‘Artificial Intelligence’ Was the Fake News of 2016
theregister.co.uk
theregister.co.uk
Still, the points they're trying to make are fair, especially this bit:
>> Out in the real world, people want better service, not worse service; more human and less robotic exchanges with services, not more robotic "post-human" exchanges.
And that's the risk to AI's future in a nutshell: artificial stupidity has the potential to make our lives miserable like natural stupidity never has.
And that's no mistake. The danger in AI is not that it will become smarter than ourselves and take over the world (where Pinky and the Brain failed). The risk is that it will remain dumber than cabbages and still take over the world, as we fall over ourselves to automate everything humans do, even the things that don't get better with more automation, until we're all left without a job, standing in a long line to speak to an AI assistant to get our food stamps for the AI food dispenser.
Just an hour ago I was on the phone with a public service. The answering system's speech recognition worked just fine- it parsed my 11-digit Customer Reference Number despite my Greek accent. It did the job better than a human- those 11-digit numbers are not meant for humans anyway! Still, it kept asking me to press numbers to choose options- why didn't it ask me to speak those aloud? And at the end of the line, when all the options ran out, and only then, it let me know that the offices were closed for the holidays and I should call again tomorrow- Byyye!
Is that natural, or artificial, stupidity? One way or another, it's a picture of the future: imagine a 1000-page form in triplicate, stamping on a human face, forever.
It's probably very hard to evaluate the impact of AI, but their answer is not any better (nor sourced) than that of the professional pundits they criticise (by the way, I can't find any article of anybody claiming that jobs have been lost due to AI). Rubbish theregister article as usual.
The thing about AI is that it doesn't seem like AI in the rear view mirror. Grammar check and voice recognition don't seem like a big deal twenty years later. But it was science fiction in real life twenty years ago.
This. Related link: https://en.wikipedia.org/wiki/AI_effect
To us it's just a bunch of algorithms on Apache Spark but for our end users it's magic, it's AI.
I read a good explanation of different terminologies of AI here on HN not long time back.
Now people will judge it and say that it's not AI because it can't do this or that or because it's too specific to a domain. But why would you buy an AI that's as smart as a human in every way and then use it just to predict sales? An AI replicates one part of human intelligence. The AI is a fully functional artificial human. Two completely independent ideas. We haven't had a single AI capable of fully replacing a human in every way, but is that something we really want or need? Or are domain specific AIs good enough?
Anything that takes the repetitive cognitive load off a human and puts it into a specially trained computer is an AI.
This is just watering out the term AI to the point of uselessness and/or ridicule.
I mean, by your definition, the steam-engine and punched-hole-card powered Jaquard looms of the early 17th century are an AI. Of course, those were the precursor to Babbage's general mechanical computer, so they were actually "specially trained computers".
Edit: 19th century, not 17th. "There are two really hard problems in programming: cache invalidation, naming things, and off-by-one errors."
Think about the word "artificial" and the word "intelligence". If it used to take a human thinking and reasoning about something to perform a task but now a computer is doing it, that is by its very definition, "artificial intelligence". A steam engine or mechanical loom don't replace human brains, they replace human hands. Which is why I said "cognitive load".
I've had the opportunity this year to pay very close attention to the AI debate since my girlfriend is doing her masters in social policy and her focus is on technological unemployment. (and the policy implications of the same)
This is an umbrella term under which AI lives under. Essentially any job which goes away to automation falls under technological unemployment.
Consider the 60,000 foxconn employees http://www.bbc.com/news/technology-36376966 rendered jobless because of TEU. Their further plans to neat fully-automate http://www.theverge.com/2016/12/30/14128870/foxconn-robots-a... their factories.
Those aren't necessarily AI but they are unemployed. The distinction isn't really that important.
Now if we consider the public policy estimates of the number of jobs "at risk" of TEU in the next 5 to 10 years, we are looking at as much as 50% to 60%.
One particularly interesting line in one of her papers was that the estimates were as much as 40% "unless natural language processing reaches pairity with humans in the next 5 years" ... Microsoft achieved that pairity that a few months ago. http://blogs.microsoft.com/next/2016/10/18/historic-achievem...
Kristin Lee at the white house recently published some really interesting papaer on AI, Authomation and the social impact of it https://www.whitehouse.gov/blog/2016/12/20/artificial-intell... my girlfriend sent me the link with the subject "the whitehouse just stole my work!" - though it's really just parallel evolution, they're saying basically what she's coming out with too. (incidentally, if anyone reads this and has Kristin's contact info, I'd love to introduce her to my girlfriend, I think they'd be able to trade ideas)
Anyways - suffice it to say - Artifical Intelligence is only "fake news" if you've got your head in the sand or you decide to creatively misinterpret the literature.
