hi all the go match was very eventful and turned out to have massive media coverage worldwide, with huge interest from technology publications, and theres a sizeable AI/ ML/ singulatarian crowd on the internet that follows these types of developments quite avidly.
wish that google would transcribe the press conferences. there was a lot of interesting details but it takes a long time to watch them and the midstream translation interruption (korean to english, english to korean) slows down the presentation also.
saw some interesting press questions in the post-3rd match conference iirc that tie in with some angles explored on this blog.
- there was a question about what are “bad moves”. hassabis seemed to concede the machine may have made a “bad move” after Sedols brilliant move that tilted the win in his favor. but how does hassabis know that? he knows that the probability of success by the machine dropped dramatically after making the move and Sedol made another. but the machine could not forecast Sedol would make the move that would plunge its win probability.
- again this seems hard to answer. even if there is a consensus by human expert commentators that the machine made a bad move, is that really provable? a postgame press query from audience was wondering whether bad moves are actually “unfathomable by humans”. hassabis nor some others do not seem to grasp the deep philosophical question of isolated bad moves.
- re bad moves, the commentators talk about “slack moves” that dont put much offense or defense into the game by the machine. but could these actually be regarded as “decoy moves” that draw the attn of the human away from areas the machine is winning? then what looks like a “bad move” may actually be a “decoy move”. again there is a lot of subtlety to this analysis. could there be any tendency by the machine to exploit human tendencies/ patterns/ repeated flaws found in the human game database?
- hassabis stated the machine hardware is distributed but not especially more powerful than the machine playing Hui because adding hardware did not seem to increase the win factor that much. he said the algorithm self-training via reinforcement learning significantly increased the win factor. but my question is, at what point did they decide after beating Hui that they should have a $1M grandmaster challenge? the announcement seemed quite soon. how much improvement had they measured in the algorithm at that point? what was their thinking/ analysis on how competitive the algorithm would be vs a grandmaster?
- a question asked about asymmetry due to alphaGo having Sedols games but not vice versa. hassabis stated the algorithm was mostly trained on amateur games over the internet and that there are not large enough databases of grandmaster games incl Sedols! this is an astonishing fact that the algorithm basically trained itself from an amateur level up to a grandmaster level.
- in some ways this feels like a once-in-a-decade type event and (contrary to singulatarian/ transhumanist hype/ froth) dont know how long the next such event might occur. again the closest event seems to be the Watson jeopardy win in 2011 (now ½ decade ago).
- the game is striking in how evenly the entire game is, and expert commentators seem to disagree for large stretches about who is winning. in one game, Sedol said in postgame analysis “at no time did I think I was winning” but commentator Redmond never sounded so skeptical. so it seems a huge element of the AI is just attempting to figure out if one has an advantage or not for long stretches. and it seems likely the AI will rate positions differently than humans because even humans cant come to a strong consensus among themselves! and if a machine rates a position better for itself than a human does, whos to say its wrong?
- wanted/ mostly predicted alphago would win but thought score might be closer say 3-2 and am surprised at its strength.
- there is an interesting angle in [a30] that finds that the chess engine Giraffe by Lai has strong similarities to AlphaGo, using neural networks to score positions and self-training to reach grandmaster level.[a32] Lai was hired by DeepMind! the article mentions the heavy nondisclosure agreements in use by Google. so we have real breakthru scientific advances here, but on other hand not necessarily open science. 😦
- SKorea coincidentally just announced a $1B govt research initiative into AI! [a31]
⭐ ⭐ ⭐
😮 👿 and now on to a vaguely connected topic. this blog has always been interested in so-called “paradigm shifts” as named by Kuhn. which evokes an old bumper sticker slogan, maybe originating from the 1960s:
subvert the dominant paradigm
in some ways all major paradigm shifts are “subversive”. a machine that just beat the worlds leading Go grandmaster literally subverted the dominant paradigm albeit in a mostly harmless way. we Americans find the word somewhat dramatic and flamboyant and its a quality routinely glamorized in our mass media/ entertainment. for example james bond 007 might be labelled as “subversive”. his own actions, and some of the non-PC elements of his movies, etc.; and in some ways all “paradigms” have a significant political component.
the matches were estimated to be seen by ~100M people worldwide, putting it on par with eg the US superbowl or soccer matches. so Sedol is a world class champion and Korean hero.
which brings me to a dark, creepy item that cannot leave alone, uncommented. (am struggling to talk about this in any way that does not sound like a non sequitur, and feel it disturbingly mars the prior celebratory tone, but feel it “must be said”.)
the south korea herald newspaper screenshot from last post has various items referring to NKorea and its dictator (“the one who shall remain nameless” for now)—and, specifically, threats of their nuclear weapons.
