| T O P I C R E V I E W |
| dr strangelove |
Posted - 11/03/2016 : 16:05:46
Mrzim otvarati nove topice i nadao sam se da će to netko drugi "odraditi", ali nitko ne spominje ništa pa očito moram ja.
Anyway, Lee Sedol je deveti dan (najviši) igre Go. Riječ je o jednom od najboljih igrača posljednjih godina, možda čak i najboljem. On se odlučio okušati protiv računala AlphaGo u 5 mečeva. Po čemu se to razlikuje od šahovskih mečeva protiv računala? Zato što je ovo sljedeći korak u računalnoj evoluciji. Ako smijem povući usporedbu: ako je pobjeda Deep Bluea nad Kasparovom značila da smo naučili dijete puzati, onda bi pobjeda AlphaGoa značila da je smo to isto dijete naučili voziti bicikl. Pretjerivanje? Dovoljno je reći da je broj poteza u Gou veći nego broj atoma u poznatom svemiru. Ozbiljno, slobodno proguglajte.
Prije susreta Korejanac je rekao da da će pobijediti 5:0 ili 4:1. Prvi susret se odigrao prije par dana i AlphaGo je iznenađujuće poveo. Lee je imao obećavajuću situaciju na ploči i vremensku prednost, ali je računalo nekako uspjelo sve to preokrenuti u svoju korist. Kažem 'nekako' jer je to iznad običnog smrtnika poput mene. Uglavnom se svi slažu da je Lee krivo pročitao stanje u donjem desnom kutu. Nakon toga je većina je smatrala, ili se barem nadala, da je Lee uspio pročitati taktiku računala i način igranja pa da će u drugom susretu izjednačiti rezultat (iako obično računala završe s ukupnom pobjedom kada povedu u seriji). U isto vrijeme, Lee je sada svoje šanse smanjio na "samo" 60%. Prije početka druge partije vidjelo se da je Lee nervozan i/ili da nije dobro spavao. Krenuo je oprezno i - izgubio. Šteta jer bilo je neizvjesno, a Lee je inače maher kad je igranje pod pritiskom u pitanju. Moglo bi se reći da je Lee izgubio na svom terenu, u samoj završnici partije. Sada Skynet vodi... pardon, AlphaGo. Sada AlphaGo vodi dva razlike i čovječanstvu se bome ne piše dobro.
Oko partija se lome mnoga koplja. Dok neki misle da je računalo vuklo Leea za nos pojedinim potezima, drugi misle da su u pitanju bili pogrešni potezi računala i svašta nešto. Uglavnom; puno teorija, malo odgovora. "Naš" ljudski predstavnik je poslije izgubljenog meča ustvrdio da je najveća/jedina šansa njegovog povoljnog rezultata u otvaranju. Jednostavno je računalo predobro kad se situacija na ploči počne kristalizirati.
Ima li Lee šanse standardnom igrom doći do pobjede? Treba li možda eksperimentirati? Kakvom otvaranju će pristupiti u sljedećoj partiji?
Čeka se treći meč.
 |
| 16 L A T E S T R E P L I E S (Newest First) |
| dr strangelove |
Posted - 14/03/2016 : 14:26:19
Once again, Lee reviewed the previous game well into the night with other top professional players, looking for a chink in AlphaGo’s considerable armor. The game plan they came up appeared to be to try a type of ‘amashi strategy’, which is among the more extreme styles of play. To put it simply, it was close to being the opposite of Lee’s strategy from the previous day. In game three, Lee developed a large sphere of influence in the opening, provoking AlphaGo to dive in and bear the brunt of a severe attack. While this was a good strategy for Black, AlphaGo managed the position surprisingly well. So well, in fact, that it was quickly able to shift out of a defensive stance and counter-attack. In contrast, Lee’s strategy in game four was to take hard profit (territory) in the corners and on the sides of the board, allowing AlphaGo to develop influence over the center in compensation.
