It is true that we are probably quite far from AGI, but nobody really knows how far. We could be closer than most experts think. Few experts in 2006 would expect deep neural networks to beat a Go champion or be able to describe visual scenes in natural language sentences within a decade. Unexpected things happen, and we are much further behind in terms of preventive technology like Friendly AI. (https://en.wikipedia.org/wiki/Friendly_artificial_intelligen...)
Given the existential risks involved, I would propose more careful handling of the technology. (I agree that opening up most other technologies is usually a positive thing.)
See "Superintelligence" for comprehensive arguments against unfettered development of AGI: https://en.wikipedia.org/wiki/Superintelligence:_Paths,_Dang...
There's an episode of Pinky and the Brain where Brain invents a magnification ray and gigantises Pinky and himself - so they're now giant rats [1]. They walk about some human metropolis, looming larger than skyscrapers and terrorising everyone. To cut the story short, Pinky happens and eventually the gun shoots everything in the world.
... so Pinky and the Brain are now normal-sized mice again. Or, well, they're giant mice, but everything in the world is also giant, so they're small.
I think of that episode during discussions of the dangers of AGI and how "democratizing" it will eliminate those dangers. The problem is that AGI, like nuclear weapons, is not an equalizing force only- it is also a very destructive force, and so there is the danger that any loon with access to it could blow us all to kingdom come (with nuclear weapons, for sure; with AGI, perhaps).
On the other hand, when nuclear weapons, or AGI, giant rays etc are not democratized we have the situation we're in today: only a few entities have access to them and they basically have free (military) reign... and there is still the danger that some loon might get to a position of power were they can push a button and -blamo.
Which is why people are really worried about this sort of thing. Because once certain discoveries are made, there's no going back and at the same time the road ahead is full of danger.
A super intelligent AI does not change the laws of physics or even get to predict the weather two weeks from now.
Sure, they could learn to be a good hacker but the worlds best hacker can't get into air gaped systems. More importantly, flawless security trumps hacking. If there is nothing to exploit there is no way in, which means a lone AI is much more powerful than lot's of AI's with different goals.
We don't know how many other people aren't working on this because of moral/ethical reasons. Of course, 99.9% of the world could be wary of genetic engineering, but that remaining 0.01% is enough to pursue research, get VC investment and drag the rest of us into that uncertain future.
To continue the nuclear analogy, the Manhattan project was probably the most impressive engineering program in the history of the world, but it was driven by survival in a World War. They didn't build and drop the atomic bombs for fun. You'd think that working on a limitless virtual brain should have similarly serious motivations, not just the examples of "how do I drive to X but also shop for Y?" or "is that a monkey?".
I know there are much grander societal goals with A.I. and the world really could become a "better place", but please sell society those goals, not the usual first-world problems. We already have enough people trying to destroy the world without these extra tools.
All the efforts of ML solve problems with a concrete input space and a concrete output space. This means they solve very discrete problems and are not generalizable, unless they have similar looking model inputs and the model has the same limited range of "moves" it can make.
I see no reason that's necessarily true. I'm a general intelligence, and I can't make myself smarter.
I don't know what the future holds, but the fact that undecidable problems always seem to involve Turing Machines either designing or inspecting other Turing machines makes me suspect that singularity won't be the explosion some expect.
We've got all kind of past fiction which plays on these concerns, more recently Ex Machina and Westworld.
But here people are making the same bold pronouncements...so certain we know or can control technology which hasn't even been invented yet.
I mean what superintelligence supposed to do? Solve the halting problem or do other plain impossible things? Chess computers can beat a human champion with sheer firepower, but they still can't do checkmate in one move.
Lets not take is as a certainty. This problem might be beyond our capabilities. Its kinda like stating: When we invent warp-drive. We don't have a proof for warp-drives same as we don't have a proof for building AGI.
But such thing cannot be tamed. As the AGI is democratized, it's bound to happen that someone somewhere will make it evolve goals.
The future of the human kind will be decided in a great measure by what kind of goals such an AGI will set first.
I disagree. Chatbots and physical robots are where AGI is going. It's basically AI that is interacting with people and the world directly, not just computing away in a cluster, broken from reality.
...
I'm sure it's written by a smart and talented journalist - but it's just too long. I cannot possibly allocate so much of my time to read a single news article!
I know it's kind of my problem, but I'm sure lots of others are just like me - it's just the nature of our tech-provoked ADD.
I skimmed through it all right, but I didn't get too much out of that. If I need to go deep, I grab a book!
Good journalism doesn't get through, because it asks too much attention of its (busy) readers.
