Hacker News Reader

Totally agree, and very happy to see this statement at the top of HN. I'm against hyping up AI or misinterpreting it just as much as the next guy (especially coming from a position where dealing with a lot of higher-management people has made me very wary of articles and vision-pieces from the other end of the spectrum that are all like "invest now in AI or lose it all"), but this is just clickbait for the sake of enjoying playing devil's advocate.

AlphaGo solved one of the hardest challenges which we believed to be very far out in the future before it came along. I'd say the field is on the right -- and very exciting -- track; people just need to learn to approach it in a clear-headed manner.

Good on IEEE though for welcoming opposing viewpoints.

I wouldn't say the challenge has been "solved" - the system may have only been able to win consistently because the programmers made adjustments and improvements to it between each game. Fitting its heuristics to a specific opponent might actually be overfitting unless it can consistently win against all challengers without being modified between rounds. [edited to clarify uncertainty about "why" it was able to win]

AlphaGo played plenty of games online after the Lee Seedol match. It went 60-0 against some of the worlds best players. Granted they were fast games but that's a challenge for AlphaGo as well as its adversaries.

Thank you for this information, it will definitely be interesting to see how the future of Go plays out. For example, will players be able to train with AlphaGo to the point of being able to beat it?

Even more interesting to me is the question of if Go players will be able to apply the novel strategies of AlphaGo which break with decades of conventional wisdom or if there's something in their execution that requires such perfect play (but leads to a more certain outcome) that they're only feasibly played by AlphaGo.

I'm not sure about human players, but it might be possible to build an adversarial neural network that beats AlphaGo (and only AlphaGo).

I thought AlphaGo is trained in an adversarial manner already (it plays against itself).

The version that played against Fan Hui was reportedly trained that way, but if I recall correctly the more recent one that played Lee Sedol did't rely on self-play as much.

Also the objective was different - they tried to train a network that would play well against any opponent. It should be easier to train a network that exploits some very narrow weakness in specific AlphaGo version.

For example, my phone consistently beats me at chess. But I found a sequence of moves in one opening where is always makes the same mistake, and I after that I can win easily.

That's the sort of thing I have wondered about. Chess continues to evolve since the famous Kasparov vs. Deep Blue matches, and from a quick look around human players do continue to draw matches against the strongest computers (even though they are regularly beaten, as well).

Humans will never defeat the strongest go playing programs ever again. AlphaGo is now clearly ahead of the top pros, and such programs will only get stronger. Humans should learn some things from the machines, but we will definitely not be able to extract the full totality of understanding, and thus will naturally continue to fall behind.

The one game that Lee managed to defeat AlphaGo was probably the last time any human player was able to beat the strongest Go program at the time. That said, there are a few interesting ideas to try, and I (being a reasonably strong amateur go enthusiast myself) would love to see Google try out:

1. how many handicaps can AlphaGo give to human players, while still come out winning? the idea of handicap is basically the stronger player lets the weaker player a few "freebie moves" before the game proper starts. For reference, a typical professional player can give 2 to 3 handicaps to the strongest amateurs.

2. Can AlphaGo defeat a group of professional who are also given unlimited time to think and discuss?

Making a group of professionals with unlimited time to arrive to a common conclusion may be a harder challenge than AI. I would love to follow those discussions, but I am not sure they would ever end.

I believe the AlphaGo team estimated it to be about 2 stones stronger than top pros.

I believe that there was a sort of "learning freeze" for AlphaGo during the Lee Sedol games and there was no manual modification of its strategy either.

If I recall correctly this was part of the commentary during the matches so I can't find a good source for it right now.

https://youtu.be/l-GsfyVCBu0?t=2866

Thank you, that's exactly what I was looking for!

Thanks for bringing this up. I was recalling more recent articles that did state there were modifications each night between matches. Perhaps those were in error; I would not have commented in that case. It's a shame nobody pointed this out sooner, I could have removed the misinformation.