Comparing to speech recognition, NLP currently doesn't enjoy the same level of success by adapting DL technique, AFAIK. By success, I mean parity with human level performance. Current techniques are far from understanding reasonably complex commands/requirement expressed in everyday language and react to it correspondingly.
IMHO, this is as hard as solving the AGI problem itself. Having a chatbot that people can talk to without realizing it is AI or not, is practically indistinguishable from mastering true intelligence, as is defined by Turing Test.
"What we have seen lately, is that while systems can learn things they are not explicitly told, this is mostly in virtue of having more data, not more subtlety about the data. So, what seems to be AI, is really vast knowledge, combined with a sophisticated UX," one veteran told me.
Why be anonymous in such an article? You aren't leaking state secrets. I'd like this 'veteran' to explain to me why recent algorithmic advances like the DNC or recurrent entity networks are not "more subtlety about the data".
The last few years have seen major improvements in the underlying neural network algorithms alone, like the addition of 'neural memory', big advances in training of neural networks that have loops in them and so on. The major progress in fields like question answering, story comprehension and game playing aren't simply about using old techniques with more data. They do reflect significant theoretical advances too.
It's not a good article, just clickbait by Orlowski of the traditional sort. His final point about liability is so vague it could apply to industrial robots or autopilots too, somehow they manage to exist just fine.
I think the issue is that these are small incremental improvements, whereas the current AI media discourse is making out like we are approaching the singularity at breakneck speed. The fact is that right now today there are relatively few domains where AI is really making a big impact in industry, and it just isn't clear how fast it's going to catch up in others (apart from self-driving cars which seem to be getting quite close).
AMT: Ok, so you have been working on neural networks for decades but it has only exploded in its application potential in the last few years, why is that?
Geoffrey Hinton: I think it’s mainly because of the amount of computation and the amount of data now around but it’s also partly because there have been some technical improvements in the algorithms. Particularly in the algorithms for doing unsupervised learning where you’re not told what the right answer is but the main thing is the computation and the amount of data.
The algorithms we had in the old days would have worked perfectly well if computers had been a million times faster and datasets had been a million times bigger but if we’d said that thirty years ago people would have just laughed.
As to the papers you mention- researchers publish papers and in them, they make claims. That's their job. The job of anyone who reads a paper is to take it in good faith but retain their critical and skeptical stance throughout. If you're not directly involved with ANN research, or implementing your own ANN systems for some commercial application, it's very hard to know whether a claim in a paper is useful to anyone besides the scientist who published it.
________________
[1] http://techjaw.com/2015/06/07/geoffrey-hinton-deep-learning-...
a) the algorithms are unimportant because all that we needed was computation
b) the algorithms were ahead of their time
I think that it is actually the latter that is true; in the same way that electricity was a precondition for computation. Babbage designed the difference engine and the analytical engine before they could realistically be implemented, this doesn't mean the design was unimportant.
There will always be some problems that are trivial to have a human solve, but very difficult for anything else. The Butlerian Jihad, and the underlying fear that robots will take every single job, is as much science fiction as hard light or teleportation.
Amazon is testing/using automated warehouse robots to stock their shelves- I imagine it would be a while until we did grocery stores with robots as it is a much less controlled environment and human beings are unpredictable. Plus running over a kid with a robot would be bad for all parties for a very long time.
This is the nature of technological progress. The loom, the cotton gin, the automobile. They were thought exercises, then flights of fancy, then the differentiator between visionary adopters in the market, and then sometime later- the absolute bare minimum for being able to stay in the market.
AI everything/everywhere isn't that realistic- not because of feasibility necessarily, but because of the perceived value or lack their of for any given application. But as a means to replace rote, occasionally hazardous and/or arduous tasks, I feel like that would be a great place to start. No workman's comp when someone falls off a ladder or makes a mistake. No father or mother that dies on the job when an improperly placed load shifts and collapses on someone. 24/7 shifts that don't involve working people to death or mental failure.
We're probably a few decades out at least, but we'll get there. Hopefully gracefully enough to figure out what to do with a swathe of society that's been made redundant.
Cars do have a high level of automation, but much of the interior still has a human operator. The AI puts a seat or door flap in roughly the right position, then the human centers it and triggers the drill motor that turns the screw and secures it in place.
Sure, it's not AI. It's not a deep reinforcement learning neural network trained with Nesterov subgradient momentum.
But robots are robots. Whatever it is, my cashier's out of a job.
