Thread

With most information hidden, the game Stratego had stumped AI until now

288 points · 149 comments · PaulHoule

dmurray · 1790968662 · focus · HN ↗
Oh no! Stratego had been on my mind as something we just hadn't tried hard enough to make a winning bot for, including the DeepMind effort from 2022. I was planning to make the first one.
I thought this was slightly less crank-coded than trying to prove the Riemann Hypothesis, but maybe these days you just ask Claude to do that and it tells you there's a counterexample at 1 + πi that no one ever noticed before.
1. NooneAtAll3 · 1790970493 · focus · HN ↗
  recently we got Starcraft 1 RL bot and Advance Wars bot, both on upper human level
  1. gregdeon · 1790970544 · focus · HN ↗
    We got an Advance Wars bot? Say more...
    zahlman · 1790984393 · focus · HN ↗
    Seems to be "JuggerMaster_BOT" on AWBW.
    NooneAtAll3 · 1790989777 · focus · HN ↗
    I only saw this vid <a href="https://www.youtube.com/watch?v=hdxAmJsspcA" rel="nofollow">https://www.youtube.com/watch?v=hdxAmJsspcA
    clickbait title and the bot isn't at "top players struggle to beat it" level - but it's no longer a casual walk in the park like the default ai is, so the progress is massive
  2. root_axis · 1790972246 · focus · HN ↗
    The frontier is moving, but there's still no human-level starcraft bot that plays entirely through vision, like a human.
    IMO it also needs to use a real mouse before I think it's a true comparison, even a casual player would have a massive advantage if they could issue selections and unit commands via query.
    xpct · 1790974567 · focus · HN ↗
    They already cap the actions per second a bot can take, sometimes to very low numbers. I don't remember if it was AlphaStar that limited how often you can move your camera or not, but some work definitely did.
    I personally don't see why vision is important, if anything I'd frame vision as useful to humans, rather than being the baseline.
    knollimar · 1790978470 · focus · HN ↗
    The capped actions cheat in EPM afaik. It does crazy micro
    dragontamer · 1790984128 · focus · HN ↗
    Your statement is severely underselling Plutos APM.
    Pluto uses Ghosts to counter Dragoons. Any SC:BW player should know how stupidly impossible this is for humans.
    For the non-players out there. Ghosts have an ability called Lockdown that can stop mechanical units (like Dragoons) from attacking or moving for long periods of time.
    The problem: it requires you to click on a Ghost, then click on a Dragoon.
    No human in the whole history of SC:BW would seriously do this. At best, Ghosts in practice are used vs capital ships (like Carriers or Battlecruisers), because in smaller numbers humans are fast enough to click and perform these feats. (Carriers and Battlecruisers are very expensive, very very powerful, units. So they are weak to specialized attacks like this as you cannot build many of them due to their incredible cost in resources).
    But dragoons? This is the mass production unit of Protoss. This is so far outside the scope of human-level APM that the games Pluto plays are useless. Lockdown simply isn't feasible to be used vs what amounts to the standard / stock infantry unit of Protoss. There's simply too many opponent dragoons to even click on.
    -------
    I dunno what kind of "APM rules" we need to give SC:BW bots to allow them to have a fair fight vs humans.
    But the rules we have in place today do NOT represent a fair fight at all. Computers have stupid amounts of speed and precision far in excess of a human player.
    boredhedgehog · 1791051161 · focus · HN ↗
    Sounds like artificial APM inflation coming back to bite the humans in the ass. If RTS players wouldn't mindlessly spam so many clicks, the target APM for the bot's limit would be lower. But as it stands, the bot just takes the inflated budget and puts it to better use.
    knollimar · 1791053768 · focus · HN ↗
    The big thing is it controls the units individually so it can do sick ghost micro
    dragontamer · 1791055286 · focus · HN ↗
    None of the pros are doing useless APM outside of maybe early-game hand-warmups (before there's anything to do, a lot of clicking is just to keep the hands warm and ready).
    Indeed, pros regularly go from 300apm average to like 800apm during critical fights, but only for a minute (maximum). Humans _can_ go well above the 300apm average and do so. (Indeed, keeping your hands warm and performing "useless" 300apm in the early game is critical to keeping the blood flowing and your fingers ready for the middle game).
