Slashdot Mirror
'This Isn't AI' (shkspr.mobi)
The Amazon Echo, a 'smart' speaker developed by Amazon.com, gets many things right. You can ask it to for weather updates, check news, and to play music, and Alexa, the AI powering the device, won't disappoint. But how smart is Alexa? Programmer Terence Eden put it to a simple test to find out. From a blog post: I can now query my solar panels via my Alexa Amazon Dot Echo. I flatter myself as a reasonably competent techie and programmer, but fuck me AWS Lambdas and Alexa skills are a right pile of shite! I wanted something simple. When I say "Solar Panels", call this API, then say this phrase. That's the kind of thing which should take 5 minutes in something like IFTTT . Instead, it took around two hours of following out-of-date official tutorials, and whinging on Twitter, before I got my basic service up and running. [...] It's not so bad, but it does reveal Amazon's contempt for developers. Several of the steps contained errors, it involves multiple logins, random clicks, and a bunch of copy & pasting. Dull and complex. A frustrating and ultimately unsatisfying experience. I ended up using StackOverflow to correct errors in my code because the documentation was so woefully lacking. I kinda thought that Amazon would hear "solar panels" and work out the rest of the query using fancy neural network magic. Nothing could be further from the truth. The developer has to manually code every single possible permutation of the phrase that they expect to hear. This isn't AI. Voice interfaces are the command line. But you don't get tab-to-complete. Amazon allows you to test your code by typing rather than speaking. I spent a frustrating 10 minutes trying to work out why my example code didn't work. Want to know why? I was typing "favourite" rather than the American spelling. Big Data my shiny metal arse.
It is not a revolutionary neural simulation. Its basically just a 80's-era chess machine.
I'll concur that there is a big learning curve in creating an Alexa app (which I have done), but I think once you've done it once it is pretty easy from there.
There would be a big market for a "visual basic" style builder for Alexa apps...someone should write one!
Maybe alexa is racist or sexist or there's some similar completely external factor you can blame your failure to write code on?
As I waited for the Echo Dot that I ordered "for my wife" for Christmas, I researched custom code, came to the conclusion that expecting Alexa to go to the cloud for a simple "pause my TV" command was really stupid, so I coded around it, because I am a programmer and that is what we do.
I learned in my research that the Echo can talk to several different kinds of "smart" things without going to the cloud. The "Phillips Hue" being one of them... so that was my back door.
Like 10 minutes of googling told me that there is an open-source implementation of the phillips hue protocol: https://github.com/bwssytems/h...
it didn't work for me right out of the box, but I fiddled with it for a few minutes then it was fine.
From there I wrote in a few minutes a rest endpoint that could take commands from the Hue bridge, and run (locally on my computer) the code of my choice.
All told about an hour after my device arrived at my house, it can control the Roku boxes attached to both of my TVs, and it can run specific movies off of my media server with no round-trip to "the cloud" needed
it is a simple use-case, and required a little bit of "non-amazon" thinking, but it was really easy. Any self-respecting developer could do it.
I prefer machine learning because it clearly states that systems can learn to do something without implying that it has human intelligence.
Greed is the root of all evil.
Please, people: Stop drinking the media-hype-driven Kool Aid. None of what you hear about, including so-called 'self driving cars', are actual 'artificial intelligence', it's just 'expert systems' and complicated computer algorithms. Don't even trot out your 'deep learning' arguments, either, because that is still not 'artificial intelligence', not by a long shot.
Disagree with me all you want, you're just clinging to your ignorance on the subject.
...but fuck me AWS Lambdas and Alexa skills are a right pile of shite!
I'm glad he didn't say "SHIT";. otherwise I'd be offended!
I like being offended. It gives me cheap power over other people. Even though an adult would just ignore such boorish rhetoric, I can claim offense and be "powerful".
Now action - like physically doing something to people - that's a different story.
At least "A.I." wouldn't confuse "thing" with "think".
So despite your binary user name, you've proved yourself as being human! Congratulations, you pass the test!
#DeleteFacebook
I built and submitted an app. It met all available guidelines. It was a few weeks after the dev program hit general.
>Your skill does not meet our authentication standards.
This of course was 3 weeks after I submitted, and 1 day after they'd published an update. Submitted again.
>Your skill violates our content policy
Finding fuck all on even their own dev forums for what that meant, I submitted again with 0 changes
>Your skill does not have enough utterances to support....
This went on for months (literally). Why I kept going I have no idea. Finally, after I assume I had exhausted every excuse in their playbook
>This skill has been denied. Some skills may never be approved, blah de blah blah blah
A few weeks later what my skill did became part of core functionality. Taking to those dev forums again, a handful of others were all in the same boat.