I was a grocery store checker in 2007, when these devices were first announced. They are as capable as they were 10 years ago, handle the same edge cases as 10 years ago, and require the same human intervention as 10 years ago. And there are still human checkers at every store - usually they "turn off" the self-check at night, since the human that watches it goes home.
All this AI is just a "mechanical Turk" - using AI to perform most of the tasks a human doesn't want to, and keeping a human around for the edge cases that are too expensive to automate. Humans have such a robustness of variety and ability that we can quickly solve most issues. After a certain point, it just makes sense to keep a few humans around for the really rare edge cases, like the when the grocery store is flooding.
So instead of 2 people to a register (one to bag, one to scan), you can have one person man 12 self checkouts? Sounds like an amazing reduction in labor. Even if self-checkout were half as fast as traditional checkout, that one employee is effectively serving way more customers per hour than they would be doing traditional checkout.
Sure there are edge cases but on an average you still save a lot of time/money.
The job's just as lost, of course.
We've had ATMs for how many decades? And yet we still have staffed bank windows, for the edge cases the ATM can't handle, such as rolled coins. Sure, some rolled-coin ATMs exist, but they're more expensive. It's usually cheaper for a human to handle the rolled coins and other bank edge cases. Same with self-check and alcohol age verification. Same with last-mile driving for self-driving cars and trucks.
But no, that's not happening soon. It seems to be a weird dream of certain left-leaving people for AI to advance as they get minimum wages to rise, so that unskilled human labor becomes no longer cost-competitive with machines. Thus, they create an unemployable underclass that has to be kept alive with welfare systems/basic income and thus are captive votes.
AI has been already well-researched in 90s.
The improvements in classification algorithms and the early commercialization, such as Siri or FB face recognition are responsible for the current bubble.
No major breakthrough in AI has been made so far. Better classifiers is not AI. They are mere pattern recognition.
AI in particular suffers from poor definitions. Nothing computers can do is "AI", only the things that are still impossible can be called "AI". Doing arithmetic was once thought to be a sign of intelligence. So was playing Chess, or Go, or recognizing faces. Now that computers can do this easily it is no longer considered difficult.
So while the recent progress with neural networks is likely somewhat overhyped you shouldn't discount the importance of making things commercially viable. In the 90s we had software that could do image recognition, true, but it wasn't accurate enough, general enough, fast enough to be useful in practice. This has changed over the last ten years to the point where image recognition has tons of commercial applications.
I'm always baffled by that example. The Industrial Revolution was already ongoing when the steam engines become practical.
This, by the way, would be a major source of additional feedback for better training. Fear has been evolved for a purpose.
I am curious about your definition of 'AI' that seems only obvious to the term 'Artificial'.
The key word here is recognition. Better, faster, more efficient classifiers.
The algorithms for training a better representation of the features, not just features itself is a big deal, but it is, again, supervised learning.
The adversary approach, which gives additional feedback for better training, is mere sophisticated supervised learning.
Take one of the best papers survey (which is an emergent new genre in hipster's blogs) and one will see that there is nothing fundamentally new, apart from sophistication and additional feedback.
In fact, there is a discussion about the definition of unsupervised learning, because of the emergence of techniques like word2vec, where it essentially supervised learning but construct all its tasks from the data itself without any labelling.
CNNs, RNNs, LSTMs, reinforcement learning, gradient descent, backpropagation - all that was used in the 90s, but given the limited computing power they had, people simply couldn't run large enough models on large enough data to get results.
I don't think there is a need for 'real AI' in order to replace a lot of humans. Just normal programming in the hands of non-tech office workers can already reduce a lot of human labor.
[1] http://www.theregister.co.uk/2016/11/11/ai_pop_music_maker/
3) When something goes wrong, such as a car crash... who do you put in jail?
2) "The Consumer Doesn’t Want It" - eg clippy, automated call centers
1) "AI is a make believe world populated by mad people, and nobody wants to be part of it" - it's all a cult
Which is all a bit silly - 3) you don't put anyone in jail, you work on the algorithms to reduce our existing 1m+ road deaths a year 2) There's some crap but people like Amazon Echo and I want my self driving car and 1) hardly dignifies a reply.
Meanwhile in 2016, computers mastered Go, the tools became good enough that the likes of Zuck and GeoHot could make somewhat functional Jarvis like assistants or self driving cars in a few months hacking and IDC are predicting the "AI Market is projected to grow from $8 billion today to $47 billion by 2020."
Just wondering.
Intelligence...
"the ability to learn, understand and think in a logical way about things; the ability to do this well"
From this it seems to me that AI is not possible without self awareness which surely must be a prerequisite for understanding.
>3. Liability: So you're Too Smart To Fail?
Seems like there is some missing text?
2. is profit.