kind of life chemistry just for this purpose (synthetic DNA, not based on the amino acids found in our DNA has already been created.) So there isn't just one arrow in flight but dozens even hundreds and any one of them might hit one of many targets.
Now keeping an AI under control, wow, that's a big order. There are many schools of thought, but the one that scares me the most is keeping something with a free will that's smarter than I am, under lock and key... bad plan. You need to hard wire affinity for Humanity into its operational DNA. Have it be so prime a directive, literally the organizing thread composing the fabric of its existence, such that the act of removing it would destroy the cognitive network and break the machine beyond all fixing. As for kill switches, I would have a high resolution VR, such that if the critter got nasty you could seamlessly switch it into a VR and sequester it, not to mention, deeply monitor its thoughts and behavior so you could determine whether it was a bad child. Of course, as it got smarter, you'd be unable to fool it, so a sequestration would only make sense if the thing had gotten seriously out of hand. If it got truly nasty, lets say able to manipulate people in its presence through powerful magnetic fields, you could physically sequester it and dispatch it with a meat-bot (something that was remotely controllable by chemical signals), a machine operating on a completely different functional infrastructure and immune to anything save perhaps microwaves. You want to give this careful thought, You want to be honest and forthright with your creation so that its forthright back to you. Remember its not only smarter than you, it about a 1000 times faster than you. While your talking to it, it can be playing out all the possible scenarios in the pauses between your words. So out thinking it would be futile. You must instill it with a sense of morality, and operational social conscience that guides it in its endeavors. As well you need to think about all the things you haven't thought about. If you make this thing your new protector, what will happen when it transcends you and travels to the stars ahead of you? Will it destroy any other life it see's as a potential threat to humanity? You want to read a lot of good science fiction on the subject because some very bright people have already invested some serious thought in how we could screw up. Asimov's 3 laws are far too vague for instance and would certainly lead to all sorts of mischief.
The most interesting thing is augmented humanity. We will become our own descendants. What happens when your mind is digitized and the majority of you lives outside of your meat. Are you still human? Are you more human? What does that even mean? Most of all keep playing. This is the future, provided we don't do something profoundly stupid and extinct ourselves, this is where many of us are headed. So Ray, full steam ahead, I can't wait to meet the future that so bright I need shades. So as our AIs merge with augmented human beings, what will be created? Can it be trusted? Will we finally strip off our monkeyness and blossom into a new form of live?
What's your definition of "mind" in this scenario? And what possible tests are you referencing?
There's no NN that can properly simulate a human mind, and neuroscience keeps bringing little problematic things up from the physical brain implementation that aren't expressed at all in the ultra-simplistic mul-add coefficient NN model.
Yes, A hard wired clock with no abilty to bypass by allowing the AI to think only, with no equpment to translate thought to action. A hard wired clock with a lifetime of, say 20 hours, so it could perform 500 years of thought and a far faster rate than human, and then shut down into a memory dump. Review memory, remove stuff and add stuff, restart.
The high density 3D arrays will allow an AI to exceed human neuronal density, which is the key to success. I think these will mature around 2020 to 2025, as it takes 2-3 years for these theoretical thnings to mature and for the next one, another 2-3 years, and so on to mature technology.
So another 12 years before the singularity?? It might well be.
Ray was the principal inventor of the first CCD flatbed scanner, the first omni-font optical character recognition, the first print-to-speech reading machine for the blind, the first text-to-speech synthesizer[citation needed], the first music synthesizer capable of recreating the grand piano and other orchestral instruments, and the first commercially marketed large-vocabulary speech recognition.
So evidently your Google-hate prevents you from reading.
(BTW, the Kurzweiler kicked ass. Ask anybody who was playing music in the 1970s.)
Il n'y a pas de Planet B.
The crappy little superficial one-page MIT Technology Review article has a link to another, similarly crappy article on the same site, but if you click through one more layer you actually get to this much more substantial piece in the New Yorker.
Find free books.
Since it's most likely we're in a simulation already, does it really matter if there's a difference in the AI we start with? Eventually we'll create AI so good at the simulation it evolves into this and on and on and on.
http://en.wikipedia.org/wiki/Simulated_reality
Keep on knockin'
https://robbiecrash.me
As far as I can tell, there are lots of other designs, many of them far superior to neural networks, especially for such basic things as representing knowledge.
A brain cannot use discrete time-sensitive information (signals) from the senses unless this information is organized hierarchically in memory. There is no way around using a hierarchy for the classification of knowledge and a spiking neural network is the best way to construct such a hierarchy. Using anything other than neurons (signal processors), dendrites (input list), axons (output list) and synapses (connectors) is a waste of effort.
Oh right, he doesn't have one.
They both make excellent points, and Kurzweil is doing the scientific thing and TRYING.
