[HN Gopher] A generalist AI agent for 3D virtual environments
___________________________________________________________________
A generalist AI agent for 3D virtual environments
Author : nuz
Score : 338 points
Date : 2024-03-13 15:22 UTC (7 hours ago)
(HTM) web link (deepmind.google)
(TXT) w3m dump (deepmind.google)
| SomeCooeyGuy wrote:
| Why link to the twitter post and not the article linked from
| within it?
|
| https://deepmind.google/discover/blog/sima-generalist-ai-age...
| GaggiX wrote:
| @dang will help
| pvg wrote:
| @dang doesn't do anything, you should email
| hn@ycombinator.com
| LeifCarrotson wrote:
| To be clear, @dang does a lot! His moderation is a big part
| of why HN is such a useful forum. But you're right that the
| text '@dang' being posted in an HN comment does not result
| in the software sending a notification to him.
| GaggiX wrote:
| I have always seen people using @dang and honestly it has
| always worked, he even thanks me sometimes.
| pvg wrote:
| There are regular comments by him, every few days, asking
| people not to rely on @dang.
| https://news.ycombinator.com/item?id=39662522
| dang wrote:
| That's random. If you want better-than-random message
| delivery you need hn@ycombinator.com.
| modeless wrote:
| I submitted that one earlier:
| https://news.ycombinator.com/item?id=39691783
|
| This one got points slightly more quickly for whatever reason.
| Usually in these cases of duplicate submissions dang will merge
| the stories together.
|
| To the DeepMind team I would say that a snappy summary
| accompanied by a video at the very top of the page instead of a
| bunch of whitespace and a static image you have to scroll past
| would likely help the blog post be more viral, as this tweet
| demonstrates.
| pvg wrote:
| _I submitted that one earlier_
|
| With a made up title which you should avoid, it's in the
| thing https://news.ycombinator.com/newsguidelines.html
| modeless wrote:
| Generally I editorialize the title when I make submissions
| in the "new" queue if the actual title is very dry. If I
| had posted the actual title of the blog post it would be
| even farther behind the tweet version. I may update the
| title now that it's on the front page, unless dang does it
| first.
| pvg wrote:
| Right, except the guidelines ask you to not-do exactly
| that and you shouldn't.
| modeless wrote:
| The guidelines are not all hard and fast rules. Dang has
| explained in comment threads (that I don't have a link to
| right now) that editorializing titles is actually OK in
| some situations. Obviously the title should not be false
| or clickbait, which my title was not.
| pvg wrote:
| No, it's the other way round. It's only ok to change it
| if it's misleading or clickbait, otherwise use original.
| You can't editorialize titles if they are _not_ clickbait
| or misleading and the original wasn 't.
|
| _please use the original title, unless it is misleading
| or linkbait; don 't editorialize._ There are regular
| moderation comments about it, they just say the opposite
| of what you think they say
|
| https://hn.algolia.com/?dateRange=all&page=0&prefix=true&
| que...
| modeless wrote:
| Again, this guideline is not a hard and fast rule. Those
| moderation comments you link to are for the very common
| case of posters adding their own strong opinions to
| article titles. The title I chose had no opinion and was
| merely a factual description of the article content. I'm
| afraid I won't be able to find dang's explanation of why
| it's sometimes OK to use different titles because it's
| drowned in the sea of moderation comments for egregious
| clickbait and highly opinionated submission titles.
| pvg wrote:
| No, editorializing is adding your opinion about what's
| important. You don't get to do that in a title, you can
| do that in a comment. That's exactly what those comments
| say. This particular thing is a pretty settled practice.
| Notice how the title was fixed. Because it was a bad
| title.
| modeless wrote:
| I changed it myself. Once the story is out of the "new"
| queue, having a snappy title is less important. I still
| did not use the exact article title as I think the
| content is better represented this way.
| pvg wrote:
| Adding 'snappy titles' to boost your submissions is
| exactly what the guideline asks you not to do. I'm not
| sure how it can be more straightforward or less prone to
| your misinterpretation.
| shadowgovt wrote:
| "Please don't complain that a submission is
| inappropriate... Don't feed egregious comments by
| replying; flag them instead."
| https://news.ycombinator.com/newsguidelines.html
| pvg wrote:
| This is neither about the appropriateness of the
| submission nor are any of the comments egregious.
| dang wrote:
| I don't want to pile on in this subthread, just want to say
| that the reason I changed the URL on the current thread
| rather than re-upping your post (as we normally would when
| the better URL was posted by someone else earlier) is that it
| was the OP's first submission to HN and I like to encourage
| newer participants.
| modeless wrote:
| Oh, that's interesting! Thanks, I'm happy to let someone
| else have the karma. I think in this case the tweet linked
| by OP was genuinely a good submission because it got to the
| point much faster than the blog post and directly linked to
| the blog post for people interested in the details.
| gitfan86 wrote:
| The deep mind team chose to post it to X because it allows
| vitality and X is also the town square now
| SomeCooeyGuy wrote:
| that wasn't the question, but thanks
| dang wrote:
| Thanks! Changed to that from
| https://twitter.com/GoogleDeepMind/status/176791851558599481...
| now.
| dtx1 wrote:
| Okay, so now i know how artists feel: AI is destroying that thing
| i love, now i will always have to fight ais when playing online
| mrnotcrazy wrote:
| I think the better deployment of this would be 1 human and 4 AI
| buddies so you could have a more tactical experience and games
| could have larger more realistic battles.
|
| I think strategy might be more important in the future than
| switchy aiming
| giovannibonetti wrote:
| Don't you already fight AIs when playing in single-player mode?
| jprete wrote:
| Single-player AIs are fine-tuned by the developers for fun
| instead of frustration. I think the GP is talking about
| undetectable online cheating.
| educaysean wrote:
| Maybe the future is that we no longer have to worry about
| players cheating online because we simply won't play random
| strangers online. You can still team up or play against
| friends, but all other "players" can be bots with varying
| levels of skills and play styles. Cheating solved.
| Janicc wrote:
| Most likely you'll be playing alongside or against AIs without
| ever knowing. Either AIs from the game developers to make their
| online game seem more popular or people who want their stats
| leveled up without any effort on their part.