A TLDR version of good articles is a must. That's a good problem for AI isn't it ? For the moment, though - I'm sure the authors themselves could make a nice summary readable in maybe 3-5 minutes.
Frankly I love it. It reminds me of the gogo days of WiReD Magazine, when so much was new in the Net it was impossible to do justice to a topic in fewer than 6000 words.
And now due to Deep Learning, AI/ML is gogo too.
What's worse is that this approach seems to be a dead end, in the sense that it is only useful for pattern recognition, which can substitute for decision making only for extremely simple processes (absent a quantum leap in a range of technologies), and even then those kinds of applications are notoriously difficult to develop and maintain.
I look forward to the enormous benefits we will get from machine learning, as we are already seeing, but again, overselling it won't do us any good.
It's just not credible to claim that you know what can be done with all possible combinations of current ideas, much less those that will be had tomorrow.
Yes, it may put you out of a job one day but it won't actively seek to destroy the human race.
This would be a massive issue in the US with the current setup.
Close to the western summit there is the dried and frozen carcass of a leopard. No one has explained what the leopard was seeking at that altitude.
Near the top of the west there is a dry and frozen dead body of leopard. No one has ever explained what leopard wanted at that altitude.
One of those is pretty stilted and not much better than lot of machine translations. The uncanny valley of being good enough but not optimal is probably something that's going to plague AI for a long time.
Google’s decision to reorganize itself around A.I. was the first major manifestation of what has become an industrywide machine-learning delirium. Over the past four years, six companies in particular — Google, Facebook, Apple, Amazon, Microsoft and the Chinese firm Baidu — have touched off an arms race for A.I. talent, particularly within universities. Corporate promises of resources and freedom have thinned out top academic departments. It has become widely known in Silicon Valley that Mark Zuckerberg, chief executive of Facebook, personally oversees, with phone calls and video-chat blandishments, his company’s overtures to the most desirable graduate students. Starting salaries of seven figures are not unheard-of. Attendance at the field’s most important academic conference has nearly quadrupled. What is at stake is not just one more piecemeal innovation but control over what very well could represent an entirely new computational platform: pervasive, ambient artificial intelligence.
That's...worrying. Given all of these companies lack of respect for privacy and consumers, it should trouble us that one of them may end up with such a world-changing innovation. Throw enough money into one place, stick a bunch of PhDs in a building, and eventually you'll get something. It's just numbers. Bodies + money. What's inspiring about that?
Does the prospect of Mark Zuckerberg having control over AI for the next five decades trouble you? Even more remarkable is he'd do it on the back of having made a marginally better social networking site in PHP in 2004 and spreading it via the best social network in the world - Ivy League universities. And now those same universities are being raided for talent by these companies...
Is this how we should be picking winners? The distinct lack of diversity and their past stances and actions are troubling. This seems to be mostly a hype piece without any regard for practical effects.
If these same companies can't anticipate or mitigate the impact of issues like fake news until after an election, what makes you think they understand the consequences and impact of something much more complex? And even if they do anticipate it, how do they hold back the pressures of shareholders?
Seven figures? For grad students?
It's still impressive, mind, and a big improvement over the previous translation, but not perfect like the article wants to imply.
https://www.kaggle.com/c/word2vec-nlp-tutorial/forums/t/1234...
"That interim also saw dedicated attempts on the part of Google’s competitors to catch up. (As Le told me about his close collaboration with Tomas Mikolov, he kept repeating Mikolov’s name over and over, in an incantatory way that sounded poignant."
"Just as the chip-design process was nearly complete, Le and two colleagues finally demonstrated that neural networks might be configured to handle the structure of language. He drew upon an idea, called “word embeddings,” that had been around for more than 10 years. When you summarize images, you can divine a picture of what each stage of the summary looks like — an edge, a circle, etc. When you summarize language in a similar way, you essentially produce multidimensional maps of the distances, based on common usage, between one word and every single other word in the language. The machine is not “analyzing” the data the way that we might, with linguistic rules that identify some of them as nouns and others as verbs. Instead, it is shifting and twisting and warping the words around in the map. In two dimensions, you cannot make this map useful. You want, for example, “cat” to be in the rough vicinity of “dog,” but you also want “cat” to be near “tail” and near “supercilious” and near “meme,” because you want to try to capture all of the different relationships — both strong and weak — that the word “cat” has to other words. It can be related to all these other words simultaneously only if it is related to each of them in a different dimension. You can’t easily make a 160,000-dimensional map, but it turns out you can represent a language pretty well in a mere thousand or so dimensions — in other words, a universe in which each word is designated by a list of a thousand numbers. Le gave me a good-natured hard time for my continual requests for a mental picture of these maps. “Gideon,” he would say, with the blunt regular demurral of Bartleby, “I do not generally like trying to visualize thousand-dimensional vectors in three-dimensional space.”