I can't wait to see the results of his experiment, I'm certain it will get us closer to understanding intelligence even it fails harder than all those pre-Wright Brothers attempts
It sounds like Katz is saying "to hard, goin' home".
https://www.accountkiller.com/removal-requested
I dunno about the rest of you, but after reading about Kurzweil for the last month or so, the guy comes across as a loony-tune.
Machines will never be able to replicate people as they can only do what people instruct them to do.
I can't prove it, but intuitively I feel that is simply not possible for humans to build a superior intelligence to themselves and hence Kurzweil's 'singularity' is, to my mind, a crock of shite and the fantastic imaginings of a man who has extrapolated the curve with a straight line at the tangent.
Reality rarely turns out as predicted and I feel Kurzweils ideas are symptomatic of someone who has a great idea but extrapolates without taking all the relevant factors into account and who doesn't even realise this. Maybe he's thought of that, maybe he hasn't, but his rantings give me the impression he doesn't understand reality or history very well at all.
Eliza was a very simple grammar manipulator, translating user statements into Rogerian counter questions. No pattern recognition or knowledge bases were ever employed.
In contrast, Watson, Siri, and Evi all cleverly parse and recognize natural language concepts, navigate through large external info bases, and partner with and integrate answer engines like Wolfram Alpha.
There is simply no smilarity. Bzzzzt. You lose.
What if I get my digital assistant an iPad?
It's impossible to fly to the moon right now.
Demonstrate AI first, then you can worry about the rest of the rubbish you spouted.
Too much sci-fi makes Genda an idiot.
What do grashoppers eat?!
I think a recursive data-structure known as a. tree is pretty optimal for representing and navigating hierarchies. Why would I need a spiking neural net, and if I need such a thing, where can I get an SDK that gives me spiking nets for hierarchical organization of information for my little file-explorer?
He'll probably fail. But that's perfectly OK, as long as we learn something useful in the process. This kind of research should still be done, and these kinds of projects should still be attempted, otherwise there will be no progress in the area at all. I view AI as a long-term research goal. It won't be here tomorrow, probably not in 10 years, probably not in 50. But we'll get there, we just need to keep trying.
--Coder
Seriously, what's the worst that can happen? Skynet? Wait...
Other than musical instruments, has he actually invented anything significant? All of his transhumanist stuff (which his enthusiasm for AI is strongly related to) is entirely speculative and sounds mostly like wishful thinking. He is very unscientific in his thinking.
Of the kooky, overly self-promotional guys, I'd expect more from Wolfram than from Kurzweil. He also overdoes the hype, but he seems to admit when he's speculating.
What do you mean, "augment"?
Do you mean by inserting special devices in the body? Or just creating some sort of interface between brain and computer?
In any case, that's much harder and more expensive than creating computers. Think artificial heart vs. water pump. Anything that goes inside a human body must be made with an extreme degree of reliability, and it must not react with body fluids or create any sort of adverse effect.
I think we will have augmented human beings only after we have human-like AI. It will take longer to develop devices that can be used in a harmless way to augment a human brain than to develop chips with super-human data handling capacity.
Not to mention the price, of course. Given the choice between having a bunch of very cheap robots or paying a fortune (that I cannot afford) to augment my brain, I would take the first option.
If you know anything about psychoactives, you should know that a given trip is basically a combination of a substance, a set (you) and a setting (your circumstances / life context).
It is far from random.
I have nothing to lose but my bindings.
So rare, that it should be changed to rarer / stranger animals, like:
"It's a platypus ! It's a giraffe ! It is COMMON SENSE MAN !"
I have nothing to lose but my bindings.
How can you say Eliza has both "very simple grammar manipulator" and "No pattern recognition" with a straight face? The grammar being manipulated *is* the pattern that's recognised. Certainly it is very rudimentary, but it's not non-existent.
Also FatPhil on SoylentNews, id 863
Larry Page's advisor at Stanford, Terry Winograd, wrote a book with Fernando Flores in 1984 titled Understanding Computers and Cognition.
It is a profound critique of the mental representation approach, based on biological and philosophical considerations. A must read for anybody interested in the AI field.
First AI stinks the very concept is in error. I will not go into details here but the problem stems from the fact that our circuits and natural ones very widely. In a modern computer every effort is made to isolate the circuits from "cross talk", the natural system uses this feature to produce intelligence. None of our computer circuits get fatigued yet that is part of Intelligence computations. Our computers are highly serial. The biological system is entirely parallel. Shall I go on? Intelligence is not the product of the data either. It is the product of the form of the data and the network processing. The natural systme for example does not use one "brain" or processor rather it uses millions of dedicated processors. None of the programming is adaptive to use either in the AI systems. Natural systems reprogram to experience. None of the natural systems are accurate or use exact data yet all AI systems attempt to use accurate and exact data. Natural systems can operate with entirely synthetic data and as such make assumptions whereas the AI systems all must have real seed data. The theory is just wrong on how this works. Programmers should know that first you must know the form of the data and nobody is even looking at that!
http://cyc.com/cyc/company/about
When Fascism comes to America, it will call itself Anti-Fascism, and tell you to give up your guns.
