History’s greatest grandmaster Garry Kasparov just wrote an in-depth review of Diego Rasskin-Gutman’s newly published “Chess Metaphors: Artificial Intelligence and the Human Mind” for the New York Review of Books. The article provides a spectacular summary of the history of computer chess programs and the AI engines that power them. What makes it a must-read, however, are Kasparov’s personal anecdotes about his own experiences playing computers, including, of course, his notorious loss to IBM’s Deep Blue in their 1997 rematch (Kasparov won the first match in 1996). This not only marked the first time a computer defeated a world chess champion, but was also seen around the world, for better or for worse, as the conclusive tipping point when raw machine intelligence overcame what we mere humans could handle.

Kasparov occupies an entirely unique post in the history of world chess champions – the tables turned during his guard. He started out in the mid-80s being able to beat the most sophisticated chess programs conclusively every single time; during the late-90s the exact reverse situation materialized. In fact, by the beginning of the new millennium $50 commercially available chess programs running on regular PCs could “crush most grandmasters” and when Kasparov last played two serious matches with such programs (in 2003) they both ended in a tie.

Of course Kasparov reminds as that just because computers can now decisively win at chess doesn’t mean that “chess is solved.” He writes:

“The number of legal chess positions is 10⁴⁰, the number of different possible games, 10¹²⁰. Authors have attempted various ways to convey this immensity, usually based on one of the few fields to regularly employ such exponents, astronomy…Diego Rasskin-Gutman points out that a player looking eight moves ahead is already presented with as many possible games as there are stars in the galaxy. Another staple…is to say there are more possible chess games than the number of atoms in the universe.”

Even if there were such a thing as “solving chess” – that is, both sides continuously playing the perfect game – these impressive numbers highlight how mathematically difficult, probably impossible it would be to actually accomplish this feat. And even if this were hypothetically possible, Kasparov notes the importance of not confusing narrow AI applications, such as playing chess, with general intelligence of the sort that allows humans to think, intuit, dream, walk up stairs, wash the dishes – to generally lead the kind of lives expected of (relatively) advanced sentient beings. Chess has been used for centuries as the ultimate metaphor for the mind, but Kasparov finally concludes that “perhaps chess is the wrong game for the times.”

Enter: Poker
Many players, including myself and several of my friends, started out playing poker directly from the chess world; yes the opportunity for profit is greater, but it is also because poker is in many ways more complex and challenging than chess. While chess is a 100% information game and entirely susceptible to computer calculation, poker is defined by hidden information and contains nuanced elements such as tells, bluffing, and emotional control which span distinct fields such as social psychology, risk-management, and game theory – all aspects that make it significantly more problematic both for players and AI programmers to master.

Something I have been advocating for several years, and it seems that Kasparov and others agree: Poker is where AI gaming researchers should be focused. Whereas chess programs can now consistently beat anyone in the world, no one has yet figured out how to beat advanced human players at poker. And it’s surely not for lack of effort – there are dozens of entities working on this problem every day. In fact, some claim that they have already built software that profitably beats online poker (for example: “How I Built a Working Poker Bot”). There are many others out there, and some make the claim that they are consistently winning. Of course it is difficult to assess the success of these programs – especially since many of them are likely not even public – yet I remain skeptical. It is probably possible to program something to beat the microlimits, but entering the realm of even mid-stakes amateur play is an entirely different ballgame.

I write this not to disregard people’s present efforts, but to encourage AI researchers to view the poker platform with the same enthusiasm and seriousness that I believe it deserves. The many disparate fields from which players must simultaneously draw information and the very social nature of the game make it perfectly suited for some truly interesting research of advanced gaming AI, and, dare I say it, even research dealing with problems of strong AI. At least it is surely more accurate than chess as a metaphor both of the mind and of the way humans interact with the world. A few brief reasons:

-Minds operate on heuristics, not algorithms.
-Our existence can be summed up as making choices in the face of limited information.
-We operate within a social context.
-Information that is presented to you may be false, and will surely be false >0% of the time.
-Individual instances of “chance” exist and do matter.
-There are limited resources and different entities have varying amounts of control over those resources.
-There are many variables from varying realms that go into making even the simplest decisions.