It's not clickbait. You're being pedantic. It's clear the author meant true / general / strong AI and was merely addressing the hype of some non-AI researchers.

It's a response to people who think because of AlphaGo that we are on the cusp of achieving true / strong / general AI.

> An AI that solves a lot of problems better than humans, and transfers that learning across many problems with relative ease, is getting close to strong AI.

We're not close. See Yann LeCun's statement on AI after AlphaGo [1] and the HN response [2]

[1] https://www.facebook.com/yann.lecun/posts/10153426023477143

[2] https://news.ycombinator.com/item?id=11280744

I have always found the idea that there's a hard line between strong and not-strong AI extremely naive. The whole thing is a long gradient of progress. Strong AI is going to creep up on most programmers like the proverbial boiling frog, and internet forums will be full of people denying its existence throughout that process.

> I have always found the idea that there's a hard line between strong and not-strong AI extremely naive

You probably haven't studied much or any AI. In my opinion, those who haven't researched a subject don't understand the details and are likely to make more inaccurate predictions than experts.

What a randomly self-serving assumption to make.

I can't agree. If we can blur the difference between weak and strong AI, that 'transitional form' (missing link?) will be enormously important to the future of AI (and mankind).

In the past 50 years, AI has seen hundreds of small successes in narrow tasks that used to require humans. But none so far has shown the potential to scale up, generalize, and serve tasks other than the narrow one for which it was designed. Like IBM Watson, AlphaGo too is likely to be consigned to the AI scrapheap in the sky.

BUT... the deep net technique used by AlphaGo shows more promise to solve the remaining unsolved AI tasks than any AI method before it. Yes, we still don't know DL's limits, like whether it can integrate one-shot learning, or build and reuse a diverse knowledgebase, or transfer specific methods to solve new more general problems. But as of right now, it's shown greater promise to solve novel weak AI tasks than any past technique I've seen. The author overlooks that potential deliberately and provocatively, and IMO, pointlessly.

Can DL scale up into strong AI too? I think the important thing here isn't that the answer isn't obviously yes (as the author posits), but that the answer isn't obviously no. And in the 50+ year quest for strong AI, that's a first, at least for me.

It's not self-serving to say PhDs are better than me at predicting the future of their field.

It would be self-serving if I lied in an interview and said I was qualified for a job making robots if I'd never had any experience doing so.

Then why not title the article "Alphago is not strong AI"?

Read the article. The title isn't a substitute for reading. It's short because your attention span is short.

You didn't answer my question. There's an advantage for using the long title "Alphago is not strong AI": less people will be angry at a perceived clickbait title. What's the advantage for using the short title "Alphago is not AI"?

You're asking me why the author chose this title? I am not the author.

I could guess that he didn't use a qualifier because there are many different ones and none universally accepted. Or, maybe he wanted to reach a non-tech audience who wouldn't know the difference between strong AI and AI.

There are plenty of good reasons, and reading the article clarifies the author's meaning.

It is easy to see if a title is clickbait or not. Just change it with some stupidly long title, such as "AlphaGo does not solve the problem of meaning, so it is not AGI" and ask yourself if this title would have generated a similar number of clicks. The title is as clickbait as it can get.

No, clickbait is "10 ways to find your spouse"

This is just a short title, and ITT people didn't read the first paragraph.

If you read the first paragraph, it mentions first AI then AGI, and does not deny that the latter is a subset of the former. In the second paragraph it even mentions that today’s AI is not AGI.

So it uses a strong claim — i.e. AlphaGo is not AI — then subsequently changes it to a smaller claim — AlphaGo is not AGI.

Reading the entire article does not change the fact that this can be seen as a technique to attract readers. Claim rare X; change X to be a subset of Y; then claim it was really new X all along. Moving the goal post.

Poppycock. Anyone who understands software knows AlphaGo used AI tech and that the title is referring to general AI. Anyone who doesn't will have their understanding clarified in the first paragraph.