| globular-toast wrote:
| Back to LAN parties then!
| cboswel1 wrote:
| Hey, maybe the bot creator will be kind enough to add a racist
| chat integration with the agent so you'll feel right at home in
| the COD lobby.
| jldugger wrote:
| Man, its not even science fiction anymore to speculate that the
| robot apocalypse happens because "exciting" violent games were
| far cheaper and more plentiful than boring real life simulators.
| seydor wrote:
| I wonder if Deepmind's quest for AGI using games as testbed will
| translate in the real world. It is quite possible that many of
| those feats rely on the ANNs learning the physics engine of the
| game. Which as impressive as it is, it 's probably nowhere near
| as complicated as the real world out there. The last mile may
| prove to be a very hard problem to solve, and i wonder if they
| have an alternative strategy to it.
| bigyikes wrote:
| Are real world physics really that much more complicated than
| video game physics? Sure, if you're talking about E&M,
| relativity, or QFT, the real world is more complicated, but for
| every day mechanics, they seem pretty comparable?
|
| Video games might be even harder to predict sometimes, since
| the physics simulations have very strange edge cases. There are
| no physics glitches in the real world.
| seydor wrote:
| i believe it is as hard as it is for robots to act exactly as
| they do during simulation
| Jensson wrote:
| > Are real world physics really that much more complicated
| than video game physics?
|
| Massively, in a game running into a wall is perfectly normal
| and valid strategy to get close to it, in reality that will
| wreck you.
|
| Or running on a fist sized rock has no consequences, in
| reality that destroys your foot. Reality is full of such
| extreme threats everywhere even in normal homes.
| tjah1087 wrote:
| SIMA author here - SIMA is betting on simulations and games. We
| use real-world open-ended language, but games/sim obviously
| have simplified physics/graphics.
|
| In terms of alternative strategies, Google DeepMind also has an
| amazing robotics team with lots of fantastic work for real-
| world robotics - including multi-robot generalists, showing
| positive effects when co-training one agent or model on
| multiple environments/bodies. Their prior work was very
| inspirational to us in SIMA!
| https://deepmind.google/discover/blog/scaling-up-learning-ac...
| robrenaud wrote:
| Simulation to reality is a huge bottleneck in a lot of
| reinforcement learning work. Reality is just super messy and
| complicated.
|
| Tesla has an alternative. If you can get your devices
| widespread and can be recording observations and actions, you
| can collect huge datasets in the real world.
|
| The first embodied, vaguely general, multi-task, useful and
| economical robots might just really open up the virtuous cycle
| of experience, learning, feedback and improvement. If I had to
| guess where it would come from right now, I'd pick Amazon
| warehouses.
| leetharris wrote:
| I will never forget being at TI7 where OpenAI revealed an AI that
| could take on pro Dota players. Dota is an insanely complicated
| and difficult game. This was an eye opening moment for me that
| led to a career shift.
|
| https://en.wikipedia.org/wiki/OpenAI_Five
| jerrygenser wrote:
| Upvoted. Not sure why this got downvotes. It's very cool that
| you were at TI7 as I only watched this on youtube. I also
| thought this was an important moment.
| jprete wrote:
| I know this might be perceived as an obvious question, but what
| exactly did you switch from, and to?
| qvrjuec wrote:
| He was a professional gamer, and switched to biodynamic
| agriculture after seeing how little hope humanity had playing
| games against machines
| WJW wrote:
| Meh. People still do athletics competitions even though
| cars exist and can outpace any human. Weightlifting is also
| still a thing even though even an entry-level forklift
| beats any human weightlifter. Chess is more popular than
| ever before, even though nobody has any hope of beating a
| computer anymore.
|
| Out of all the fields that human do professionally, sports
| will be one of the last ones to disappear. The fact that it
| is (unaugmented) humans competing is the entire point.
| tsumnia wrote:
| > Out of all the fields that human do professionally,
| sports will be one of the last ones to disappear. The
| fact that it is (unaugmented) humans competing is the
| entire point.
|
| This is my thought/hope for what we'll expect in the
| coming years as AI's automation becomes more commonplace.
| Society's interests will start going towards activities
| that showcase human ability - sports, livestreaming (very
| much its own industry now, but mostly for socializing,
| art, and gaming), performance, dance, etc. Sure AI can
| 'do' these things, but not at the level elite performers
| can or with the subtle nuisances in human personalities.
| OkayPhysicist wrote:
| I don't think the unaugmented qualifier is accurate. What
| matters is that there are well-established rules defining
| scope. People racing cars is still a very widely enjoyed
| form of entertainment.
| FrustratedMonky wrote:
| LOL. Yes, exactly.
|
| Even in the future when the AI is provide everything and
| we are no longer able to understand it, humans will be
| doing human competitions, playing chess, etc... The human
| on human action will be only thing left, and only thing
| humans care about. Chess is already unwinnable, but
| humans still want to measure themselves against other
| humans.
|
| Chess, Go, what next? Pizza delivery? Accountant
| Simulator? Humans are already being outclassed one
| feature at a time.
| leetharris wrote:
| A slow transition (over 5+ years) from web/desktop
| development into HPC and AI/ML.
|
| I still consider myself a 4/10 at best compared to my amazing
| peers who studied this from the start, but you have to start
| somewhere!
| calderarrow wrote:
| If you don't mind sharing even more, what did you do to
| learn HPC/AI/ML? Any suggestions for getting started?
| Agingcoder wrote:
| https://en.algorithmica.org/hpc/
| Jwsonic wrote:
| I'd love to hear more about your approach to the shift. How
| did you pitch yourself during interviews without prior
| experience?
| alumic wrote:
| I'd echo the same sentiment as the other commenters, if you
| don't mind me throwing my hat into the ring. Considering a
| MS in Data Science with a focus on ML
| nsypteras wrote:
| So glad you said this! For some reason that's always stuck out
| to me as having been my biggest personal "wow" moment while
| watching AI development progress. ChatGPT is awesome but for
| some reason I've never felt as awed by it.
