28-Jan-16 World View -- China's game of Go beaten by Google's AI software, bringing the Singularity closer

[John]

28-Jan-16 World View -- China's game of Go beaten by Google's AI software, bringing the Singularity closer


Google's AI technology will soon start taking people's jobs

** 28-Jan-16 World View -- China's game of Go beaten by Google's AI software, bringing the Singularity closer
** http://www.generationaldynamics.com/pg/ ... tm#e160128



Contents:
China's game of Go beaten by Google's AI software program AlphaGo
Google DeepMind's Go algorithm uses massive pattern matching neural networks
Google's AI technology will soon start taking people's jobs


Keys:
Generational Dynamics, Google, artificial intelligence, AI, DeepMind,
AlphaGo, Singularity, minimax algorithm, neural networks,
value networks, policy networks, Fan Hui, Lee Sedol

[Enough]

I'm not sure if we will ever truly reach Singularity, but I have serious doubts that every job will be replaced by computers.

First of all, mass unemployment will lead to revolution. But that's the best case scenario in my book.

Secondly, let's say that there is no revolt and people end up living in small concrete studio apartments with a smart phone (or smart chip in heads). To what end? More social isolation? People need human interaction. Social media is already driving people mad! I don't want a phone app teacher or a robot sales girl. That's nightmarish. That's hell. Unless none of this will matter because people will soon not matter anyway...

I'm going to start a luddite colony somewhere. There will always be people who still value human interaction, home grown food, and hand made goods. And the way things are going, it might just be the only way to survive, physically and mentally.

[jldavid47]

I'll be more impressed when a team of computer partners can win a World Bridge Championship against a team of top human experts. Chess and Go are games where there is complete information in front of you. The only real question is what is your opponent going to do. Bridge is a game of incomplete information and inference. You only see your hand and the dummy. You can only communicate with your partner using bidding and carding agreements. Even though all of those agreements are disclosed to the opponents, the human can deviate from those agreements and part of the game is deciphering when that might be happening. Frankly, bridge is a much more difficult game to "solve" than chess or go. I think the same is true with poker. There will be no singularity until computers conquer humans in those games.

[John]

> I'm going to start a luddite colony somewhere. There will always
> be people who still value human interaction, home grown food, and
> hand made goods. And the way things are going, it might just be
> the only way to survive, physically and mentally.

Once again, using the ape analogy, apes don't intermingle with
humans, but have their own non-human colonies in the jungle.

> I'll be more impressed when a team of computer partners can win a
> World Bridge Championship against a team of top human
> experts. Chess and Go are games where there is complete
> information in front of you. The only real question is what is
> your opponent going to do. Bridge is a game of incomplete
> information and inference. You only see your hand and the
> dummy. You can only communicate with your partner using bidding
> and carding agreements. Even though all of those agreements are
> disclosed to the opponents, the human can deviate from those
> agreements and part of the game is deciphering when that might be
> happening. Frankly, bridge is a much more difficult game to
> "solve" than chess or go. I think the same is true with
> poker. There will be no singularity until computers conquer humans
> in those games.

> Game theorists crack poker

> An 'essentially unbeatable' algorithm for the popular card game
> points to strategies for solving real-life problems without having
> complete information. ...

> A new computer algorithm can play one of the most popular variants
> of poker essentially perfectly. Its creators say that it is
> virtually “incapable of losing against any opponent in a fair
> game”. ...

> The strategy the authors have computed is so close to perfect “as
> to render pointless further work on this game”, says Eric Jackson,
> a computer-poker researcher based in Menlo Park, California.

> “I think that it will come as a surprise to experts that a game
> this big has been solved this soon,” Jackson adds. ...

> But poker is harder to solve than draughts. Chess and draughts are
> examples of perfect-information games, in which players have
> complete knowledge of all past events and of the present situation
> in a game. In poker, in contrast, there are some things a player
> does not know: most crucially, which cards the other player has
> been dealt. The class of games with imperfect information is
> especially interesting to economists and game theorists, because
> it includes practical problems such as finding optimal strategies
> for auctions and negotiations. ...

> Bowling and colleagues designed their algorithm so that it would
> learn from experience, getting to its champion-level skills
> required playing more than 1,500 games. At the beginning, it made
> its decisions randomly, but then it updated itself by attaching a
> 'regret' value to each decision, depending on how poorly it
> fared. ...

> As part of its developing strategy, the computer learned to inject
> a certain dose of bluffing into its plays. Although bluffing seems
> like a very human, psychological element of the game, it is in
> fact part of game theory — and, typically, of computer
> poker. “Bluffing falls out of the mathematics of the game,” says
> Bowling, and you can calculate how often you should bluff to
> obtain best results.

> Of course, no poker algorithm can be mathematically guaranteed to
> win every game, because the game involves a large element of
> chance based on the hand you’re dealt. But Bowling and his
> colleagues have demonstrated that their algorithm always wins in
> the long run.

http://www.nature.com/news/game-theoris ... er-1.16683

[zorbopondo2]

Until someone writes a general purpose AI that can play chess and go and bridge and quickly discover, write, and publish in peer-reviewed journals in various languages new, interesting and original research in each of 20 physical, biological, chemical, literary, historical, economic, philosophical, and mathematical niche fields that I might specify after the AI has been fully coded, I won't even admit to the possibility of singularity.