There's a clear and meaningful difference between "not AI" and "not strong AI".

With respect, you are being pedantic.

look, if i walk up to somebody and say alphago is not AI, they'll understand what I'm saying, because the layperson doesn't distinguish AI and strong AI. Only techies do.

I think that's what the author was thinking when he wrote the title, and I'm giving him the benefit of the doubt.

I'm being flexible, not picky. Disagree with me? Fine, I really don't care.

> It's clear the author meant true / general / strong AI

Perhaps we need a new term to refer to "true" AI, preferably one that matches people's general notions of what an AI should be. "Synthetic Thought" maybe?

People just need to read the article. The author describes what he meant in the first paragraph.

If that were true, the article is useless – because nobody ever said that AlphaGo was 'strong' AI.

RE:the embodiment/robotics argument, notice that he has a vested interest in this field:

> Several research labs are now trying to go further into acquiring grammar, gestures, and more complex cultural conventions using this approach, in particular the AI Lab that I founded at Aldebaran, the French robotics company—now part of the SoftBank Group.

> ...requires embodiment, that they have a physical body in space through which they connect to the world. I don't believe either of these are necessary requirements for AI.

This is very interesting. In the human realm, there is the celebrated case history example of Helen Keller, who progressively developed her blindness and deafness over a period of years. She would arguably "pass" whatever tests that were administered for evaluating a strong AI.

However, walking backwards from that, towards case histories where patients were born deaf and blind, they also could pass those kinds of tests. But add more sensory disabilities, and it becomes murkier. Has anyone performed a detailed survey of medical histories? Is there a point beyond which lack of a specific set of senses leads to sapience incapacitation 100% of documented instances in humans? Could that be a promising lead to build a virtual environment that simulates a minimum threshold of senses?

Also, AIs are not constrained by senses shared with humans. Our information passing is mediated by few inputs — a stream of light as a matrix within a very narrow range of possible spectra; a stream of air pressure decoded into frequency series; touch etc. AIs could have many more modes of input and output.

Also… it seems weird to consider having “human” inputs embodied but having direct access to information stored elsewhere as not — surely there’s disk platters, SSDs and other devices existing in the real world, no?

Absolutely correct, more senses are definitely available to an AGI. In the context of building an AGI that possesses "like a human" implied or directly in the evaluation criteria, and the thesis that embodiment of senses is a prerequisite to fulfilling that criteria, we don't know whether or not more senses than human might thwart the development of the first AGI. We currently believe that more senses than "normal" tends to lead to neuro-atypical development like autism, so there is some evidence for caution against tossing in all the inputs we can imagine, based upon current thinking (I wonder if that thinking is based upon empirical fMRI scans). We do have the corpus of data built up by humans however, that says "no known cases in medical history of sapience failing to develop with this minimal set of senses".

My cursory search says hearing and sight are not necessary. Leaving us with smell, taste, and touch. Smell and taste are closely interlinked, so there is a possibility that a simulated smell-taste and touch are all that is required to "bootstrap" the first AGI, so to speak.

Your point recalls to mind the scene in Battlestar Galactica (the reboot) where Cavil fulminates "I don't want to be human". [1] I suspect once we boostrap the first AGI, you are absolutely correct, we will rapidly add more senses to the AGI's repertoire. But we only have one data point for a template to build AGI off of, ourselves, so I also suspect the first AGI will somewhat resemble us. If so, and if the embodiment principle is one of the right routes to take, then it makes sense to me to simulate what we empirically determine as the minimum sense set, and then expand from there. Walk before we run, so to speak.

It wouldn't hurt for someone investigating the embodiment principle to build for as many senses as they can envision, though. Many eyes, and all that, for attacking the problem space. I'm just expressing an uninformed gut feeling here, as I am not in the AI research space.