| sanxiyn wrote:
| I too was impressed at first, but got disillusioned after
| learning how they did it. It was much more similar to chess
| AI with piece-square tables than I first thought.
|
| https://gist.github.com/dfarhi/66ec9d760ae0c49a5c492c9fae939.
| ..
| adtac wrote:
| If Dota was twice as complex, do you think an AI would be more
| than 2x better or less compared to your scenario?
|
| I suspect the more complex the game, the bigger the advantage
| over humans.
| mminer237 wrote:
| A big advantage of AIs is instant reaction time. OpenAI
| programmed in an artificial reaction delay to most skills,
| but they were still generally much faster than any human
| would be. Overall strategy is where the AI was lacking, but
| its technical flawlessness makes up for it.
| adtac wrote:
| If we model the game as someone flicking switches, strategy
| is ability to know which switches to flick when whereas
| technical skill is the ability to quickly and precisely
| flick the chosen switches.
|
| In more complex games, there are more switches and the
| current set of best switches changes faster. With more
| switches, it's harder to know which are the best switches
| because the future is less predictable. And even if we
| figure out the best ones, they might change before we flick
| them. And even if we get around to it in time, we might fat
| finger it and accidentally flick an adjacent switch. And
| our opponent never gets tired or injured.
|
| This is why I suspect AIs have a much higher ceiling even
| if we limit them to half the APM pros have. Better strategy
| matters less, but I admit it's our only chance lol.
|
| FWIW, I've never played Dota but I've played a lot of AoE2
| and from what I know they're similar enough (but maybe
| someone can correct me).
| CuriouslyC wrote:
| If you're talking immediate mechanical complexity, then yes.
| If you're talking delayed strategic complexity, then
| definitely not.
| jsheard wrote:
| As I explained in my sibling comment the version of Dota they
| played was heavily simplified, because the full combinatorial
| explosion of mechanics was far too much for the AI training
| to overcome. They didn't even get close to playing normal
| Dota at a high level, nevermind a hypothetical version of
| Dota which is twice as complex.
| everforward wrote:
| You would need to define what kind of complexity.
|
| I would broadly break it into things that are complex to
| perform (crazy APMs or accuracy), things that are complex to
| understand (the stack or layers in MTG), and things that are
| complex to predict (e.g. time-delayed abilities and the
| correct time to use them, like Baptiste's lamp in Overwatch).
|
| AI have basically constant performance across a performance
| complexity curve, because the complexity typically derives
| from physical interfaces the AI doesn't use anyways. E.g.
| their APM is not limited by how fast their fingers can
| physically move.
|
| AIs do very poorly on tasks that are complex to understand.
| The best Magic: The Gathering AI's I've seen are awful
| (though also likely far less well-funded). Best-case scenario
| is basically an AI who makes plays that don't make any sense,
| but are at least valid plays. It's a crazy difficult problem.
| E.g. there are various ways to make infinite mana with
| combinations of cards, and the AI needs to a) realize that it
| can use those cards in order to create infinite mana, and b)
| that it can do this multiple times (I.e. it can pay for a
| spell that costs more mana the loop generates by going
| through the loop multiple times). That's very hard thing to
| do; human players somewhat frequently don't realize when they
| have loops.
|
| Add on top of that that a game of Magic can enter a state
| where a loop of effects becomes recursive but doesn't result
| in either player winning. The game is a draw, because it
| cannot progress anymore. Detecting these can be non-trivial,
| because they might involve side effects that look like
| someone should win (I.e. you lose a life and I gain one, then
| I gain 2 life, then you deal 1 damage to me, then I gain 1
| life, then you deal 2 damage to me. Life totals shift around,
| but net to 0 by the end of the loop).
|
| I think the AI do well at complex prediction tasks as well,
| by nature of their response times and access to prior
| information. I would expect an AI to beat humans by a wider
| margin the more complex the prediction gets. Humans have
| finite time and thus experience; the AI is going to have more
| "experience", and be able to recall it at a faster rate.
| jsheard wrote:
| The caveat being that the scope of the game was _significantly_
| pared down for the sake of the AI. Specifically the team
| compositions were pre-determined, meaning the AI only had to
| understand 10 heroes in two specific arrangements of 5, when
| there 's normally >100* heroes which can be chosen in any
| permutation, and certain game mechanics were also declared off-
| limits for the human players because the AI wasn't able to
| understand them. Beating pros in that subset of the game was an
| impressive achievement but there was a _huge_ gulf between what
| they did and doing it in the full version of the game, which
| they quickly gave up on trying to do after collecting their
| marketing trophy of beating Dota pros in something resembling
| Dota.
|
| * I'm not sure exactly how many heroes there were at the time,
| it was less than the 124 there are today, but it was certainly
| a lot more than 10.
| avree wrote:
| There were 112 heroes available at that time. It's also worth
| noting that two of the heros chosen for OpenAI to use, Viper
| and Sniper, are considered some of the mechanically 'easier'
| heroes, as they rely primarily on autoattacks to do damage,
| as oppose to decision-making around when to use spells.
| Crystal Maiden, Lich, and Necrophos, the other 3 of the 5
| OpenAI heros, are similarly considered 'easier' as they have
| spammable, very forgiving abilities that can be used almost
| indiscriminately.
| FrustratedMonky wrote:
| baby steps. Still impressive.
|
| AlphaGo was strait up same.
|
| AlhaStar did have some limits placed to narrow it down for
| the AI. But was still imperfect information, and wildly
| complicated.
|
| And those were all 3+ years ago.
|
| Games are a lower resolution representation of the 'real'
| world. And we haven't seen any slowing down of AI scaling
| up for more and more complex world views.
|
| Eventually the 'map' will be the 'real', as real as the
| human brains internal map of reality.
| Jensson wrote:
| > And we haven't seen any slowing down of AI scaling up
| for more and more complex world views.
|
| We absolutely have. We have superhuman performance on Go,
| we have human expect level performance at Starcraft, and
| now we get human baby level performance at 3d games. The
| more complex the task/game the worse the AI gets relative
| humans it seems, I don't see how this shows the AI
| scaling up, to me this is all moving horizontally.