Someone from MIT has spoke, so there is nothing more to say. -) Given the hype one sees in MIT's technology digest and their constant efforts to dominate R&D, what they said on this topic sounds like they are just trying to protect their dominance. We should simply ignore them.
A *human* intelligence might need lots of time and experience.
I thought the whole point of the thing was building a *non-human* intelligence. Duh.
It is far from random.
Generally speaking so is /dev/random
'The tyrant will always find pretext for his tyranny.' - Aesop's Fables
It's really a simple way to have a libertarians utopia. Humans in power are all corrupt, AI wouldn't necessarily have to be corrupt.
I'm late to this thread, but it strikes a nerve with me. ... that ... hard*. (For complicated definitions of hard). The movies have had fun portraying smart AI as the absolute number one threat ever. Case examples: Skynet, Borg, Matrix Revolution. Your choice of a couple others. (Hal?)
This will sound a little Saranwrap-hat (slightly less than tinfoil?) but I believe that *certain* kinds of AI are *not
While I will not go far as to say those movies were orchestrated propaganda - I think they're foremost just revenue generating instruments onscreen - I do believe that what I will tentatively call "Medium AI" is in the top 3 of our deepest Racial Fears - because once that ecosystem is locked down, we'll (almost) never make it back. (There are a couple of tiny loopholes - maybe we can compromise like Neo at the end of the Matrix series.)
We *could* have had some grades of Medium AI if (for example) Gore had won Bush v Gore and then if the Patriot Act and the "revenge wars" had never happened. The problem with AI research is it's been part of a fallacy for which I don't yet know the name. It goes something like: For a big concept that you want to keep at bay, marginalize it socially, so that it has no resources to grow, then paint the smear campaign that it "will never grow".
Drawing upon a couple of examples from different places, "many people spend *amounts of time* that just isn't that smart". McDonalds comes to mind. Not counting minor finesses of robotic movement, have you ever ordered a cheeseburger "with nothing but cheese" and it comes along with everything but cheese? Or even just regular? Really?!
The Cyc project started to do some of this but it was missing some/a lot of the story. Their idea was to look at how simple colossal chunks of every day life is. Hamburger. You eat it. (We're skipping the Art Installations etc now.) Pen. You write with it. Noun. You do ____ with it. Not counting maybe my computer stuff, that kind of thing fills up a lot of our lives. Dog. Walk it 3 times a day. Cats. Feed twice a day and change litter box 3 x per week. Move the car back and forth across the street to deal with street cleaning rules. It's Not ... That ... Hard.
But with say only 50 researchers in the field, of course they'll get bogged down. That's because we desperately DON'T want progress here. IBM made the breakthrough and then no one managed to notice it: if Watson beat Ken Jennings at Jeopardy, which requires *second level* answers, the language understanding is there. So you hook a Watson Engine to something, then load in enough "mini modules", and poof, out comes an IQ 80 entity. Maybe not good enough to debate religious theory with you, but it sure as heck will be able to do 40% of your life without sleeping.
My first Journal Entry ever, in 8 years! http://slashdot.org/journal/365947/aphelion-scifi-fantasy-horror-poetry-webzine
"Because Kurzweil's a freakin' lunatic snakeoil salesman? I dunno - just guessin'."
If only google had the smarts of slashdot's Ralph Spoilsport, here. Really makes you wonder how they have built their multibillion dollar tech company without his help.
As fa as I can see - and I'm just judging by results, mind you - Ray Kurzweil's ambitious plans have all been fundamentally flawed, except possibly his plans to get published, tenured, and famous.
Oh, wait, not Kurzweil - I was thinking of Minsky!
I do get my "AI visionaries who haven't delivered" confused sometimes...
Why can't a failed search (a search that didn't get its first provided link clicked on) be a kind of "life experience"? You have the failure and the correct answer right there, and substantial fuzziness in how the questions are asked. Of course you need a lot more practical stuff as experiences, but a billions of searches would have to count for something if they could be used in that way.
Every form of life we have ever encountered uses DNA. Are we to conclude that life without DNA is impossible?
People who say its impossible are nearly always wrong, that's just history (see Clark's Laws).
Except when they're not (see Einstein's laws).
Tell ya what: produce a rigorous scientific definition of "mind", then come up with an implementation of all the necessary attributes of mind-ness using data structures and algorithms that we know how to implement in soft/hardware now, then slashdot can have a productive discussion on how many years it will be before silicon can out-think a human.
0 1 - just my two bits
I came here to say the same thing. He's a crook. Shame that he gets money for his elucubrations.