So why did Lee choose this plan? Well, we know from the first three games that AlphaGo is very good at estimating its probability of winning. It appears to be able to do this even more accurately than the best human players. With the help of this skill, AlphaGo seems to be able to manage risk more precisely than humans can, and is completely happy to accept losses as long as its probability of winning remains favorable. The Japanese have a name for this style of play, as it closely resembles the prevalent style of Japanese professionals over the previous few decades. They call it ‘souba‘ Go, which means something like ‘market price’. The essence of the idea is that you accept whatever seems to be the fair value of a position, rather than wagering the whole game on a single, complicated negotiation. This typically leads to a drawn out game where you seek to get the slightly better end of the deal in the majority of trades, and end up with a winning position in the endgame. As John Fairbairn has pithily put it, it’s like trying to win by arbitrage. Of course, stock traders don’t stand a chance of beating modern trading algorithms at their own game, so we shouldn’t expect Go players to do so either. What Lee and his friends had realized, was that they needed to completely upend the market.
With his cash firmly secured under his mattress, Lee invaded Black’s large sphere of influence deeply with White 40, brazenly daring AlphaGo to attack him. Continuing up to 48, Lee lightly scattered stones throughout AlphaGo’s potential territory at the top. He was not intending to save any particular stone. He only sought to flexibly establish a presence in this space and tank AlphaGo’s earlier investment. This came across as somewhat unreasonable, as AlphaGo had paid Lee good money for that potential! However, this was what Lee wanted — to force an all or nothing battle where AlphaGo’s accurate negotiating skills were largely irrelevant. AlphaGo didn’t have much choice in the matter. Either it could fight back and seek to gain compensation by attacking, or it could lose.
If AlphaGo had attacked any of Lee’s scattered stones directly, he would have been delighted. It would be easy for a player like Lee to dodge any straightforward attacks, and leave AlphaGo with relatively little to show for its efforts. A better strategy is to attack indirectly, threatening to surround all of the invading stones on a large scale and swallow them whole. This puts more pressure on the opponent to defend somehow and was exactly what AlphaGo did with the shoulder hit at Black 47. This move leaned against White’s stone at R11, while eyeing White’s stones at the top indirectly. It was similar to a strategy Lee had tried on the previous day. In the moves that followed, Black sought to get in front of White, preventing Lee’s group on the right side from connecting up with and rescuing its allies in the center. This tactic was a success for AlphaGo up to Black 67, at the cost of an acceptable four stone sacrifice on the right side. AlphaGo took the lead with the knight’s move at Black 69.
Rather than completing the transaction, by playing 70 at O6, Lee immediately reduced Black’s potential territory in the center with White 70. AlphaGo defended firmly with Black 71, which appeared to say “I’m winning,” but Lee probed its weaknesses fiercely from 72 to 76. Finally, as commentators were lamenting that the game seemed to be decided already, Lee unleashed a brilliant tesuji at White 78 – the only move that would keep him in the contention. AlphaGo failed to play the best response with Black 79, and its stocks suddenly crashed to pennies on the dollar. Tesuji are the close range tactics of Go. If you imagine Go as a mental martial art, tesuji are the techniques of hand to hand combat. They are clever moves which contribute to Go’s beauty as an artform. Strong players can usually spot a tesuji in the blink of an eye, through years of training and experience. So, I imagine, can AlphaGo’s policy network, otherwise it wouldn’t be able to play so well. But not all tesujis are equal. Some can be found by most players. Others are so rare, so exquisite, that even the majority of professionals don’t see them. The move Lee played was the latter kind. Most other professionals who were commenting the game live didn’t see the move. Gu Li 9p, a top Chinese professional who is Lee Sedol’s friend and fierce rival, was commenting on the game in China. Gu described White 78 as the “hand of god”, and claimed that he didn’t see the move coming. Neither, it would appear, did AlphaGo.