Let’s go beyond poker bots that can at best squeeze out profits playing for pennies, and begin thinking about a poker AI than can adapt to individual players, can make unwarranted bluffs only to establish a crazy image for uncertain future gains, can vary its playing style – in short, can outthink its opponents. Perhaps while venturing down this road of creating a social, competitive agent, one driven by the same impulses of survival and will towards progress that guide biological life, we will come one step closer to the birth of a true thinking machine. Just be careful: never trust a poker player.

[The Chess Master and the Computer, Garry Kasparov - New York Review of Books]
[Cylons Playing Poker via Anthony J. Cox]

Virtual worlds are a natural setting for experimenting and testing various AI applications and appear especially suited for “general” intelligence development because of their broadness, openness and theoretical lack of limitations and rules. Ben Goertzel and his company, Novamente, have been working on virtual pets and the results are pretty interesting. So while projects such as these are something to get excited about, it’s hard not to dismiss bigheaded goals as worthless hype without seeing the real work that has gone into specific planning and execution – at least something that does not recall the infamous Underpants Gnome plan of 1)Collect underpants 2)??? 3)Profit. So while I don’t think much will come out of this, I have to issue the small warning that the distinction between “virtual” and “real” is pretty nonexistent. All that data is comprised of real code stored on real servers with real physical locations managed by real people… And again while I do remain skeptical of anyone achieving anything close to sentient intelligence anytime soon, I am also a bit concerned about singular commercial entities working secretly on a closed AGI system. Not involving all stakeholders – that is, ALL OF US – raises some ethical quandaries that at least need to be debated.

I would put someone like Philip Rosedale (proven record, visionary thinker, deep pockets and connections) pretty high on the list of people that could make some serious headway in the field, especially in potential commercial applications; he certainly tackles the big issues. Let’s just hope Philip Rosedale doesn’t resemble Watchmen’s Ozymandias…

[New World Notes - Philip Rosedale Attempting to Create Sentient Artificial Intelligence That Thinks and Dreams in Second Life!]

Ben Goertzel, an AGI researcher and CEO of Novamente, organized the conference and provided a very good roundup to readers of h+. He explains that the AGI conference is smaller and more intimate than other related gatherings because it specifically focuses on the “big picture:” human-level thinking machines.

“Since its founding in the 1950s and 1960s, the AI field has achieved great successes – to name just a few: the AI linguistics underlying Google and other search engines; the AI planning and scheduling software used throughout the military and industry; the AI fraud detection software underlying modern credit card operations; and the AI gaming software underlying everything from Deep Blue to the bots in massively multiplayer online games. Yet all these wonderful achievements have a common narrowness of scope, which is why inventor and futurist Ray Kurzweil has characterized them as “narrow AI.” The grand goal of the original AI researchers – the creation of thinking machines with general intelligence at the human level and ultimately beyond — remains largely unaddressed. Many, both within and outside the AI field, have complained about this situation; the mission of the AGI conference series is to help do something about it.”

Reading Ben’s report makes one realize just how difficult it is to piece together highly-specialized aspects of learning, cognition, computer science, etc. to form something akin to a tapeworm – let alone a human mind. However, I’m confident that the eureka moment will come in our lifetime, even if we don’t fully understand how we got there, or how our creation will actually work…

[Ben Goertzel's Report on AGI-09: The Second Conference on Artificial General Intelligence]

The school is designed to bring together the brightest minds from a multiplicity of academic fields and provide an interdisciplinary framework to better understand (and bring about) the future of technology. Cost for the 9-week summer course is slated at $25,000; since it’s the first year, it’s hard to say how hard it will be to get accepted and how much they will give out in aid, though my guess is hard and little.

The curriculum follows 10 tracks:

* future studies and forecasting
* networks and computing systems
* biotechnology and bioinformatics
* nanotechnology
* medicine, neuroscience and human enhancement
* AI, robotics, and cognitive computing
* energy and ecological systems
* space and physical sciences
* policy, law and ethics
* finance and entrepreneurship

Though this is pretty exciting stuff, and I most definitely support its existence, it seems a bit of a misnomer to call it a “University;” it appears to be more of a business trade show/conference than a school. It’s also quite likely that people who are willing to shell out 25K on a summer course possess ulterior business motives and are piggybacking on nerds to sniff out the “next big thing” while establishing the proper Valley connections. Finally, based on current info, it looks like the SU will just be teaching interested parties about various Singularity-related academic topics, and not performing ANY actual research whatsoever. And isn’t the point of the Singularity that it will pretty much happen no matter what we do… that the seeds are sown, and barring some massive catastrophe that blasts us back to the Stone Age, we are just waiting for the trigger? Personally, I’ll be following the people at the trenches, who, for example, are working and thinking extremely hard on how to actually create a seed AI, rather than the “students” attending a glorified Google networking event.