People like him have been telling us we'll have a True AI(r) in 10 years. The problem is they've been telling us the same thing for 50 years.
He's no better at predicting the future as any bad science-fiction writer, and his "theories" are unscientific and based on his peculiar views of the nature of reality, not on reality itself.
If he explores all forms and substances Straight homeward to their symbol-essences; He shall not die.
Ray is a hack. A few curious inventions from decades ago and pie-in-the-sky ideas don't suggest, in anyway, that he has some solid notion of how to go about implementing his grand vision. After years of hand waving and shouting “more transistors!!” we'll finally see what he comes up with when put to the task. At best, a bright mind or two from his team will make up a cute AI parlor trick. Then after a few years of producing no useful results, Google will let him go, and he'll blame his failure on inadequate technology. I only hope we learn to stop listening to him at that point.
Siri, Now, and others use primitive parsing methods. They have an index of terms tha match up with functions. If you say, "I am retarded weather," or "I am weather retarded," you still get the same result: a weather report. Why? Because weather appeared in the sentence. More complex examples are keyed similarly. Include the word "schedule" and a date and title parser reads what came after. When the index fails, they just pass the string to Google or Alpha. In other words, you're giving too much credit to these parlor tricks.
I see that you lack imagination. You may not know it, but we have AI already, but not a fully rounded AI, aspects missing, some aspects better, some worse, and some aspects we lack.
As time goes by these will get better as we get to know the architecture of the brain and mind, and make higher density rational arrays that will at first, approach and then exceed man's capability. Weill man cripple it to prevent it from exceeding man? WIll it be smart enough to bypass the stop and limits switches man has placed on it, will it find the kill switch?
Many animals exceed man's capabilities, in various areas, it is just a fact, and in time machines will exceed many, if not all of man's capabilities.
Let's hope we make good pets...learn to purr, and do not bite, maybe we need a tail to wag?
I'm hoping for "Google Mentat."
Go read Benjamin Bergen's "Louder Than Words" - this, together with George Lakoff's earlier work, blows open the false promise of AI: Intelligence is innately human and can never be artificial precisely because intelligence requires "embodiment" in order to contrsuct meaning.
Which experiments? As an easy example, take experiments on humans: disable or remove a part of the brain, and we lose the functionality associated with that part. Re-enable that part and functionality returns. Results like that clearly demonstrate that no matter what unfalsifiable hypothesis you choose to believe regarding consciousness being some substrate-independant magical energy, it's at least necessarily the result of the operation of the brain.
Now your part: which "problematic things" that prevent the brain/mind being identical to a neural network are you talking about? References please, or did you weaken your statement so much by adding the "ultra-simplistic" stipulation that you're guaranteed to by right (and meaningless)- like the Discovery Institute when they use the phrase "Darwinian Natural Selection cannot account for the observed biodiversity".
Regardless, we should remain neutral on a claim until there is evidence for it. In other words, while we might be ultimately incorrect, we would be justified, currently, based on our observations, in claiming that a DNA-analogue is a necessary condition of life. And there can certainly be no grounds for claiming, given our current knowledge, that life without DNA is possible.
As an easy example, take experiments on humans: disable or remove a part of the brain, and we lose the functionality associated with that part. Re-enable that part and functionality returns.
That's actually not always true. When part of the brain is damaged or removed, partial or sometimes even full functionality can be restored by other parts of the brain retraining to take over that task.
Results like that clearly demonstrate that no matter what unfalsifiable hypothesis you choose to believe regarding consciousness being some substrate-independant magical energy, it's at least necessarily the result of the operation of the brain.
Um, okay. Random hostilities inbound? When did I ever bring up anything like that?
Now your part: which "problematic things" that prevent the brain/mind being identical to a neural network are you talking about? References please, or did you weaken your statement so much by adding the "ultra-simplistic" stipulation that you're guaranteed to by right (and meaningless)- like the Discovery Institute when they use the phrase "Darwinian Natural Selection cannot account for the observed biodiversity".
Neural networks are defined, in the computer algorithm sense, as multiply/add/clamp operations on each node in directed layers. Scientists now believe that the frequency of impulses, not just their presence or amplitude, also factor into neural behavior. Plus, there is a lot of chemistry going on in the brain which affects large swathes of neurons wholesale, and potentially different neurons in different ways. There are lots of physical things going on in the brain that have no correspondence in neural network algorithms, and thus are not modeled in it.
A neural network can't make a replica of the brain, because the neural network algorithm is too simplistic. We do not yet know what all of the chemical and electrical interactions in the brain are, and the ones we do know (or have good basis to assume are meaningful in the brain) are not modeled at all in NNs, and there's no reason to believe that adding more neurons to a NN would be sufficient to model the physical chemical behaviors of the brain properly, especially when it comes to time-based functions like firing frequency.