After 79, Black’s territory at the top collapsed in value. White’s invading stones had managed to escape through a hidden tunnel, and White 92 made miai of H14 and J10 (meaning White could play one or the other). This was when things got weird. From 87 to 101 AlphaGo made a series of very bad moves. We’ve talked about AlphaGo’s ‘bad’ moves in the discussion of previous games, but this was not the same. In previous games, AlphaGo played ‘bad’ (slack) moves when it was already ahead. Human observers criticized these moves because there seemed to be no reason to play slackly, but AlphaGo had already calculated that these moves would lead to a safe win. The bad moves AlphaGo played in game four were not at all like that. They were simply bad, and they ruined AlphaGo’s chances of recovering. They’re the kind of moves played by someone who forgets that their opponent also gets to respond with a move. Moves that trample over possibilities and damage one’s own position — achieving less than nothing. The game continued for another 80 or so moves, but it ended with AlphaGo’s eventual resignation. A jubilant Lee Sedol had scored his first win against the machine. This game was a masterpiece for Lee Sedol and will almost certainly become a famous game in the history of Go. After his brilliant move at 78, Lee’s play was perfect.
The question on many observers’ minds right now must be, what happened? AlphaGo’s lead developer, David Silver, attended the post-game press conference, but he didn’t explain the problem in detail. Silver and Hassabis said that they would have to wait until they returned to their office in the UK (after the match) to analyze the problem in detail. We, in turn, will have to wait until they do so before we find out what happened.
Can Lee Sedol do this again? Certainly it’s possible, but it may be very difficult. You need a special kind of position to pull this kind of stunt, and such situations don’t arise in every game. AlphaGo may also be more resistant to this problem than previous Go AIs, because it also uses its policy network to select moves. However, it must be pointed out that the policy network failed to properly score White 78 in the first place, probably because it was such an unusual move. We will have to see what happens on Tuesday when Lee plays the final game of the match. If anyone can repeat this feat, he can. |
| dr strangelove |
Posted - 14/03/2016 : 14:25:44
Sklepao sam nabrzaka jako zanimljiv razvoj četvrte partije od nekoliko različitih tekstova pa toplo preporučam svima da pročitaju, čak i onima koji ne znaju pravila. |
| Dranki |
Posted - 14/03/2016 : 00:24:53
http://www.theguardian.com/world/2016/mar/13/go-humans-lee-sedol-scores-first-victory-against-supercomputer
Interesantno je da čak ni tvorci AlphaGO ne mogu objasniti zašto je AlphaGO napravio pogrešan potez koji mu je osigurao poraz. Neki eksperti nagadjaju da se AlphaGO u medjuvremenu naučio i džentlmenskim manirima pa je, nakon što je sebi osigurao dovoljan broj bodova za ukupnu pobjedu, dozvolio Lee Sedolu da pobjedi kako bi izbjegao poniženje.
Neki ozbiljniji eksperti čak tvrde da je Lee pobjedio zato što AlphaGO nakon osigurane ukupne pobjede nije htio gubiti vrijeme na Lee Sedola, pa je učio šah, ili neku drugu igricu, dok se jadni Lee mučio smišljajući slijedeći potez. |
| Dranki |
Posted - 13/03/2016 : 18:21:38
Ljudska rasa uzvraća udarac:
https://www.rt.com/news/335411-champion-beats-computer-go/ |
| nagor |
Posted - 12/03/2016 : 18:55:35
Pobijedite ga:
http://www.cleverbot.com/ |
| Dranki |
Posted - 12/03/2016 : 18:24:43
http://www.spiegel.de/netzwelt/gadgets/alphago-sieg-wendepunkt-der-menschheitsgeschichte-a-1082001.html
Evo zašto je potpisana ona famozna peticija:
quote:
AlphaGo je učio ne samo iz odigranih partija koje su ljudi nekada igrali, nego je takodjer trenirao, neumorno i neprekidno, u partijama protiv samog sebe. Iz toga je očigledno proizašao jedan novi način igre, koji je trajno zbunio GO-elitu. AlphaGo je povlačio poteze, koje su ljudski profesionalci prvobitno smatrali za grešku. A radi se u stvari o tome da je mašina našla jedan novi način da pobjedi u igri, način do kojeg ljudi nisu došli tokom hiljada godina povijesti GO-a.