[1] https://www.youtube.com/watch?v=pM3CptVZDYU

Is AlphaGo really what we want to classify as 'intelligence'? It is a calculating engine – one which outperforms humans, to be sure, but no more intelligent than a properly-wielded pocket calculator.

Yes – stuff like this is confusing 'intelligence' with 'sentience'.

Playing go or chess would have been classified as activities clearly requiring intelligence in the recent past. We're just moving the goalpost whenever AI advances. At some point, it'll probably divide mankind into fractions which either want or don't want to grant certain rights to AI (possibly obsoleted by the great AI uprising of 2025-20-11 6:45:23am to 2025-20-11 6:45:25am)

I don't think it is quite that simple.

The 'real world' is not nearly as nicely ordered and specified as a turns based game is. So a computer will be ideally positioned to take advantage of its ability to:

- remember perfectly, forever

- execute algorithms at extreme speed

- in parallel

So you end up with a race that is comparable to a sparrow against a jet airplane. Technology can and does beat biology regularly in all sorts of domains.

Slowly but surely computers are making their way into the real world: their programmers - and by extension the systems - learn how to adapt to the messiness of interacting with life as we know it.

Just like you'd first get kids to play in a playground (the games) and then let them loose in the real world (self driving cars and other much harder tasks).

In the end, all of these amount to a slow but steady march towards human equivalence on a large range of tasks, and excelling humans on plenty of them.

The hard AI question is one where we ask ourselves whether or not there will one of these days be a system that is better than humans on essentially all tasks without exception, including learning, creativity and so on. Once you reach that level all bets are off, until then the slow-but-steady march continues.

The difference is evolution vs revolution, but we've already seen enough progress in limited AI domains that we could conclude we're already seeing a minimal revolution. Whether it will result in a major one is still an open question, it could easily be that progress will at some point plateau.

It uses intuition to play, which is how humans play, but it does it by considering way more positions that humans.

>It uses intuition to play,

Not by the normal definition of intuition. It reasons about millions of moves.

Could you provide a citation for that figure? It's not one I've read but I don't necessarily disbelieve it. It is worth considering this perspective from one of the Deepmind Nature papers:

"During the match against Fan Hui, AlphaGo evaluated thousands of times fewer positions than Deep Blue did in its chess match against Kasparov; compensating by selecting those positions more intelligently, using the policy network, and evaluating them more precisely, using the value network—an approach that is perhaps closer to how humans play. Furthermore, while Deep Blue relied on a handcrafted evaluation function, the neural networks of AlphaGo are trained directly from gameplay purely through general-purpose supervised and reinforcement learning methods"

Those are two independent sentences which are not mutually exclusive.

Your intuition is probably based on millions of events happened to you and your ancestor (parts that are developed through evolution and social construct).

Equivalent on the side would be million of games in the past either simulation or real human ones, both of which such systems use

The policy network (intuition) is already high dan level, which is stronger than many strong amateurs. That's right, with zero reading it's already something like 5d level.

The other part that increases its strength is the positional evaluation through the value network. It's ALSO intuition based on which points are going to belong to which player.

Those two combined give it a high dan level, maybe around 8d.

Only the last component (Monte-Carlo tree search) is not quite as similar because it uses random game roll-outs to ALSO give an evaluation.

AFAIK, value network evaluation is an overestimate of territories, while the MCTS is an underestimate so using both it's stronger than professional level.

The levels I put is the version that played Lee Sedol. The version that went 50-0 on the Internet is clearly stronger and might already be professional level without MCTS.

Not only does it use a form of intuition, but it's an intuition that it's developed itself, rather than just copying others. Professional Go players are learning from the creative strategies of AlphaGo and if that doesn't imply some level of weak intelligence then I'm not sure why anyone talks about a weak/strong AI divide to begin with.