That makes sense, considering that Basic English has only a thousand words yet can express most concepts with enough words.
The problem with word embeddings, or any distance-based model really, is that language doesn't work that way.
Chomsky has a standard example he uses to make this point: "Instinctively, Eagles that fly swim". He points out that in this phrase, the "instinctively" goes with "to swim" (as in "instinctively, they swim") even though the phrase, and the attachement, mean nothing (the phrase is nonsensical by design).
If the relation was really based on distance, we would expect "instinctively" to attach to "fly". The fact that it doesn't suggests that there is something else that makes us pick the correct association out of all the possible interpretations in that sentence.
Word vectors in their original form also have trouble with homonyms etc "faux amies": for instance, the word "cat"- is it referring to the animal, or to the Linux command? In vector space, there wouldn't be any difference, so the animal would be associated with the symbol ">" and the Linux command with "small" and "furry".
If you believe that Moore's law is only on life support and not totally dead, there will be more processing power to harness in the future. The number of researchers and investment are clearly growing very quickly. The models used can endlessly be improved.
But on the other hand there are so many things that even today just aren't captured as digital data. I work as a mechanical engineer and there are many nuances to mechanical design that appear nowhere in print (or youtube video, or blog post for that matter). Learning these things takes a complex combination of of sight touch and intuitive leap. Even unsupervised learning requires some input to feed the net. I just don't see where it will come from.
Anyone think I'm way out of line here?
What's missing in this article as well as most reporting on AI is the differentiation between artificial intelligence and artificial consciousness, otherwise known as individuality or self awareness.
To me there's a whole smoke and mirrors phenomenon going on in the AI topic, especially the idea of "emerging AI", and the supposed potential danger that poses, and it's tied to the tendency we have as humans to anthropomorphize inanimate objects, and to believe in supernatural effects.
That tendency allows the idea of artificial self awareness to always float behind the scenes in these conversations, and let's normal reporting on AI be magically conflated with a different topic.
It's important to realize that AI is nowhere near self awareness, or conscious, or "awake" and won't be no matter how far the field and implementation goes. No matter how many Turing Test they pass, intelligent machines will be no more conscious or self-aware than the Mechanical Turk!
That's because solving the problem of self-awareness, or consciousness is a different engineering challenge than solving problems of AI. Consciousness is a more complicated, and specialized a thing.
Were we to build an artificial self-awarene machine we would not expect it to pass a Turing Test. Instead we might expect different things of it and ask different questions to determine if it is self aware: can it adapt and survive without human help, ie can it trap and store energy and reproduce itself, and what purpose does it find for itself is what objective does it pursue ...
These are things machines are capable of as well, but as I said: it's a different engineering challenge than producing information that is organized to be sensible to human mind, which is the AI challenge, and the Turing Test.
That isn't to say machine learning isn't potentially dangerous, on the scale of atomic weapons or greater, especially in conjunction with automation, however the idea of an artificially emergent consciousness wth intelligence greater than our own is hogwash: we would do better to pay attention to our own emergent lack-of-intelligence systems and worry about them taking over first.
> No matter how many Turing Test they pass, intelligent machines will be no more conscious or self-aware than the Mechanical Turk!
I see. Well, this is just a rephrasing of the "Chinese Room" discussed in the article. Taken to it's logical conclusion, I am certainly self aware, but the rest of you are all just acting out complex behaviors encoded in chemical and electrical gradients, successfully mimicking consciousness.
I think that if any entity exhibits the behaviors associated with conscious thought, it would well behoove us to treat such entities as conscious, or we may very well find ourselves holding the short end of that particular stick sooner than we'd like.
> That's because solving the problem of self-awareness, or consciousness is a different engineering challenge than solving problems of AI. Consciousness is a more complicated, and specialized a thing.
Since there is no doubt that ML/AI has a long way to go toward AGI, and along the way we can expect the discipline to evolve considerably in many unexpected directions, this assertion of yours is close to tautological.
> Were we to build an artificial self-awarene machine we would not expect it to pass a Turing Test.
Why not?
> Instead we might expect different things of it and ask different questions to determine if it is self aware: can it adapt and survive without human help,
So, anyone severely ill to the point that they cannot do without assistance is not conscious and self aware?