I have in mind niches like propagation of radar wave in crystal lattices, or behavior of nematodes missing a specific ganglion, or an improvement to international trade policy, or a better account of the thirty years war,...

I mean, if an AI can't even do these little tasks, how on earth is a singularity even possible?
Or rather if the singularity is possible, then long, long, long, long before it happens, all the tests I mention above will have become commonplace.

Of course no one has the slightest idea how to write such an AI. The best anyone can do is make these single-purpose toys, which in the larger historical picture will be of no more significance than similar projects from the past. The Turk for example https://en.wikipedia.org/wiki/The_Turk.

[John]

> Until someone writes a general purpose AI that can play chess and
> go and bridge and quickly discover, write, and publish in
> peer-reviewed journals in various languages new, interesting and
> original research in each of 20 physical, biological, chemical,
> literary, historical, economic, philosophical, and mathematical
> niche fields that I might specify after the AI has been fully
> coded, I won't even admit to the possibility of singularity. ...

> Of course no one has the slightest idea how to write such an AI.

A lot of this stuff is already done, and the rest will be done within
a few years.

It's already been five years since IBM's Watson software was able to
beat human Jeopardy! champions in a wide variety of subject areas,
including physical, biological, chemical, literary, historical,
economic, philosophical, and mathematical niche fields.

** 19-Feb-11 News -- IBM's Watson supercomputer bests human champions on Jeopardy!
** http://www.generationaldynamics.com/pg/ ... tm#e110219

AI algorithms can already do original research in very limited
domains, such as mathematical theorems. These algorithms will be
expanded to all domains in the next few years as sufficient computer
power and storage becomes available.

I suspect we're very close to a breakthrough where technology like
IBM's Watson can be used to "read" and absorb everything on the
internet, and be able to provide expert analysis on subjects in any
domain. Add to that existing algorithms to do original research, and
everything you've described will be accomplished.

[Guest]

Test

[Guest]

Once again, using the ape analogy, apes don't intermingle with
humans, but have their own non-human colonies in the jungle.

What is that supposed to mean?

[JimN]

As a life-long student of AI, I find myself in a camp that lies somewhat between John's (and Ray Kurzweil's) position (that the Singularity is somewhat near) and the various naysayers doubtfulness that it will come to fruition.

For instance, having just completed the IBM-sponsored 'Introducing Watson' college course, I can say that the hard part of Watson is finding and curating the documents (source material) - so that perhaps X jobs doing some task might be replaced with Y jobs curating the data so that Watson can do the task (where presumably Y < X but Y is not asymptotically approaching zero).

And although the emerging new Data Science (the area I'm currently most interested in), which uses (among other techniques) neural-network based approaches to machine learning, will make great strides in the next 15 years, I suspect it will only increase the demand for software engineers. I don't think that all current computer programmers should run for the hills and start their own Luddite colonies. Perhaps some of the more repetitive computer jobs might go the way of the abacus. There certainly is the prospect of net job loss due to continually encroaching technology - which when added to robotic manufacturing might compound that job loss. That overall phenomenon is what I fear the most (right after idiotic terrorists, that is). The real problem hear is Malthusian in nature - that is, we're procreating millions of uneducated, unemployable humans at an astounding rate, and in the face of less and less jobs for uneducated people - and this very well might lead to anarchy and revolution of some sort, somewhere down the line. But as a long-time AI-interested person, I don't think pulling the plug on AI research is the solution either. The only solution is social education on the global impact of a local (unprotected) sex act. One could argue all day whether there will be a Singularity, and/or whether it will be 5 or 15 or 30 years from now. The early AI researchers predicted 10 years, and that was in the 1960s - so they were way off. And although the Go example is a neat breakthrough in the use of neural networks and computer-self-instruction, that is not new either. We (Newhouse and Grant) used millions of training games to nearly solve Qubic in the 1970s, and Samuel employed similar techniques to eventually solve the game of checkers. And neural nets have been with us for a long time, although they've had a recent resurgence in the context of Data Science - so that the neat application of that technology to Go is a sign of the fact that we have (a) more well organized efforts these days (on the scale of IBM's Watson effort); and (b) giant advances in computer speed, memory size, and sophisticated distributed architectures employing thousands of loosely coupled computer CPUs. Cheers to Google.

[John]

> I'm going to start a luddite colony somewhere. There will always
> be people who still value human interaction, home grown food, and
> hand made goods. And the way things are going, it might just be
> the only way to survive, physically and mentally.

> Once again, using the ape analogy, apes don't intermingle with
> humans, but have their own non-human colonies in the jungle.

> What is that supposed to mean?

I was responding to a message about starting a luddite colony
somewhere by "people who still value human interaction, home grown
food, and hand made goods." The implication was that the alternative
was living among the robots, with no human interaction or home grown
foods. I was saying that apes still live in their own colonies in the
jungle, can still interact with each other, and still get to eat what
they want. So humans can act in a similar way with the robots.