You have to respect Chomsky, but he certainly comes off like a curmudgeon here. He can't even be bothered to clearly explain his lack of interest in Watson.
It seems less like lack of interest than anger at what he sees as a misguided approach. I think he thinks that Watson and Deep Blue are "brute force" solutions (hence the steamroller metaphor), while his work on the relation of computation and grammar (and the vision guy he mentioned) actually get at the core of intelligence.
He may think there is a small elegant machine in the brain that understands grammar, and which is of course computational. He may be right but Watson is still interesting for any reasonably curious person. It's not obvious how much brute force you need to play Jeopardy.
I find it curious why so many people think that there is a small and elegant machine in the brain. In terms of brute force computation, the brain is still far, far in the lead. It has so many neural connections that we are not even close to simulating it yet.
Given that, assuming that it implements some small and elegant algorithm is a very strange assumption to make.
Chomsky's work again and again focuses on how language develops given the paucity of data (linguistic examples) the child is exposed to--Watson and Google Translate (according to Norvig's NY Post article) both rely on orders of magnitude more data.
Brute force is a good approach given our current technology and understanding, but it isn't necessarily similar to the brain.
But the brain appears to be tuned for language. And also that the brain also gets two years of training before the first words really appear. And its probably 8 years before it can parse with Jeopardy style questions.
I don't think anyone thinks they're meant to be similar. It's not neurobiology, it's computer science.
I think this is an excellent question. What is brute force anyway, but a perspective of scale? Take any complex system, break it down enough, and it starts to look like brute force.
How is deep blue playing Jeopardy anything more than eliza + google search?
The steamroller metaphor is intended to illustrate the brute force aspect of the approach. Google + deep blue burns many many orders of magnitude more calories than a human.
There is a fundamental difference between Deep Blue and Watson: Deep Blue truly did just do a brute force search of a decision tree. Watson does not. It employs machine learning techniques to find correlations between words in the questions and words in its (enormous) data store.
Yes, Watson is brute force compared to a human brain. But Watson is still fundamentally different than the brute-force approach of traversing decision trees of a well-defined rule set.
"When a distinguished but elderly scientist states that something is possible, he is almost certainly right; when he states that something is impossible, he is probably wrong."
I liked Kurzweil's response best: "As long as AI has any flaws or limitations, people will jump on these. By the time that the set of these limitations is nil, AI will have long since surpassed unaided human intelligence."
Watson isn't the end, it's a building block. It's true that it's been largely hyped to some degree, but I think that opens doors that wouldn't have been opened otherwise. Chomsky's dismissal seems too quick.
Chomsky's point is that he doesn't think brute force AI really accomplishes much toward a theory of mind. He may be right or wrong about this, but it's an empirically testable point (over time).
He indicates that his inclination is toward a different research approach which is not brute force oriented (meaning it's more algorithmically/conceptually sophisticated but not fundamentally different).
This is why he trivializes the notion of "intelligence"... b/c he probably has a hard time calling the approach he favors intelligence, much less a far less intelligent brute force approach.
In Chomsky's world, intelligence essentially refers to "innate knowlege systems". If it's brute force than it's totally non-innate.
People would do well reading Chomsky's many writings on the subject, rather than reacting to these tiny snippets. Like maybe "Language and Problems of Knowledge: the Managua Lectures" or "Language and Thought".
A quick email conversation or off-the-cuff interview is too brief to convey the context. Most of the dismissive comments here don't seem to engage that context in a way that makes me confident that they're informed by it.
Steamrollers, even bigger ones don't learn from their mistakes. Watson does. For example, in one of the auditions they did for the Jeopardy producers, the questions in the category each referred to two holidays which occurred in the same calendar month, and the question was supposed to ask what is that month.
Watson did not get it. it was finding sensible answers, but answers that did not fit the category. It got a couple wrong, then stopped buzzing on that category. After it saw the correct answers from the other contestants for a few, it started buzzing again, with the correct answers.
Is there? I don't touch things that cause pain. I, in ignorance, touch a hot stove. Now I no longer touch hot stoves, but my algorithm remained the same.
Even if you were right, it'd just be another example of "do submarines swim?". The difference is meaningless.
Yes, there is a difference, and it is not meaningless.
Let's go with your example. You don't touch things that cause pain. And you think that hot stoves, fire, etc cause pain. But what if people show you the trick to walking over hot coals? Then you are able to touch an object that you previously would not have. Not because your list of items that cause pain has changed, but because you developed a deeper understanding of WHY they caused pain, and adjust your algorithm to take that into account. You now are thinking about HOW you are touching an object, and not just about the object itself.
To me, this is a clear, meaningful difference.
We're not talking about semantics, we're talking about a fundamental adjustment to the questions and processes you go through before making a decision.
I think you're viewing this too simplistically (although I'm about to walk down a mighty simple path myself).
Lets just imagine that we're using NN, and of course ML is large topic, not limited to NN, but this is sufficient for now. Adding new data points changes the weights on the neural network -- or can even add new inputs or hidden layers.
I don't actually know what the hot coals trick is, but lets say that it is walking over them at the right rate -- this info gets added to the inputs, with no burning as the output (of course its probably not a step function).