| msp26 wrote:
| One of the constraints ended up being implemented into the
| main game: separate couriers for each player. But generally,
| agree with your point.
|
| But it's very cool how the OpenAI matches ended up making mid
| players reevaluate how they used consumable regen.
| nindalf wrote:
| Explanation - health potions cost a small amount of in game
| money and have to be ferried by a courier to the player.
| Most pros (and good players copying the pros) didn't do
| this because it wasn't considered cost effective. They
| would rather save up for a larger purchase. Until they
| repeatedly lost to OpenAI bots spending absurd amounts of
| money on health potions.
|
| The AI didn't follow "best practice" because it wasn't
| trained on human games, found a better way and that was
| quickly adopted by all, becoming the new best practice.
| murderfs wrote:
| League of Legends players discovered this like 15 years
| ago (the "13 health pot start"), I wonder why this didn't
| cross over. I suppose the player bases don't actually
| intersect very much?
| evandale wrote:
| I think maybe the games are a bit different and it wasn't
| viable? I was pretty into the original WC3 Dota and
| starting with tangos for healing was a pretty popular
| strategy for supports and solo lane players.
|
| caveat: my Dota 2 knowledge is lacking because I haven't
| followed the game for about a decade now and I have
| essentially 0 experience with League.
| cptn_badass wrote:
| It's mostly because it's was a different scarce resource
| at that time that was seen as non-optimal use by the
| players, the courier. It can ferry item to you, in a
| normal game there was only one of them for your whole
| team, which mean using it would take that ability away
| for your teammate during the ferry time.
|
| One constraint to those showmatches at the time was that
| every heroes had their own courier, and player at that
| point were not accustomed to using it for "low value"
| travel, unlike the AI that was using it liberally.
|
| In a later patch, the 1 courier per hero feature was
| added, and now pro players are much better at managing
| it, but at that time it was truly a heavy opportunity
| cost.
| Etherlord87 wrote:
| Very interesting! Shows how much AI is over-hyped (even
| though, as you say, it was very impressive anyway). It was
| even worse in case of Starcraft 2, where the AI had a much
| wider view than humans, and while the AI was supposed to show
| its strategic superiority, by limiting the APM (actions per
| minute), the limit was still very high, inspired by the max
| APM achieved by humans - whereas this max is achieved only
| for a very short period of time (a single minute), and
| consists mostly of insignificant click spam (had APM been
| limited to half that for humans, the effect would probably be
| negligible, and very minor for a quarter...). So as a result
| the AI would win by being able to micro-manage more units,
| rather than having a better strategy. But again, it was very
| impressive anyway.
| FrustratedMonky wrote:
| How do you have 'over-hyped' and 'very impressive' in same
| sentence. Which is it?
|
| I think you are not giving AlphaStar the correct spin.
|
| They came back and changed it to only have the same
| viewport as the human, it could not see all of its units
| simultaneously, it had to move the cameras like a human.
|
| BUT importantly, it NEVER had perfect information. It could
| only see exactly the same as the human, just at one point
| they were letting it see the whole map without changing the
| camera, but it still could not see enemy units without
| sending a probe.
|
| And. Little unsure on what the argument about APM is
| saying. It was slowed down to match the human speed, but
| somehow that makes it less impressive? That is just making
| it more 'human-like'. Kind of like people today want to put
| guardrails on AI, but if it was unleashed, it beats them
| easily. That isn't a knock on the AI. The AI would still
| have to think about every move, and form a strategy. They
| slowed it down to human level inputs, handicapped it, to
| make it playable to a human. But to your point, if AI could
| make 400 APM and human had 400 APM (both limited to same),
| then that is better measure about the 'thought' behind each
| individual move.
|
| I still remember watching one match where the human was
| winning, the AI was down, and the AI really did fight back
| very aggressively from a loosing position, like a human. by
| expanding and adapting, and it looked very scary.
| Etherlord87 wrote:
| > How do you have 'over-hyped' and 'very impressive' in
| same sentence. Which is it?
|
| I'm stunned; how would you think they are contradictory?
| Imagine a transportation that moves with the speed of
| 1000 km/h. Very impressive, right? Now imagine media
| everywhere say it moves with the speed of light. Wouldn't
| this be over-hyping?
|
| > BUT importantly, it NEVER had perfect information. It
| could only see exactly the same as the human
|
| Maybe we're speaking about different events... In the one
| I'm commenting on, the AI had some zoom-out, I think 2x
| (meaning it would see 4 times more at once). Yes it had
| fog of war, but a zoom out like this is a very
| significant advantage.
|
| > And. Little unsure on what the argument about APM is
| saying. It was slowed down to match the human speed,
|
| No it wasn't, not exactly. Imagine that you measure a
| human racer speed in km/minute, every minute. Then you
| take the highest measured "average per minute", and
| program AI to move with that speed at all times. Then you
| praise AI for its pathfinding algorithm, because using
| that speed, it beats the human racers.
|
| Yes, if a human racer has to slow down, because e.g. the
| human is unable to avoid obstacles at maximum speed, it
| does make the AI being able to move faster, impressive.
| But few people here would be impressed by a high reflex
| of a computer, because we all are used to the fact
| computers can react much faster than humans. It is
| misleading, however, to allow AI to move faster, and then
| give it the "spin", as you say, that the AI has won
| because it was smart, as opposed to being fast.
|
| BTW, I think the AI was either only using one race, or
| was playing only against one race. This one thing was
| actually mentioned in the event (once). The APM was
| mentioned too, I think, but the nuance I describe
| unfortunately wasn't mentioned.
|
| It makes me sad, because as I said, it is a very
| impressive technology. But it's hard to fully appreciate
| something when it is so blatantly over-hyped and when you
| see so many people around you being mislead and praising
| AI for achievements that it didn't exactly accomplish.
| chse_cake wrote:
| That's a fair assessment but it was also 6 years ago. Back
| then transformers had recently come out. Tricks to do DL at
| scale were still brewing. Even achieving what they did for 10
| heroes showed DL could work in "non-deterministic-ish"
| problem settings.