Like Scott Aaronson said, "You can look at any of these examples -- Deep Blue, the Robbins conjecture, Google -- and say, that's not really AI. That's just massive search, helped along by clever programming. Now, this kind of talk drives AI researchers up a wall. They say: if you told someone in the sixties that in 30 years we'd be able to beat the world grandmaster at chess, and asked if that would count as AI, they'd say, of course it's AI! But now that we know how to do it, now it's no longer AI. Now it's just search."

> Then he makes the point that this requires embodiment, that they have a physical body in space

He makes even a stronger curious claim: That "This realization is often called the “embodiment problem” and most researchers in AI now agree that intelligence and embodiment are tightly coupled issues."

That makes it sound like it is a consensus between researchers, that strong AI requires a robotic body. Which I doubt.

So do I. But I'm mindful of the flawed assumptions made by the best minds in symbolic AI from the 1950's to the 1980's that AI required no more than raw facts and logic. Let's not repeat that mistake.

I'm intrigued by the potential of deep nets to 'simulate' the bases for symbolic grounding using something like thought vectors (TVs), especially if they can be revised and augmented through personal experience. Bootstrapping one's knowledgebase using someone else's rich set of TVs may go a long way for an AI to become grounded vicariously.

The first Chomskyan argument looks plausible (though still speculative) to me, but I share your skepticism of the ‘embodiment problem’ claim that a body and rich set of senses are necessary prerequisites. Anyone seduced by that argument should consider the life story of Helen Keller, who did have two-way interaction with the world, but only through the biological equivalent of a teletype.

That and virtual environments are continuing to grow in richness.

>I share your skepticism of the ‘embodiment problem’ claim that a body and rich set of senses are necessary prerequisites. Anyone seduced by that argument should consider the life story of Helen Keller, who did have two-way interaction with the world, but only through the biological equivalent of a teletype.

Helen Keller had smell, taste, touch, and proprioception all functioning. In what way did she have no senses but a teletype?

I was just looking for an analogy suggesting low bandwidth and experientially lean.

I mean, yes, it's low bandwidth, but it's actually high dimensionality, which has information-theoretic advantages.

Low-bandwidth is something of an understatement, especially when we consider activities like language acquisition, where smell, taste and proprioception play little part. Are the advantages of the dimensionality of this case sufficient to rescue the author's assertion that the embodiment problem requires robotics for its solution? Note that he used vision in his examples.

It occurred to me later that one can argue that Helen Keller, like the rest of us, was born with a brain shaped by hundreds of millions of years of interacting with the environment. I do not doubt that interaction is probably an essential part of us getting to AI, but I doubt that physical embodiment is the only way to do it.

Very disappointing article indeed.

Alpha Go is an AI project with historical relevance, that had a well specified and measurable goal, which is beating humans in a game that requires to develop an intuition for moves rather than just reading moves in advance.

Having said that I don't understand what's the point of trying to delegitimize the significance of Alpha Go. Implementing what we call "intuition" is not a minor thing.

In contrast, I encountered the Pepper robot in a mall in SF. It was almost impossible to interact with it, so if there is going to be a critique, why not starting from one of his own projects.

I think we can agree to disagree on the definition of intelligence.

Intelligence implies a lot more than pattern recognition. Current state is just a better way to treat large quantity of data. The smart is to feed the computer the right way with good data, iterate through different ways the computer can guess a model from the data and finally interpret the final results. Computers doesn't know how to do that. Yet.

> The smart is to feed the computer the right way with good data, iterate through different ways the computer can guess a model from the data and finally interpret the final results.

That's exactly how AlphaGo was trained.

Not very up to date are you...

So, basically the Chinese Room brouhaha all over again...

Just wait, next they'll discover the trolley problem and claim that self-driving vehicles will have to be philosophers.

Tis already happening. http://www.businessinsider.com/self-driving-cars-already-dec...

And this will be the outcome: http://www.smbc-comics.com/comic/self-driving-car-ethics

> things everyone knows

Everyone in the field. There are many more people outside of the field though.