[Singularity University]
[via TechCrunch]

New Scientist did a short review documenting contemporary approaches in solving the Turing Test, a simple experimental question posed by mathematician Alan Turing in 1950: can a computer intelligence involved in casual conversation convince human judges that it too is human? For over half a century researchers have attempted to build something that can pass the test; as the New Scientist review points out, contemporary attempts have not been limited solely to conversational analysis, but in fact span many variations, including musical (a Jazz piano that can improv with a human), aesthetic (life-like androids, and also see our post on virtual actors), military (weapons deciding when to fire themselves)…

…and of course, pwning at Unreal Tournament! Realistic AI engines are a huge component in game development and greatly contribute to players’ enjoyment and engagement levels. Proving that these issues are not only limited to research labs, moneyed corporations, and theoretical musings of cognitive scientists and philosophers, Australian game developer 2K offered anyone that could design a bot that passes the ‘bot Turing Test’ $7,000 cash and an all-expense-paid trip to their studios down under. The contest was recently held at the IEEE Symposium on Computational Intelligence and Games in Perth, Australia. Slashdot reports:

“Teams from Australia, the Czech Republic, the United States, Japan and Singapore competed in the final. Competitors created bots to play a specially modified Unreal Tournament 2004 Death Match. Expert judges then tried to tell whether they were playing a bot or a human, just from their observation of the way they played the game. Judges included AI experts, a game development executive, game developers, as well as an expert human player. The result? The winning team AMIS, from Charles University in Prague, managed to fool 2 out of the 5 expert judges, and achieved an average ‘human-ness rating’ of 2.4 out of 4. All the human players were judged more human than the bots overall, but the judges were fooled often enough to suggest that in next year’s contest, some bots may be able to pass the test by fooling 4 out of 5 judges. AMIS won $2,000 cash plus an all expenses paid trip to 2K’s Canberra studio. You can check out the full results and competition videos, and try an online video quiz that lets you judge for yourself.”

[New Scientist: Tests that show machines closing in on human abilities]
[botprize 2008], via [Slashdot]

“What will change everything? What game-changing scientific ideas and developments do you expect to live to see?”

Well, since they asked, I’ll chime in with my plebeian perspective: the creation of recursively self-improving AI. The actual results of such an event are totally unimaginable (beyond our current abilities) but the fact that it will “change everything” is almost certain. Nick Bostrom, a philosopher at Oxford, articulates my personal beliefs quite well in his answer to Edge’s big question (read after the jump).

In addition to Nick Bostrom, a few other participating individuals that I have read or followed in recent years include Aubrey de Grey and Brookyln’s own Douglas Rushkoff. There is a great overview done by Next Big Future, which points out that 30 of the 150 responses are explicitly transhumanist. However, I encourage readers to browse many of the responses; the thoughts are as widespread as they are interesting. What do YOU think?

[Edge.org Big Question 2009]
[Next Big Future: Analyzing the Edge 2009 Question and Answers]

“NICK BOSTROM
Philosopher, University of Oxford; Editor, Human Enhancement

SUPERINTELLIGENCE

Intelligence is a big deal. Humanity owes its dominant position on Earth not to any special strength of our muscles, nor any unusual sharpness of our teeth, but to the unique ingenuity of our brains. It is our brains that are responsible for the complex social organization and the accumulation of technical, economic, and scientific advances that, for better and worse, undergird modern civilization.

All our technological inventions, philosophical ideas, and scientific theories have gone through the birth canal of the human intellect. Arguably, human brain power is the chief rate-limiting factor in the development of human civilization.

Unlike the speed of light or the mass of the electron, human brain power is not an eternally fixed constant. Brains can be enhanced. And, in principle, machines can be made to process information as efficiently as — or more efficiently than — biological nervous systems.