|
| The more interesting question is: can we train a Dota model
| that plays with all 124 heroes today?
| iamdelirium wrote:
| Not really. By only playing a certain subset of the game, the
| AI could use heroes that they were good at (micro intensive)
| while disallowing heroes that could counter the strategy the
| Open AI chose. Hardly a fair game.
| pdimitar wrote:
| How did this change your career?
| bitcharmer wrote:
| What is TI7?
| tsumnia wrote:
| Dota's yearly competition, TI7 means "The International 2017"
| pests wrote:
| That's confusing.
| injuly wrote:
| Oh, that was wild.
|
| Though to be fair, the human players had to rely on muscle
| memory to win lanes (CSing, blocking waves, pulling, trading
| hits, cutting waves, stacking, etc.); whereas the AI could
| perfect the timings down to the fraction of a millisecond.
| ballenf wrote:
| Was there any control input limit on the AIs? Like if the AI
| couldn't click buttons or move the mouse faster than a very
| fast human.
|
| In a similar vein, it would be fascinating if the AI had to
| also evade bot detection, that is appear (nearly)
| indistinguishable from a human player.
| kelvie wrote:
| If I remember right there was a reaction time for the
| OpenAI team they could tweak, if I remember right it was
| around 200ms (and a short search I think confirms that).
| Jensson wrote:
| The bot did perfectly dodge a skill that humans almost
| never dodged because it didn't have a good visual cue.
| Against the AI that skill became useless, really screwed
| the humans over and made it clear to everyone that the AI
| didn't really play with the same limitations as humans.
|
| In the next game it played they had made it react even
| slower and then it no longer beat tournament teams.
| blharr wrote:
| Somewhat related, AlphaStar has such a control input limit
| for Starcraft limiting the actions per minute by some
| amount.
| garyiskidding wrote:
| Greg Brockman from OpenAI inserted the program drive for a game
| of Dota vs Dendi (popular Dota 2 player). Really memorable.
| swolpatrol wrote:
| funny, I was there and got my only openAI shirt. I even
| competed againts the bot in the 1v1 and won. best memories from
| Seattle
| 12345hn6789 wrote:
| I don't quite think it's that impressive. AIs in video games
| are specifically "nerfed" simply due to the fact they can make
| decisions much, much faster than humans. Open AI didn't do
| anything special in this case.
|
| See @deep blue for more. Or, any strategy game made in the last
| 20 years with AI difficulty mods.
| 7734128 wrote:
| AI in most strategic games are provided massive advantage
| over human players. In CIV V, for example, the AIs start with
| several units and techs on higher difficulties.
| nurettin wrote:
| In short, the game was dumbed down and computer programs can
| micro perfectly, so the players had no edge.
| sdrg822 wrote:
| Dang they use Transformer-XL from 2019 haha - didn't realize
| people still used that / XLNet-like architectures
| ado__dev wrote:
| I hope devs are able to use this to give more life to NPCs. So
| many times we've been promised NPCs in RPGs that have their own
| lives, do their own things independent of the player, etc. and
| it's never really materialized into anything notable. With AI
| though, I feel like we may be getting close.
| digging wrote:
| I'm suspicious of whether that will actually make games more
| fun or interesting.
|
| For example, the more realistic human character animations get,
| the deeper we seem to fall into the uncanny valley. The fact is
| that humans themselves naturally move in ways that look weird
| when put on a stage, so mocap tends to look sillier the better
| it is. Which is why we have theater, where movements are
| exaggerated.
|
| Anyway, with AI characters, I expect it will actually be more
| frustrating and boring than not if they have realistic lives
| and schedules. All the littler irritations that we deal with
| and accept form real people just become friction in a game.
| Games, as movies and books and shows and plays and
| illustrations, don't need to be more realistic to be better.
| Media is caricatures of real life, with important information
| _intentionally presented_ to give us a good experience. Taking
| _inspiration_ from real life can give us better mechanics but
| blindly mimicking real life will give us shitty games.
| timlod wrote:
| Well, weird movements in games should be a thing of the past
| in the near future, as we can begin to extract motion capture
| data from videos of normal people acting normally.
|
| I think it depends on the type of game you have, but I
| wouldn't underestimate this type of technology for say, open
| world games where it might make the game more immersive due
| to convincing realism.
| digging wrote:
| > Well, weird movements in games should be a thing of the
| past in the near future, as we can begin to extract motion
| capture data from videos of normal people acting normally.
|
| I think you misread my posts. We don't have awkward
| animations because our mocap isn't good enough, we have
| awkward animations because typical human motion looks
| awkward - our brains just mostly ignore that.
|
| People are awkward; we don't actually want characters in
| games/movies/etc to be like real people. Very few movies,
| for example, would be well served by conversations
| frequently and for non-plot-related reasons being
| interrupted by loud noises, having people talk over each
| other and nonverbally try to figure out who gets to speak,
| having characters ask "What?" and then begin to reply
| without waiting for the answer because their brain caught
| up half a second later, etc.
| ryandrake wrote:
| > Anyway, with AI characters, I expect it will actually be
| more frustrating and boring than not if they have realistic
| lives and schedules. All the littler irritations that we deal
| with and accept form real people just become friction in a
| game.
|
| I remember Ultima V in the 80s had this. NPCs had their own
| daily routine and went here and there throughout the day. A
| couple of side quests relied on this mechanic--you had to
| learn their schedule and catch them somewhere at some time.
| hbn wrote:
| Majora's Mask did something like that too. The game was
| centered around a 3 day cycle you had to do things in
| because at the end the moon would crash into the world and
| everyone dies. So you'd get stuff done, then time travel
| back to the beginning of the cycle repeatedly. NPCs all did
| the same things in those 3 days, so you could help people
| out with their sidequests in each cycle, like in Groundhog
| Day.
| Legend2440 wrote:
| Everyone's a skeptic these days.
|
| I'm pretty confident that people will find ways to make fun
| and interesting games using them, just like they have with
| every other computer technology over the last 40 years.
| digging wrote:
| > Everyone's a skeptic these days.
