On Fri, Oct 25, 2013 at 12:24 PM, Telmo Menezes <te...@telmomenezes.com>wrote:
> On Fri, Oct 25, 2013 at 1:11 AM, Platonist Guitar Cowboy > <multiplecit...@gmail.com> wrote: > > > > > > > > On Thu, Oct 24, 2013 at 11:29 PM, Telmo Menezes <te...@telmomenezes.com> > > wrote: > >> > >> On Thu, Oct 24, 2013 at 7:02 PM, Craig Weinberg <whatsons...@gmail.com> > >> wrote: > >> > > >> > > >> > On Thursday, October 24, 2013 12:43:49 PM UTC-4, telmo_menezes wrote: > >> >> > >> >> On Thu, Oct 24, 2013 at 6:39 PM, Craig Weinberg <whats...@gmail.com> > >> >> wrote: > >> >> > > >> >> > > >> >> > > http://www.theatlantic.com/magazine/archive/2013/11/the-man-who-would-teach-machines-to-think/309529/ > >> >> > > >> >> > The Man Who Would Teach Machines to Think > >> >> > > >> >> > "...Take Deep Blue, the IBM supercomputer that bested the chess > >> >> > grandmaster > >> >> > Garry Kasparov. Deep Blue won by brute force. For each legal move > it > >> >> > could > >> >> > make at a given point in the game, it would consider its opponent’s > >> >> > responses, its own responses to those responses, and so on for six > or > >> >> > more > >> >> > steps down the line. With a fast evaluation function, it would > >> >> > calculate > >> >> > a > >> >> > score for each possible position, and then make the move that led > to > >> >> > the > >> >> > best score. What allowed Deep Blue to beat the world’s best humans > >> >> > was > >> >> > raw > >> >> > computational power. It could evaluate up to 330 million positions > a > >> >> > second, > >> >> > while Kasparov could evaluate only a few dozen before having to > make > >> >> > a > >> >> > decision. > >> >> > > >> >> > Hofstadter wanted to ask: Why conquer a task if there’s no insight > to > >> >> > be > >> >> > had > >> >> > from the victory? “Okay,” he says, “Deep Blue plays very good > >> >> > chess—so > >> >> > what? > >> >> > Does that tell you something about how we play chess? No. Does it > >> >> > tell > >> >> > you > >> >> > about how Kasparov envisions, understands a chessboard?” A brand of > >> >> > AI > >> >> > that > >> >> > didn’t try to answer such questions—however impressive it might > have > >> >> > been—was, in Hofstadter’s mind, a diversion. He distanced himself > >> >> > from > >> >> > the > >> >> > field almost as soon as he became a part of it. “To me, as a > >> >> > fledgling > >> >> > AI > >> >> > person,” he says, “it was self-evident that I did not want to get > >> >> > involved > >> >> > in that trickery. It was obvious: I don’t want to be involved in > >> >> > passing > >> >> > off > >> >> > some fancy program’s behavior for intelligence when I know that it > >> >> > has > >> >> > nothing to do with intelligence. And I don’t know why more people > >> >> > aren’t > >> >> > that way...” > >> >> > >> >> I was just reading this too. I agree. > >> >> > >> >> > This is precisely my argument against John Clark's position. > >> >> > > >> >> > Another quote I will be stealing: > >> >> > > >> >> > "Airplanes don’t flap their wings; why should computers think?" > >> >> > >> >> I think the intended meaning is closer to: "airplanes don't fly by > >> >> flapping their wings, why should computers be intelligent by > >> >> thinking?". > >> > > >> > > >> > It depends whether you want 'thinking' to imply awareness or not. > >> > >> Ok. I don't think we can know that in any case. > >> > >> > I think > >> > the point is that we should not assume that computation is in any way > >> > 'thinking' (or intelligence for that matter). I think that 'thinking' > is > >> > not > >> > passive enough to describe computation. It is to say that a net is > >> > 'fishing'. Computation is many nets within nets, devoid of intention > or > >> > perspective. It does the opposite of thinking, it is a method for > >> > petrifying > >> > the measurable residue or reflection of thought. > >> > >> Ok but let's take a human grand master playing chess. You don't think > >> a computer can play like him? > >> > > > > This relates to what you said earlier which I agree with: > > > > They are more intelligent in > > the sense that they can play competitively while only considering a > > small fraction of the scenarios. How do we do this? There is almost no > > real AI research nowadays because people gave up on answering this > > question. > > > > The answer lies somewhere in building branch histories and databases that > > are for now only partial. The computer cannot beat humans without > databases > > for openings, middle, and endgame. I believe this is what freaked out > > Kasparov in the questionable game and what gives his suspicion of human > > intervention in the code, which IBM never ruled out or proved negatively > > between games, some substance. Kasparov lost because IBM eventually > accrued > > enough understanding of Kasparov's database (dozens of years of notes and > > logs that make up his holy grail secret) to not let it fall for > Kasparov's > > gambit. > > > > Kasparov's and any GM's