Znači mašine mogu naučiti da budu daleko inteligentnije od ljudi, i pitanje je šta onda.
Onako na prvu, moja pretpostavka je da će Amerikanci proglasiti Deepmind za prijetnju po američku nacionalnu sigurnost, i bombardovaće sjedište firme i sve firme u svijetu koje se bave razvojem AI. |
| Obi-wan |
Posted - 12/03/2016 : 12:36:16
quote:
Originally posted by Just_Charlie
Zar tu ne vrijede Asimovi zakoni robotike?
Trebamo li se bojati?
Naravno da ne.
Dok se stvori vojska robota, globalno zagrijavanje će nam već odavno doći glave.
Uglavnom, ja sam optimista... |
| nagor |
Posted - 11/03/2016 : 23:58:44
https://youtu.be/-iZyPamkEUs?t=39 |
| Dranki |
Posted - 11/03/2016 : 22:09:13
Autor teme je izostavio da spomene u čemu se zapravo sastoji senzacija u ovom dogadjaju, po uvodnom postu ispada da je samo još jednom neki računar pobjedio nekog čovjeka u nekoj tamo igri.
Senzacija se sastoji u tome što algoritam koji je igrao protiv šampiona nije bio preprogramiran za tu igricu, nego je učio igricu dok je igrao, prepoznajući i pamteći obrasce, te razvijajući strategiju za pobjedu u igri, odnosno primenjujući iste one principe po kojima funkcioniše ljudski mozak dok uči.
Senzacija je znači taj algoritam koji se kvantnim skokom približio vjernoj simulaciji ljudskog mozga, a na kojem radi stotinjak doktora za AI i neurologiju u Deepmind-u, koji je prije koju godinu kupljen od Google-a. Mislim da je osnivač te firme rekao da duboko vjeruje da se svi procesi u mozgu odvijaju po jednom "jednostavnom" algoritmu. Svi eksperti za AI pokušavaju da prokljuve taj algoritam. Tvorac C jezika je jednom prilikom rekao nešto što se može primjeniti i na ovaj slučaj:
quote:
UNIX is very simple, it just needs a genius to understand its simplicity.
Upravo taj mamutski napredak AI je ponukao najveće umove sadašnjice da nedavno potpišu peticiju protiv zloupotrebe AI, izmedju njih Stephen Hawking i Elon Musk. |
| Just_Charlie |
Posted - 11/03/2016 : 21:55:50
Zar tu ne vrijede Asimovi zakoni robotike?
Trebamo li se bojati? |
| Obi-wan |
Posted - 11/03/2016 : 21:51:35
Mislim da je najbolje da pozovu Bobu Fišera! |
| Mali_Mate |
Posted - 11/03/2016 : 21:33:18
quote:
Originally posted by KRAS
Jel to dobro il lose? Mislim jos uvijek postoji opcija-cekic..i razlupas racunalo ako se bude previse pametno pravilo...
nisi gledao Terminatore?  nema lupanja čekićem jednom kad se internetom raširi na milijune servera po svijetu, eventualno isključit kompletnu svjetsku struju, al tad smo ionako najebali  |
| KRAS |
Posted - 11/03/2016 : 20:34:34
Jel to dobro il lose? Mislim jos uvijek postoji opcija-cekic..i razlupas racunalo ako se bude previse pametno pravilo... |
| Sawaar |
Posted - 11/03/2016 : 20:27:48
quote:
Originally posted by izivko
Šta je ovo? Pomislio sam na šah a vidim nije?
Igra Go? Nikad čuo
Igra kojih 1000 godina starija od šaha, nastala u Kini. Jedna od najstarijih društvenih igara. |
| tihomirduras |
Posted - 11/03/2016 : 18:27:48
quote:
Originally posted by izivko
Šta je ovo? Pomislio sam na šah a vidim nije?
Igra Go? Nikad čuo
baš si me preduhitrio sa pitanjem, |
| izivko |
Posted - 11/03/2016 : 18:26:23
Šta je ovo? Pomislio sam na šah a vidim nije?
Igra Go? Nikad čuo |
|
|