|
| You'd be surprised at how untrue that is, unfortunately.
|
| > I'm pretty confident that people will find ways to make
| fun and interesting games using them
|
| I agree - but also, that's a different phenomenon than
| simply inserting AI naturalistic characters and expecting
| them to be fun to interact with.
| ca_tech wrote:
| There is a delicate balance though. The draw of the RPG is that
| you play a character who usually has an outsized impact on the
| world in comparison to most everyone else. If we expect the
| NPCs to operate as independent entities, how do we manage their
| ambition so that they don't individually, or in aggregate,
| impact the world more than you? Without that, your character
| becomes another cog in the machine; it's their world and you
| are simply living in it... Which I could see having a certain
| appeal and may be just another genre.
| rebuilder wrote:
| Sandbox games are pretty big. Not my thing at all, but I'm
| sure there'd be a big market for believable open-world RPGs
| for example. They'd probably be more simulators than
| narrative-driven games.
| russfink wrote:
| Maybe give them a "Robocop Directive 4 [HIDDEN]" that
| prevents significant actions...?
| sfjailbird wrote:
| That sounds really cool, as a genre, actually. You would have
| to figure out what given NPCs actually want, and how you may
| be able to use that, to make them do what _you_ want. And not
| based on some lame script, but complex world and character
| models. I 'm sold!
| Tossrock wrote:
| At a certain point, you're just reinventing real life with
| worse graphics.
| Jensson wrote:
| Real life where you are the chosen one with super powers.
| silveraxe93 wrote:
| Real life with worse graphics +
|
| - Ability to reload when you fail
|
| - You can choose your gender
|
| - Actually, be anything. A dwarf, elf, dog, tall, short,
| muscled, etc...
|
| - Novel physics (magic)
|
| - Sooooo much more
| PartiallyTyped wrote:
| What makes cyberpunk 2077 such a great game is that the world
| doesn't revolve around you, it exists despite you, and you
| see this in all of the side-quest events that are happening
| in the background. The world being alive despite you, and not
| for you adds a whole new dimension to the universe.
| summerlight wrote:
| The problem here with highly capable agents is that it's not
| predictable and controllable in the point of designer's view.
| That might work for certain types of games, but many cases
| designers want to have a certain degree of control on their
| games.
| Legend2440 wrote:
| Then we'll make new kinds of games where the unpredictability
| of the NPCs is a core mechanic.
|
| Seriously, have you no imagination? Why sit around coming up
| with reasons it won't work instead of ways to make it fun?
| TillE wrote:
| Nobody actually wants to read LLM-generated soup. That's
| just a waste of everyone's time.
|
| If you're making a highly interactive, dynamic game, you
| don't even need detailed language for NPC interaction. You
| may as well use simple templates or even symbols.
| Jensson wrote:
| Figuring out what wont work is how you figure out what do
| work. All his points are good, they are things you would
| have to work around in some way.
|
| > Then we'll make new kinds of games where the
| unpredictability of the NPCs is a core mechanic.
|
| This is impossible, you need the NPC to be predictable on
| some level to make a fun game. Even unpredictable NPC needs
| to have a predictable personality on some level, total
| randomness isn't fun. Like, imagine a terrain generator
| that just randomizes terrain on each tile, that wont be fun
| at all, that is what a basic random personality would be
| like.
|
| Think of a human opponent, they are very predictable, just
| looking at a human player and what he does and I can
| predict what he will do in the future. Not perfectly, but
| players aren't that random. To make an AI that feels good
| it has to be very predictable.
|
| The main problem with "smart" bots is that they have so far
| always been way less predictable than humans, they get a
| strange edge cases and bugs where they start to act very
| dumb and strangely, that feels like a bug to the player and
| isn't fun. Or their smartness makes them do the same thing
| every time making them even more predictable than basic
| scripting, either way they are worse than basic scripting.
|
| Getting over these issues is a really hard problem, LLMs
| hasn't helped solve that so far.
| evandale wrote:
| > Even unpredictable NPC needs to have a predictable
| personality on some level, total randomness isn't fun
|
| tongue in cheek counterpoint: Rimworld players love
| Random Randy :P
|
| I think it really depends on the game though, but you're
| right 100% random in an RPG could be really annoying.
|
| Right now I'm into games like Factorio and Captain of
| Industry and they've both recently had blog posts about
| how they do terrain generation and CoI stuck out because
| you can manually plop features like mountains and then it
| procedurally generates the mountain range[1].
|
| There's been a lot of games recently that seem to be
| doing procedural land generation, is there not a way this
| can be applied to AI personalities as well or is there no
| overlap between them? It kind of feels like procedurally
| generated personalities should be do-able but it sounds
| like there's something more going on that complicates
| that?
|
| [1] https://www.captain-of-industry.com/post/cd-42
| Jensson wrote:
| > tongue in cheek counterpoint: Rimworld players love
| Random Randy :P
|
| Even randy random isn't entirely random, people love it
| since it sends you big threats, so it is coded to ensure
| it throws you big threats. If it randomly didn't send big
| waves people wouldn't like it as much.
|
| "If Randy has not fired a major threat after 13 days, the
| next Randy fired event becomes a major threat."
|
| https://rimworldwiki.com/wiki/Randy_Random
|
| > There's been a lot of games recently that seem to be
| doing procedural land generation, is there not a way this
| can be applied to AI personalities as well or is there no
| overlap between them?
|
| I'm certain that is possible, but we don't have nearly as
| much intuitive understanding how to generate full fledged
| personalities hooked into an LLM that changes how the
| character acts and his motivations etc that will actually
| work well when put in a world and interacting with other
| NPC's in that world.
|
| Terrain is just really easy to generate well enough,
| almost everything else is way harder.
| FloorEgg wrote:
| A pretty material portion of the population actually
| doesn't have much of any imagination, and these people,
| especially when they have had some success and developed
| "expertise" will always assume if they can't imagine it
| then it's impossible, or if they don't know how to do it
| then it can't be done.
|
| Surely you've met some of these pessimists before.