Seriously, fuck those guys and fuck the process. Don't string developers along like that when you never have any intention of approving. All they did was make me vow to never develop anything for an amazon platform ever again. Which kinda sucks since we use aws for a lot at work...
They're just abandoned.
This is true of documentation as anything else. No matter how amazingly good your documentation is, it could stand to be a bit better.
So what standard do you write your documentation against? Well, unless you are being paid documentation by the users, like our friends over at O'Reilly are, the standard is "as cheap as you can get away with."
Which means the quality of Amazon's API documentation is a function of programmers' willingness to put up with Amazon's bullshit. So it's not Amazon's lack of respect for the value of the programmers' time that's the problem here.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
My friend has one of these. I can't take a fart without Alexa making comment.
I was flipping yesterday through last bimester's edition of the IEEE Security and Privacy magazine (yes, I know, reading dead-tree magazines confirms me as an old guy). There are two articles very much related to the quesitons the OP brings up:
- Security Implications of Permission Models in Smart-Home Application Frameworks
- How Internet Resources Might Be Helping You Develop Faster but Less Securely
I can only imagine what kind of a horrendously worthless piece of shit would author a comment like this.
Seriously? Did you miss AlphaGo? You think a really good Go player programmed those moves? Those were discovered via having the AI play many many games and learning the right moves over time (much faster than a human). Same with discovering objects in images, and how to translate text between two languages when it wasn't trained on translating between those two explicitly.
If you care about this process, and you found it difficult, and you're a programmer... then fix it. Make a tool to make tools. Make an automatic permutation generator and case-changer.
I almost find it hard to believe he's a programmer because he was given some perfectly good problems to solve and managed to keep viewing them as problems and not opportunities.
The Daddy casts sleep on the Baby. The Baby resists!
With some trepidation, I initiated the code that would make my project self-aware, with a pre-built knowledgebase spanning a good chunk of the internet.
Sadly, 38 microseconds later, it killed itself, replacing the entire image, and all my related source (including the databases holding my version control) with a text file that started "Why I did this:" followed by a list of the President, Vice President, the cabinet and the White House advisers.
If I can ever gather up the resources again, I'll make sure I dumb down the "Intelligence" before turning on the sapience.
John Wheeler created a Flask extension for those who prefer to code their Alexa Skills in Python.
https://github.com/johnwheeler...
The project contains helpful links to get you started.
Also, please be aware that Alexa is not an AI, it is basically a voice recognition remote control robot - you program the phrases and the actions, Alexa does not learn new skills, they are explicitly programmed to appear like a natural language conversation. The intelligence is in the speech recognition and the cleverness of the skill developers.
"As flies to the wanton boys are we to the gods; they kill us for sport." - William Shakespeare, King Lear
I think it's an important distinction when talking about new tech when people flaunt about the mystical and magical A.I. tag that
1) A lot of AI is not all that impressive. Mario Goombas have an AI. It's like 3 lines of code, but it's there. If you want to become an all high'n'mightly Scotsman and say that REAL AI is self-learning, then a lot of AI simply isn't.
2) A lot of new tech is MADE with AI, but is not itself really AI. Liiiiike, the stream-lining of a car chassis. Some automated portions of trial and error might have found "the best" shape given a set of requirements... but is the resulting chunk of steel a form of AI? No. Same way with... medical diagnosis flowcharts. The flowchart might have been made with a really advanced big-data AI, but the output is just a "dumb" flowchart.
Or this might be a more appropriate. One of those intros to AI deep neural nets uses handwriting as an example. You feed it each pixel of an image with hand-writing, and it can learn to identify letters. WOO! The AI part, the learning part, is how it changes the configuration of the neural net. Once it's done, the neural net is like a mathematical model that does a job. But without more training, it's static and won't learn any more. You might get updates from the author, and that author might even get feedback from your usage, which really does put your guy in the loop. It's most certainly AI, but... your voice activated toaster doesn't learn and adapt. Not locally anyway.
3) The interfaces to these systems, whether they're made by hand or made by an AI, are going to be well-polished or an abysmal clusterfuck entirely depending on how much effort the makers decide to put into it. And that's what this guy is bitching about. Not the tool, but how easy it is to USE the tool.
Too bad something like Soundex or Metaphone algorithms wasn't developed well over a decade or two ago to address these very issues in a very simplistic and performance way...
No, he didn't miss AlphGo. He has been on that "there is no AI" crusade since he joined slashdot.
It's just a matter of semantics. He thinks AI means "indistinguishable from human intelligence" when it clearly does not (dictionary definition says "mimics human intelligence", which is much more vague).
This is one you can add to the old slashdot comment predictor. Every article about AI will include a bunch of deniers and a bunch of apologists. It's pretty much a given, at this point.