recently “right next door” in “next door neighbor” North Korea there was a huge/ mindblowing international incident. 21 year old US university of Virginia student Otto Warmbier was arrested by the North Korean govt for “subversion” for stealing a propaganda poster from his hotel in a restricted area. did Warmbier know the area was restricted? did he know the posters were highly guarded? [b]
talk about youthful hijinx vs clash of cultures or also from the casual vernacular
sometimes youre the windshield, sometimes youre the bug
he was just sentenced to 15years in a prison labor camp. shocking/ unbelievable. the crime was subversion against the state. the NKorean govt released blurry surveillance footage. in some mideast countries/ regions/ autocracies eg Saudi Arabia, Taliban, ISIS, the hands of young thieves (some teenagers… maybe even preteens? some of them stealing/ starving for food) are cut off. maybe one might say the NKorean govt is slightly )( more humane. (so now we live in that upside down world (somewhat forecast/ foretold by Orwell) where its hard to tell the difference between reality and an onion headline/ satire article! or where maybe reality is even more bizarre/ outlandish than the satire…)
now, am not an overly religious person, but think the only response possible to this at the moment is to “say a prayer for Warmbier”.
the US govt weakly protests the sentence as “unduly harsh”.
❓ ❗ ❓ WTF?!?
is that an understatement or what? actually it sounds like, to borrow a centuries old phrase, cruel and unusual punishment. “shocking, outrageous, horrific, nearly monstrous” would be nearly understatements also.
but we live in interesting times as the chinese curse goes. on the other hand (re the expression “people who live in glass houses shouldnt throw stones”) Guantanamo or Abu Ghraib seem to involve cruel and unusual punishment.
the entire NKorean govt, which apparently has even more political prisoners than in Russia (does post-USSR russia even have them any more?) seems to be based on the principle of cruel and unusual punishment.
am thanking my lucky stars at this moment that we dont live in a totalitarian system.
yes, have gotten into trouble myself in my life a few times in various memorable incidents for “subverting the dominant paradigm”.
all that pales mightily in comparison to Warmbiers “youthful indiscretion” turned living dystopian nightmare.
am thankful that can blog freely in cyberspace without fear of danger/ retaliation.
NKorea has also shown up in this blog recently in its coverage of cybersecurity issues and their apparent hack of Sony.
we live in politically tumultuous times and am thankful a peaceful match showing the international camraderie between cultures was possible in our at-times dark climate.
- 1. Artificial intelligence: Google’s AlphaGo beats Go master Lee Se-dol – BBC News
- 2. Go champion Lee Se-dol strikes back to beat Google’s DeepMind AI for first time | The Verge
- 3. Human Player Finally Beats Google’s AlphaGo AI Program At ‘Go’
- 4. Lee Sedol defeats AlphaGo in masterful comeback – Game 4
- 5. How Google’s AI Viewed the Move No Human Could Understand | WIRED
- 6. Go Grandmaster Lee Sedol Grabs Consolation Win Against Google’s AI | WIRED
- 7. Google’s Go Victory Is Just a Glimpse of How Powerful AI Will Be | WIRED
- 8. ‘Huge leap forward’: Computer that mimics human brain beats professional at game of Go | Science | AAAS
- 9. Google’s AI is the first to defeat a Go champion
- 10. Google’s AI Takes Historic Match Against Go Champ With Third Straight Win | WIRED
- 11. AlphaGo beats Lee Se-dol again to take Google DeepMind Challenge series | The Verge
- 12. Deep Blue developer speaks on how to beat Go and crack chess | The Verge
- 13. After AlphaGo, what’s next for AI? | The Verge
- 14. Why the Final Game Between AlphaGo and Lee Sedol Is Such a Big Deal for Humanity | WIRED
- 15. Google’s AlphaGo AI beats Lee Se-dol again to win Go series 4-1 | The Verge
- 16. Google artificial intelligence program beats S. Korean Go pro with 4-1 score | Reuters
- 17. Why AlphaGo Changes Everything
- 18. AlphaGo Beats Lee Sedol Final Score 4-1
- 19. AlphaGo Has Lost A Game – Score Stands At 3-1
- 20. Google’s AI Wins Fifth And Final Game Against Go Genius Lee Sedol | WIRED
- 21. AlphaGo’s battle with Lee Se-dol is something I’ll never forget | The Verge
- 22. In Two Moves, AlphaGo and Lee Sedol Redefined the Future | WIRED
- 23. What Google’s winning Go algorithm will do next : Nature News & Comment
- 24. Game-playing software holds lessons for neuroscience : Nature News & Comment
- 25. The Go Files: AI computer clinches victory against Go champion : Nature News & Comment
- 26. Digital intuition : Nature News & Comment
- 27. Google AlphaGo ‘can’t beat me’ says China Go grandmaster – Telegraph
- 28. The Go Files: AI computer wraps up 4-1 victory against human champion : Nature News & Comment
- 29. Why Neural Networks Look Set to Thrash the Best Human Go Players for the First Time
- 30. The Chess Engine that Died So AlphaGo Could Live | Motherboard
- 31. (LEAD) Gov’t to invest 1 tln won in artificial intelligence
- 32. Deep Learning Machine Teaches Itself Chess in 72 Hours, Plays at International Master Level
- 1. Moment Otto Warmbier steals propaganda poster bearing the name ‘Kim Jong Il’ from North Korean hotel | Daily Mail Online
- 2. Otto Warmbier: US student in North Korea dragged away by armed guards was travelling with group of young people ‘out drinking vodka till 5am’ | Asia | News | The Independent