    Nonetheless, the ghost micro that Pluto demonstrated is demonstrably superhuman. No one clicks that fast or precisely with the mouse.
    ---------------
    Indeed: the only way humans reach 300apm consistently is because of _keyboard_ shotcuts. Tapping buttons on a keyboard (which don't move and are always consistent) still counts as an "action".
    A "mouse action", which requires you to move the cursor, and then click accurately on a moving unit, is far more intensive than clicking "55" to recenter your map upon your starting Hatchery (assuming classic handstance: Zerg often put the starting hatchery on 5 and their middle or index-finger on the 5 so that their pinky can reach s and d, for "select larvae" then "drone").
    The keyboard shortcut of "55sd" is 3 actions, but far easier to perform than the 3 actions of "select ghost -> lockdown -> target".
    ---------
    I believe Slayer Boxer's famous ghost micro was accomplished with hotkeys 1234567890, so he could set off 10 lockdowns with 1l-click 2l-click 3l-click 4l-click 5l-click 6l-click 7l-click 8l-click 9l-click 0l-click. And that was a long time ago: it probably was a SlayerBoxer exhibition match (ie: he agreed with the opponent to go ghosts and the opponent to go battlecruisers before the match started, so that it was more of a crowd entertainment game rather than a serious tournament).
    This is the human limit of ghost micro, 10 precise lockdowns into a nuke. And only if _EVERYTHING_ was lined up for it and the opponent is even (mildly) working with you to try to make the most entertaining game.
    What Pluto is doing is so far outside of normal... that its nearly useless for us to study. I dunno, maybe some deeper analysts will look at some other tidbits of Pluto's build orders but I personally think its APM feats / clicking feats are just too far outside of human that there's no way anything else is worth studying.
    I think there's an argument that Pluto's TvT game is at least somewhat realistic and maybe humans can learn from it. But the jury is still out, people are still trying to analyze these games and understand the techniques, strategies, and tactics.
    grog454 · 1791071589 · focus · HN ↗
    > I personally don't see why vision is important
    There are at least 2 reasons vision is important in SC1
    1. Cloaked units are close to impossible for humans to notice when they are not consciously looking for them but once identified, its relatively easy to track and manipulate them indirectly (e.g. via body blocking). To be fair, I have no idea how the SC1 API works with cloak.
    2. Missing short blips of opponent unit movement on the minimap is common and sometimes detrimental enough to swing the game.
2. bananaflag · 1790971966 · focus · HN ↗
  I wonder why people act as if Riemann hypothesis is somehow already solved, when it is still highly probable that it is impossible for humans and slightly less impossible for AI overminds.
  (Of course, tomorrow Google might announce that it has solved it.)
  1. andrepd · 1790979044 · focus · HN ↗
    RH would require completely novel insight and a brilliant original thought. Unlike Navier-Stokes it's not a gradual effort and you cannot piggyback off of other mathematician's works, so "AI" will not solve it anytime soon.
    yorwba · 1790979473 · focus · HN ↗
    There has been a lot of intermediate progress on the Riemann hypothesis, for example ruling out particular kinds of zeros, and some of those partial results even involve LLMs <a href="https://www.anthropic.com/research/riemann-zeta" rel="nofollow">https://www.anthropic.com/research/riemann-zeta so I don't see why I can't possibly be a gradual effort.
    literalAardvark · 1791002357 · focus · HN ↗
    On what basis do you make these claims? Honest question.
    andrepd · 1791030136 · focus · HN ↗
    Which claims? The general feeling in the mathematical community was that significant progress towards Navier-Stokes was being made, and that the last "piece" to settle the question was imminent for some years now. Riemann Hypothesis seems virtually as unreachable today as it was 50 years ago.
3. Davidzheng · 1790999088 · focus · HN ↗
  I think now it's a good time for people to work on the Riemann hypothesis. There are actually a lot of ideas, no one knows if any of them goes anywhere--but I don't think it's completely impossible to make progress especially with new tools haha.

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.