algorithm for beating chess engines often runs > along > > the lines of: > > > > Keep position closed via Botvinnik type openings and middlegame so the > > computer will have to contend with billions of possible move > continuations > > instead of a few dozen million. Then implement precise, but highly > complex, > > long term strategy that offers both positional and material gambit for > > twenty or so moves which is designed to flip at exactly the point of the > > computer's computational horizon, and the computer loses. > > > > This doesn't work today, because human GMs have fed the databases with > every > > line/variation up their sleeves (from hundreds of years of recorded > games) > > and consequently we feed the software with every refutation. Once a > > refutation is implemented, it's our loss in terms of raw summing value, > > because we are forced into unexplored territory and can't compute those > > positions optimally anymore (no more strategy, just local computational > > tactics; computer wins). > > > > This is vague, so I'll give concrete example: Marshall Attack for black > > against white Ruy Lopez. This one is odd and even powerful chess engines > > that don't include this in their databases, that can brute force > billions of > > position assessment sums (more than deep blue could on a desktop of > today), > > will loose even against yours truly! :-) > > > > Why? Because black gives up central control, sacrifices a central pawn > and > > can get away with savage sacrifices of minor and major pieces and loss of > > positional strength; all things the computer is summing highly peering > and > > brute forcing 10 or 20 moves in advance to calculate an undecidable > flurry > > of too many best possible moves, in function of finding itself in an > > incredible advantage. It doesn't see black's long term but explosive > > minority attack king side unless it has a database to draw from at a much > > earlier stage in the game, innocently conspiring against pieces that are > > irrelevant to the plan. > > > > At the precise point my engine sees that branch in which black mates > white > > suddenly, it's already too late. The board is already set and the > computer's > > positional and material advances are irrelevant. I let my engine evaluate > > positions without database, and it sees black losing clearly a couple of > > points (half a point can tip a game in GM level chess) until on some > > billionth branch it discovers black's long term intent, at which > suddenly, > > in computing on one move, black gets a ten point lead despite material > and > > positional sacrifice because the computer sees suddenly that unavoidably > > white is mated. > > > > Of course, with database all the computer has to do is refutation move > like > > pawn a4 early enough, which prevents the mess above at its root in the > game. > > But we don't have algorithms to find these really vague (in terms of > > position and material) but precise variations (in terms of fuzzy felicity > > conditions for long term strategy; which pieces of black become crucial > is > > open depending on particular game, save the queen, so you can't define > "if > > black sacrifices piece x this indicates gambit" independent of position > for > > example), even though now computers are helping GMs find these types of > > variations! > > > > Therefore, the whole pitting "human vs. computer" makes little sense to > me > > in chess until we have computing power to solve chess completely or > prove it > > undecidable. Human and computer are playing the same game, and without > the > > right memory and branch, both are lost. I guess a GM panel with complete > > access to code, databases, and hardware specs could, given sufficient > time > > that nobody would finance, find lines long and complex enough to beat a > > computer, and a GM panel without the code and in an unfamiliar branch > > doesn't have a chance against an entertainment engine they sell on > itunes! > > > > The question regarding superiority should be: what are we even proving > here, > > given we write the code and we designed the game? Right, the distinction > > between machine and human becomes blurry. Every discovery of line > enriches > > the game up to this point. Even when or if a super computer solves > chess, it > > will not have done so without the notes of monks' games in the 15th > century. > > PGC > > PGC, > > Thanks for the detailed explanation from someone who actually knows > something about chess. I will keep this in my files :) > > My high-level objection is very simple: chess was an excuse to pursue > AI. In an era of much lower computational power, people figured that > for a computer to beat a GM at chess, some meaningful AI would have to > be developed along the way. I don' thing that Deep Blue is what they > had in mind. IBM cheated in a way. I do think that Deep Blue is an > accomplishment, but not _the_ accomplishment we hoped for. > I don't disagree with this. Deep