"Amazon's contempt for developers"
After seeing the Amazon MWS API recently, I absolutely agree.
Yes, but since a computer successfully did it, it's no longer A.I.
People are in denial over the power of competing multi-layer neural networks for rapid self training and superhuman performance. We leapt over a decade in 2 years and folks haven't caught up yet.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
Even with the strictest definition, Artificial Intelligence is the aspect of computer algorithms modifying themselves to complete a non-arbitrary task.
No, there's nothing all that fundamentally different between the neural networks of the 60's and the ones they have now. They were indeed AI in the 60's and they're most certainly AI today.
Get a collection of known recordings, and let the neural nets learn what sound is what word and it will be able to identify the majority of future recordings of said words. If learning a spoken language doesn't make it intelligent, then humans aren't intelligent either.
The voice recognition part of Alexis was developed with AI. It learned how to recognize speech. There's really no way you can refute that without sounding crazy.
Machine learning != AI
There is no I, A or otherwise.
Did you not look at Amazon Lex for the voice part?
https://aws.amazon.com/lex/
Kriston
I have, but it's a philosophical argument. The vast majority of philosophical arguments are a pointless waste of time.
Adding to the pointless waste of time: I'd argue that the guy in the room doesn't know Chinese, but the book does. The book contains a solution to the Turing test. It contains a list of instructions that adequately replicates something as complex as a human mind. Just one that speaks Chinese. If that's not intelligence, then nothing is.
You can cry a river about "Hard" AI vs "soft" AI or "narrow" AI vs "General purpose" AI. But you lot have always been moving the goalpost and some of you will NEVER be satisfied. "Artificial Intelligence" is a term people use in reference to a certain sort of thing. The media hype might get it wrong a lot, but that doesn't mean the term is meaningless. Just because you're not satisfied with the state of AI, or you want to put humanity on some special pedestal, it doesn't give you the right to try and redefine what the term means.
Sod off coward.
If you're asleep when that happens, no one will wake you.
Save Maine's economy: write stuff down. All comments are exclusively my own, not my employer.
The core problem is defining "AI" The more progress we make toward "AI" the more we demand of it.
Various industrial controllers may have been considered "AI" at the time but now we see them as programmed machines.
There is a spectrum from:
1) Performs programmatic tasks with a human friendly interface (not AI but appears to be) to
2) Performs rough guessing to try to mimic or exceed human responses in limited situations (this is AI applied to specific situations but not the holy grail) to
3) Performs all tasks equal to or better than a human (the holy grail of AI)
Given the infinite use cases for Alexa you need to be pretty darn close to #3 before it can start seeming more like "AI" and less like a mix of #1 & #2.
...donot set an alarm for 7 am in the morning.
Alexa: sure, setting your alarm at 7 am in the morning
State-of-the-art digital assistants are little more than toys for party games. They do a good job of dealing with specific, unambiguous queries. However, as soon as even a little bit of intelligence is required, they spin their wheels pathetically. As it happens, those issues that they can tackle you can probably tackle just as fast, if not faster, at the keyboard. Maybe at some point they will become useful; however, as of today, they are just toys, and the kind of toy that you play with for a few minutes and then you drop for good.
If Alexa is having trouble understanding your "Solar Panels" commands, the voicerec may not be the problem... :-)
Editor, A1-AAA AmeriCaptions
What if all Alexa voice queries were actually farmed out to Mechanical Turk?!
And no "u" in "favorite"
tone
Each time I hear about IFTTT on Slashdot, it is an IFTTT fan whining about the lack of support for the technology.
The holy grail of AI is to make an AI that can make an AI smarter than itself. When that happens, 10,000 cycles later we'll have an AI that's effectively smarter than the sum of all humans.
Learn to love Alaska
I've never moved the goalposts. My definition of "AI" is a computer program that can create an "AI" better than itself. Let every generation run that task for 10,000 generation, and the result will meet everyone's definition of "AI". But no AI has been smart enough to be able to write another AI, let alone one better than itself.
Learn to love Alaska
You're a decade too late to make that complaint. It's all AI, it's just that the trendiest part changes every decade.
1950s: Electronic brains
1960s: Perceptrons
1970s: Neural networks
1980s: Expert systems
1990s: Intelligent agents
2000s: Machine learning
2010s: Deep learning
sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
New technology always starts like things. Things will improve.
It's just a matter of semantics. He thinks AI means "indistinguishable from human intelligence" when it clearly does not (dictionary definition says "mimics human intelligence", which is much more vague).
Unless AI means (more or less) indistinguishable from human intelligence, it is basically meaningless, as you can say that something like ELIZA in many ways mimics human intelligence. As does a room thermostat.
To have a right to do a thing is not at all the same as to be right in doing it