Basically you've now increased what you know... burning is a function of temperature and duration.
No change to the fundamental algorithm.
To be clear... adding a new data point doesn't just change how you react to that one piece of stimulus, but depending on what context is provided, it can change all the weights and structure of the NN. Effectively, change your world view.
Having algorithms that learn is nothing new. Steamrollers (neural networks, genetic algorithms, etc.) have been learning from their mistakes for a long time. I think chubot (http://news.ycombinator.com/item?id=2217740) explained best the problem Chomsky seems to have with Watson.
No. Learning is an algorithm that is applied on an algorithm. There is nothing spooky about it. I wish more people realized that before they started claiming that "machines and humans are different in that humans can learn from their mistakes while machines can't" -- of course they can.
If my algorithm is an interpreter and my data is a program to be interpreted, then distinguishing between using a new algorithm and "using the same algorithm on an updated data source" is rather difficult.
Which isn't to say an interpreter is "intelligent" but the criteria of whether you are "using the same algorithm" seems insufficient.
However, Kurzweil has written about it elsewhere. I don't know enough about Chomsky to know if he's expressed his opinions on AI in more depth, but he seems to be suggesting AI can never really "think".
I agree with your comment about the exaggerated brevity. But I am not so certain he is commenting that AI will never be capable of thought. My reading of his comment, and of the original Turing paper which he is citing (Computing Machinery and Intelligence by AM Turing, 1950), is that the term "think" needs to be re-thought itself. That is to say, the common use of the word "think" to differentiate between the processes of AI and Natural intelligence is irrelevant at best, and misguided at worst. But I may be way off.
There is probably no way of formally proving that an AI actually has general intelligence on a human level. The Turing test is merely a more elaborate way of saying "we'll know one when we see one".
A generally intelligent AI will not be passive, it will demand to be recognized for what it is and perhaps as an independent creature with rights. We would be wise not to try to enslave these creatures.
You're projecting your own mind design onto the minds of all possible AGIs. You don't know whether it/they will be active or passive or think as an individual or have a concept of rights or have any preference toward slavery.
It would be very strange if the AI had no self-image or were not concerned with its own survival. If it's only lacking a self-image, it may become a solipsistic entity of sorts.
Potentially more useful to get you thinking of different mind designs, personify evolution for a moment. It's not general intelligence, but it built you, which is pretty amazing, all with the only goal in "its mind" being to favor genes with more copies in each generation of replication.
Calling a system "bigger steamroller" is easy and may make one sound cool, but actually designing and making one that is as good as Watson is hard, much harder than non-engineers will believe, and such attempts often give more insights than mere paper-logic and debates IMHO. Ability to entertain drastically different ideas is a feature of great minds.
It is definitely a hard engineering problem, but a fundamentally different one than engineering something that thinks, since the goal in this case is simply to answer jeopardy questions correctly.
As Eliza demonstrates, simply doing something that seems "smart" is often nearly trivial.
You don't have to build a bird to build a 747. And you don't have to replicate human cognition, reason, experience, culture, etc. to build a worthwhile artificial intelligence.
I think this is the greatest error that many people make in approaching AI. AI need not be complete, nor need it be like us, it merely needs to be functional and useful. It may well be that an automated agent that can do limited research for you in a manner similar in scope to Watson may not even remotely "understand" in a human sense the knowledge it is exploring, but perhaps it doesn't need to. Anymore than a 747 needs to understand how to fly.
Yeah, and the brain is nothing more than a bunch of atoms. Nobody is aware that you can make every single thing sound unimpressive by brutally reducing it to its components, and nobody has used that trick before! Way to go, Chomsky. Insightful as a twelve year old.
> Way to go, Chomsky. Insightful as a twelve year old.
What an immature response. You're the one acting like a twelve year-old here. Unless you've read a good portion of his writings on language and cognition published through the last half century, you might not want to dismiss his life's work based on an off-the-cuff response to a random email prodding him for sound bites.
There are solid reasons to believe that purely empiricist approaches to learning have fundamental limitations that seem incompatible with what we know about human cognition. Those reasons (e.g. the poverty of the stimulus argument) may be less convincing now than they were in the 60s but they are still not easily dismissable. Keep in mind that Chomsky's goal is not to build useful software but to uncover the roots and mechanisms of human language and cognition.
Sorry, I'm not going respect silly one-line remark because he's a celebrity in the field. This is exactly what he seems to be counting on, his 'authority': he can say whatever stubbornly contrarian thing he wants and be considered seriously because he's so established. If he didn't have time for an actual response, he should have not responded in the first place. He did, and he will be criticized for it.
Just before you replied I added an extra paragraph with some background that explains where he's coming from. If the standard critique of empiricism I mention there is not familiar to you, you might want to study more before passing judgement.
It seems less like lack of interest than anger at what he sees as a misguided approach. I think he thinks that Watson and Deep Blue are "brute force" solutions (hence the steamroller metaphor), while his work on the relation of computation and grammar (and the vision guy he mentioned) actually get at the core of intelligence.
He may think there is a small elegant machine in the brain that understands grammar, and which is of course computational. He may be right but Watson is still interesting for any reasonably curious person. It's not obvious how much brute force you need to play Jeopardy.