Blue was pure PR to push PC revolution. GMs play each other, more for fun than anything else, with strong chess engines by their side. This combines long term strategical experience with the computer offering best possible moves. The GM then either decides that one of the computer's suggestions is actually better than what she/he had in mind strategically, or that the computer is deviating from the GM's strategy because they don't see far enough, or understand psychology, aims, strategy of the opponent. That's what it boils down to. Given ceilings in performance and precise limitations of software (some computational horizon), the computer psychology can be pried open and lines can be found that exceed these limitations, like with any other opponent! A recent example of how a powerful engine is unusually defeated by a GM that understands its psychology is found in the following link, which is unfortunately tainted by "Human is better than computer"- ideology nonesense; nonetheless, you can get picture about how a GM recently proceeded to beat, in this case, to humiliate an engine that can evaluate more positions in a game, than he can in his life. http://www.chess.com/article/view/computers-in-chess-good-or-evil-part-two Then again, based on this game, you would tweak the software accordingly to avoid the trap. And the GMs would have to exploit other limitations and vulnerabilities. PGC > > I believe there will be an AI renaissance and I hope to be alive to > witness it. But for this renaissance to take place, I think two > cultural shifts have to happen: > > - A disinterest with the "science as the new religion" stance, leading > to a truly scientific detachment from findings. Currently, everything > that touches the creation of intelligence is ideologically loaded from > all sides of the discussion. This taints honest scientific inquiry; > > - New economic structures that allow humanity to pursue complex goals > outside the narrow short-term focus on profit of corporatism or the > pointless status wars of academia. > That's the real tragedy imho. Even if somebody were to develop highly effective learning algorithms for more general sets of problems, they'd have no interest to share it, because NSA, Google, Facebook, defense departments, banks pay better short term than public interest in the long run. But there will be rebels and double agents, as always... and they need your money, lol! PGC > > Best, > Telmo. > > > > > > > > > > > > > > >> > >> > > >> >> > >> >> > -- > >> >> > You received this message because you are subscribed to the Google > >> >> > Groups > >> >> > "Everything List" group. > >> >> > To unsubscribe from this group and stop receiving emails from it, > >> >> > send > >> >> > an > >> >> > email to everything-li...@googlegroups.com. > >> >> > To post to this group, send email to everyth...@googlegroups.com. > >> >> > Visit this group at http://groups.google.com/group/everything-list > . > >> >> > For more options, visit https://groups.google.com/groups/opt_out. > >> > > >> > -- > >> > You received this message because you are subscribed to the Google > >> > Groups > >> > "Everything List" group. > >> > To unsubscribe from this group and stop receiving emails from it, send > >> > an > >> > email to everything-list+unsubscr...@googlegroups.com. > >> > To post to this group, send email to everything-list@googlegroups.com > . > >> > Visit this group at http://groups.google.com/group/everything-list. > >> > For more options, visit https://groups.google.com/groups/opt_out. > >> > >> -- > >> You received this message because you are subscribed to the Google > Groups > >> "Everything List" group. > >> To unsubscribe from this group and stop receiving emails from it, send > an > >> email to everything-list+unsubscr...@googlegroups.com. > >> To post to this group, send email to everything-list@googlegroups.com. > >> Visit this group at http://groups.google.com/group/everything-list. > >> For more options, visit https://groups.google.com/groups/opt_out. > > > > > > -- > > You received this message because you are subscribed to the Google Groups > > "Everything List" group. > > To unsubscribe from this group and stop receiving emails from it, send an > > email to everything-list+unsubscr...@googlegroups.com. > > To post to this group, send email to everything-list@googlegroups.com. > > Visit this group at http://groups.google.com/group/everything-list. > > For more options, visit https://groups.google.com/groups/opt_out. > > -- > You received this message because you are subscribed to the Google Groups > "Everything List" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to everything-list+unsubscr...@googlegroups.com. > To post to this group, send email to everything-list@googlegroups.com. > Visit this group at http://groups.google.com/group/everything-list. > For more options, visit https://groups.google.com/groups/opt_out. > -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/groups/opt_out.