| summerlight wrote:
| Easier said than done. I assume you have no prior
| experience on professional game development? Many game
| designers tried to tame chaos as a game mechanic in the
| name of "emergent gameplay" and only a few of them survived
| through numerous iterations in an extremely limited format.
| I would recommend you to do your own research before making
| such a bold statement. It is not that people cannot come up
| with the same idea; many cases they tried it and there's a
| good reason not to do that.
| mike_hearn wrote:
| You can control LLMs to any arbitrary degree of specificity
| by a mix of retraining and changing the decoding strategies.
| They can be as predictable as needed, I think the bigger
| issue is more like how do you write stories when the
| possibilities get so much larger.
| summerlight wrote:
| Yes, less predictability is a part of the problems from the
| unconstrained search space. I think technically there is a
| room for improvement, but this usually needs ML expertise,
| which most of game designers and engineers do not have at
| this moment.
| treprinum wrote:
| Game AI already had the problem of too tough opponents two
| decades ago. Nobody (OK, outside Rainbow Six players) is going
| to play a game where AI demolishes any player within
| (milli)seconds.
| 2OEH8eoCRo0 wrote:
| Have you played Dwarf Fortress?
| ordinaryradical wrote:
| This is a death knell for MMORPGs where botting is already a
| massive problem which distorts the player economies and degrades
| everyone's playing experience.
|
| The cat-and-mouse game of stopping these goldfarmers just became
| exponentially harder.
| xypheran wrote:
| Maybe it's better that these time wasters die, speaking as
| someone who played them way too much.
| malux85 wrote:
| It's not just about time.
|
| During the teenage years while the neocortex is growing,
| teenagers are practicing and honing fine motor control. Video
| games help develop that, as well as learning about social
| interactions, emergent system behaviour and strategic vs
| tactical thinking.
|
| I'm not saying they don't have downsides, but there are some
| upsides too.
|
| Sure, if you're an addictive personality using video game
| escapism to ignore your life problems, that's a whole
| different thing (and even without video games this type of
| person would just find another form of escapism)
|
| So I don't agree with your generalisation of games as "time
| wasters" - maybe for you they are, but not for everyone, I
| don't play them anymore very much (just a bit of chess every
| now and then) but they provided me with a lot of
| understanding in my formative years
| JohnMakin wrote:
| > as well as learning about social interactions
|
| A large percentage of social interaction in a game like
| world of warcraft is profoundly negative and maladaptive.
| I'm not sure I would want my child learning about that in a
| MMO.
| malux85 wrote:
| Mine were overwhelmingly positive, and got me friends I
| still have 20 years later today, it depends on the MMO
| Tyr42 wrote:
| I mean I'm happier to get scammed over selling cow hides
| in game than used cars in real life.
|
| We don't give kids as many opportunities to make mistakes
| in real life, I dunno.
| digging wrote:
| I believe they are specifically calling MMORPGs time-
| wasters, which I second (as a video game enthusiast).
| They're not unique in being so but they are heavily
| designed to, basically, waste your time grinding.
|
| I also agree that even MMORPGs have their upsides. But as a
| genre they're pretty unhealthy.
| rincebrain wrote:
| I'm not sure I agree they're any more unhealthy than the
| non-MMORPG games full of microtransactions trying to get
| more money than the upfront cost in this day and age.
|
| We've reinvented how arcade games used to try and extract
| maximum quarters, but the iteration cycle is so much
| faster now that we can't really play whack-a-mole on all
| the pathological human manipulation strategies people
| deploy now, and with people not being able to physically
| walk away from their phones or other devices in many
| cases, it's Bad(tm).
| ryandrake wrote:
| > I'm not sure I agree they're any more unhealthy than
| the non-MMORPG games full of microtransactions
|
| I used to think: "Anything that got rid of the 'pointless
| grinding' aspect of RPGs would be an improvement." And
| then micro transactions and pay-to-win got invented. I
| didn't think it was possible but game designers somehow
| actually managed to make RPGs even worse!
| digging wrote:
| > I'm not sure I agree they're any more unhealthy than
| the non-MMORPG games full of microtransactions
|
| I didn't say they were
| LunaSea wrote:
| They are about as much a time waster than any other hobby
| really
| ytx wrote:
| But in terms of other benefits of that time spent, (imo)
| they're probably somewhat better than micro-transaction-
| gambling-mobile games (or just plain gambling), but likely
| worse than a sports league or chess club.
|
| Not sure how it'd compare against similar amounts of
| youtube/netflix though.
| LunaSea wrote:
| > chess club
|
| How? A lot of games could be seen as a sort of 3D chess.
| ytx wrote:
| Specifically meant in-person chess clubs as opposed to
| only playing lichess from home for hours every day. I'd
| probably have a less negative view of "time wasting" if
| video games were played more in-person too.
|
| I have fond memories of LAN parties growing up, where
| socializing was as big a part as the actual gaming - it's
| not like we were sitting there harvesting wood for hours
| on end!
| LunaSea wrote:
| Socialising is still a very important part of games
| (eSports, dungeons and raids, Discord / TeamSpeak /
| Ventrilo, forums, guilds, etc), especially for MMOs.
|
| Much more than chess which is mostly a individually
| played game whereas an MMO is a cooperative game.
| wtf_is_up wrote:
| >The ability to play chess is the sign of a gentleman.
| The ability to play chess well is the sign of a wasted
| life.
|
| - Paul Morphy
|
| One of my favorite chess quotes. As an avid chess player,
| I can't agree more.
| stronglikedan wrote:
| Why should I weigh the benefits of time spent on
| something that I enjoy doing? That would make it
| unenjoyable.
| LeifCarrotson wrote:
| What do you propose as an alternative?
|
| It feels fun to be rewarded for something you accomplish in-
| game.
|
| In many singleplayer games, you can slide difficulty up or
| down to change the effort:reward ratio.
|
| In an MMORPG, though, you have different groups of players
| with different amounts of time. You want to make it fun for
| both the kid on summer vacation who is happy to spend 80
| hours a week on a game (not a choice I want for my kid, but I
| was a kid once too) and an adult who has a 60-hour work week
| and exchanges 2 hours of sleep after the kids go to bed to
| play.