> ie can it trap and store energy and reproduce itself,
So, a single-celled organism is conscious?
> and what purpose does it find for itself is what objective does it pursue ...
Ah, this seems a relevant criteria, but keep in mind that humans can be subjected to operant conditioning ("brainwashing") to impose external goals, not to mention that humans actually require a couple of decades of such conditioning (albeit rather more gradual and haphazard) before being considered competent members of society, but we don't consider humans to be less conscious or less self-aware on either side of that particular divide.
> it's a different engineering challenge than producing information that is organized to be sensible to human mind, which is the AI challenge, and the Turing Test.
Given that people have to be specially educated to produce information that is organized to be sensible to a computer, I don't see why an AGI, whatever it's capabilities "out of the box", so to speak, shouldn't be expected to be capable of learning to be sensible to humans.
Yes of course a single celled organisim is conscious.
Exactly the way an amoeba is self aware is how an self-conscious intelligence system would need to be to pose any kind of threat: organized to find energy sources and metabolize, replicate, etc.
I'll tell you: a single celled organism is way more self aware, and way more functionally complex than any computer or software - in fact it's orders of magnitude more complex of a machine.
That's my point: solving problems that make a machine capable of producing intelligence that is sensible to you and I is not solving the and problems that make a machine like a single cell organism, which is to say vertically integrated from the atom upwards to be a self-sustaining, self propagating, energy trap.
A self-aware human who is disabled and can't live without intervention of other humans, can't self-sustain without others and therefore will not pass the test of being able to self-sustain. It's a test, and so one failure isn't validation of hypothesis. It can still be a great test affairs fail a percentage of the time.
In general we know that all self conscious organisms self-sustain, even social, super-organism ones that need each other to survive so a criteria for a self aware organism is that it be capable of self sustaining. We don't even have a good test for that yet. But a test that would fail a perfectly self-aware disabled human wouldn't be a good one.
We could very well administer a Turing Test to an artificial consciousness, but my point is that it wouldn't be a very accurate test. A Turing Test only proves the accuracy of a facsimile of human intelligence. It proves nothing about self consciousness systems. An amoeba would fail it in an instant as would a parrot or dolphin - and if you tell me these organisms aren't self-aware and conscious then we are definitely not on the same page.
I could be wrong. I'm absolutely interested in anyone who can make a convincing argument otherwise, however until then I'm pretty certain that no emerging conscious machine will happen by accident. Rather it would take a Manhattan project or greater to produce an artificial consciousness on par for sophistication with an amoeba. And we don't have much motive to attempt it either, so I'm doubting we will do it anytime soon.
"Hey, I'm interested in AI maybe this is worth some investigation"
-Sees that it's from nytimes-
"Uh oh"
Let's use it to translate what I just wrote, into Spanish:
"Ese viaje de ida y vuelta el artículo comienza con es mucho mejor que cualquier cosa que he obtenido de Google Translate."
That's readable but the "el articulo comienza con" is bush league — a clear sign Google is translating word for word. No one would ever mistake this translation for real Spanish.
If we translate back to English, the result is better than the Spanish!
"That trip back and forth the article starts with is much better than anything I've gotten from Google Translate."
So, amusingly, a weakness on one-way translation — the word-for-word method — becomes a strength on round trip translations. (Not that round trip translations are going to be useful to anyone.)
I did Spanish so that a lot of people here would understand. But now let's do Hebrew. I get
תרגום הלוך ושוב כי המאמר מתחיל עם הדרך דרך טוב יותר מכל דבר שאי פעם שיתקבל מ- Google Translate.
That's beyond merely bad, it's pretty unintelligible as Hebrew. (In fact the only way a Hebrew speaker would make any sense of it is if he knew English and tried translating word for word back to English.)
And indeed, once again the round trip translation is better, though the original meaning is pretty much lost:
"Translating back and forth that article begins with a way better way than anything I've ever received from Google Translate."
Google brain requires a patch.
https://www.theguardian.com/world/2002/jun/17/humanities.int...
I not quite comfortable on how the Borges quote was translated. To show the difference, the new TRANSLATE version, translates again to spanish as:
Tu no eres lo que escribes, eres lo que has leido. A bit different from Borges original phrase meaning.
A sentence with a closer meaning (but a bad translation) would be:
One's way of being is not (caused) for what you have write, but for what you have read.
>Grinning, Pichai read aloud an awkward English version of the sentence that had been rendered by the old Translate system: “One is not what is for what he writes, but for what he has read.”
>To the right of that was a new A.I.-rendered version: “You are not what you write, but what you have read.”
edit: formating