There are multiple paths to greater intelligence. By “intelligence” I here refer to the panoply of cognitive capacities, including not just book-smarts but also creativity, social intuition, wisdom, etc.

Let’s look first at how we might enhance our biological brains. There are of course the traditional means: education and training, and development of better methodologies and conceptual frameworks. Also, neurological development can be improved through better infant nutrition, reduced pollution, adequate sleep and exercise, and prevention of diseases that affect the brain. We can use biotech to enhance cognitive capacity, by developing pharmaceuticals that improve memory, concentration, and mental energy; or we could achieve these ends with genetic selection and genetic engineering. We can invent external aids to boost our effective intelligence — notepads, spreadsheets, visualization software.

We can also improve our collective intelligence. We can do so via norms and conventions — such as the norm against using ad hominem arguments in scientific discussions — and by improving epistemic institutions such the scientific journal, anonymous peer review, and the patent system. We can increase humanity’s joint problem-solving capacity by creating more people or by integrating a greater fraction of the world’s existing population into productive endeavours, and we can develop better tools for communication and collaboration — various internet applications being recent examples.

Each of these ways of enhancing individual and collective human intelligence holds great promise. I think they ought to be vigorously pursued. Perhaps the smartest and wisest thing the human species could do would be to work on making itself smarter and wiser.

In the longer run, however, biological human brains might cease to be the predominant nexus of Earthly intelligence.

Machines will have several advantages: most obviously, faster processing speed — an artificial neuron can operate a million times faster than its biological counterpart. Machine intelligences may also have superior computational architectures and learning algorithms. These “qualitative” advantages, while harder to predict, may be even more important than the advantages in processing power and memory capacity. Furthermore, artificial intellects can be easily copied, and each new copy can — unlike humans — start life fully-fledged and endowed with all the knowledge accumulated by its predecessors. Given these considerations, it is possible that one day we may be able to create “superintelligence”: a general intelligence that vastly outperforms the best human brains in every significant cognitive domain.

The spectrum of approaches to creating artificial (general) intelligence ranges from completely unnatural techniques, such as those used in good old-fashioned AI, to architectures modelled more closely on the human brain. The extreme of biological imitation is whole brain emulation, or “uploading”. This approach would involve creating a very detailed 3d map of an actual brain — showing neurons, synaptic interconnections, and other relevant detail — by scanning slices of it and generating an image using computer software. Using computational models of how the basic elements operate, the whole brain could then be emulated on a sufficiently capacious computer.

The ultimate success of biology-inspired approaches seems more certain, since they can progress by piecemeal reverse-engineering of the one physical system already known to be capable of general intelligence, the brain. However, some unnatural or hybrid approach might well get there sooner.

It is difficult to predict how long it will take to develop human-level artificial general intelligence. The prospect does not seem imminent. But whether it will take a couple of decades, many decades, or centuries, is probably not something that we are currently in a position to know. We should acknowledge this uncertainty by assigning some non-trivial degree of credence to each of these possibilities.

However long it takes to get from here to roughly human-level machine intelligence, the step from there to superintelligence is likely to be much quicker. In one type of scenario, “the singularity hypothesis”, some sufficiently advanced and easily modifiable machine intelligence (a “seed AI”) applies its wits to create a smarter version of itself. This smarter version uses its greater intelligence to improve itself even further. The process is iterative, and each cycle is faster than its predecessor. The result is an intelligence explosion. Within some very short period of time — weeks, hours — radical superintelligence is attained.

Whether abrupt and singular, or more gradual and multi-polar, the transition from human-level to superintelligence would of pivotal significance. Superintelligence would be the last invention biological man would ever need to make, since, by definition, it would be much better at inventing than we are. All sorts of theoretically possible technologies could be developed quickly by superintelligence — advanced molecular manufacturing, medical nanotechnology, human enhancement technologies, uploading, weapons of all kinds, lifelike virtual realities, self-replicating space-colonizing robotic probes, and more. It would also be super-effective at creating plans and strategies, working out philosophical problems, persuading and manipulating, and much else beside.

It is an open question whether the consequences would be for the better or the worse. The potential upside is clearly enormous; but the downside includes existential risk. Humanity’s future might one day depend on the initial conditions we create, in particular on whether we successfully design the system (e.g., the seed AI’s goal architecture) in such a way as to make it “human-friendly” — in the best possible interpretation of that term.”