|
| That means the person with more money than time will want to
| buy things from someone with more time than money. But this
| causes all kinds of distortion in the game balance and
| economy.
|
| I don't know that this is solvable, whether you're trying to
| balance against cheap labor or AI bots.
| nozzlegear wrote:
| As someone who met his wife in an MMORPG (World of Warcraft)
| and still actively plays it each week with the same group of
| friends from 15 years ago, I'm not so quick to hope they die.
| wongarsu wrote:
| They can be problematic, but for many they are just a
| convenient excuse to spend time with friends in a voice chat.
| It's not really more or less of a time waster than most other
| group activities.
|
| Now playing alone for the dopamine rush of successfully
| grinding repetitive tasks: yeah, that's a bit of a time
| waster. Maybe therapeutic for some, and definitely not the
| most harmful way to spend time and get validation, but also a
| bit pointless. But I would argue that if you play an MMORPG
| alone you're doing it wrong. If you don't have friends at
| least get engaged in a guild and spend countless hours
| improving real-life social and leadership skills.
| vasco wrote:
| You seem to attribute more value to an activity just
| because it's done in a group but I'm not sure that holds.
| geometriccan wrote:
| Sometimes I can't tell if people on this site are joking
| or genuinely this out of touch. Go touch grass.
| alfiedotwtf wrote:
| I'm with Vasco on this one. What's wrong with enjoying
| your own company
| Jensson wrote:
| That isn't very easy to realize though, not too long ago
| game designers thought that multiplayer was mostly about
| having other players as fun challenges to overcome. But
| people like having other people even if they never really
| interact, just having others there that you can show what
| you did to and talk about stuff is fun.
|
| But point is, that realization isn't that simple, it took
| a long time for cooperative games to become common. In
| early days game consoles had cooperative split screen to
| let two players play at the same time, not because that
| was more fun, so it took a really long time for
| cooperative modes to become standard in online gaming
| because it wasn't at all obvious that people liked
| cooperative play.
|
| MMORPGs were the main cooperative online games for a long
| time. Today we have dedicated short session cooperative
| games, those are still very popular.
| vasco wrote:
| Instead of feeling superior in a drive-by snark, explain
| to me why gaming for 3 hours after work while voice
| chatting with 3 friends is not a waste of time but if you
| do it on your own it is?
| dreamworld wrote:
| I'll bite. First, each human is kind of a separate
| universe, another 80 billion neurons to converse with,
| each with our own histories and vastly different
| knowledge and experience. In a conversation, we learn a
| lot from each other, and better understand how we can be
| different in skills, and even in basic things like
| emotion, motivation, etc.., better understanding what it
| means to be a human, and better understanding what it
| means _to be_ in general. Also, it 's very important for
| us to maintain some kind of social contact (I think
| written counts as well), because our brains language
| ability will degrade and we will lose critical skills
| including reading social cues.
|
| Speaking of social cues, interacting with others
| specially in a complex environment where there can be
| severe competition as well as cooperation and difficult
| coordination, is something that also is worth practicing.
|
| I have nothing against solo games, but this kind of thing
| is not practiced in a solo game.
|
| Finally, I think other kinds of games (e.g. in
| competitive games) tend to have very simple interactions
| and objectives, compared to an MMO: there's a clear
| objective to win that's shared by everyone. Some MMOs
| have much more interesting interactions, where each
| person is interested in a different thing, and I think
| this contributes to a very rich atmosphere that isn't
| just 'Go win, try to win match, go out', i.e. more life-
| analogue (without other limitations of life, like you
| can't actually die, and being poor isn't as terrible as
| it often is IRL :( ).
| dreamworld wrote:
| So just be more in the real world? (I mean, IRL) Well,
| yes, but there are advantages to virtual worlds, as long
| as they're not designed to be simply addictive time
| sinks. And there are advantages to the real world.
|
| The inputs to a computer game are more limited, you can't
| see people, their faces (and sometimes voices), the
| graphics are still a far cry from the more beautiful
| places.
|
| Also, real life is full of responsibilities and large
| parts of it still, well, suck (bad jobs, exploitative
| practices, etc.). I think we're improving somewhat
| (greatly hampered by greed and power games).
|
| If you have interesting activities IRL, like a great
| fulfilling job and hobbies (that are also potentially
| useful in other ways, like charity work), then by all
| means, but I think virtual worlds have their place in our
| lives.
| Kerb_ wrote:
| 3 hours of gaming alone can be valuable relaxation and
| entertainment but doing it while voice chatting can be
| both as well as social engagement. Just because one is
| more valuable from most perspectives doesn't mean one is
| a waste and one isn't. You don't need to be in a lobby
| with friends to enjoy or be good at CS:GO or R6, but I
| think it makes you more likely to become a better player
| and cooperate with your teammates if you do, and I think
| engaging in cooperation in one realm of your life can
| lead to easier cooperation in other areas. They are both
| wastes of time from the perspective that you could be
| building something or doing a creative hobby with an
| actual output, and they're both effective uses of time
| from the perspective that recreation and skill are
| important even if they aren't essential skills. In the
| end, I think playing MMOs without engaging in the social
| aspect is a waste and you might as well be playing Cookie
| Clicker, but that doesn't mean I think MMOs or Cookie
| Clicker are waste of time in and of themselves.
| RohMin wrote:
| I guess you're not necessarily improving your social
| skills playing alone?
| jonReadingNews wrote:
| Another idea is this facilitates (re)creating these kinds
| of worlds and moments. Imagine your favorite MMORPG at your
| favorite time and imagine being able to recreate what feels
| like that time and place with other "players" being agents
| behaving in manners consistent in that context. Invite some
| friends. Have a good time. Throw it away.
| squidsoup wrote:
| We're an irrelevant blip in the natural history of our
| planet. Everything is a waste of time.
| suby wrote:
| We are arguably in one of the most important blips in the
| history of mankind.
| Jensson wrote:
| We are also in among the most important events of Earths
| history, just releasing all that trapped CO2 back into
| the atmosphere reversed many millions of years of robbing
| the biosphere of CO2, it will affect earth for many
| millions of years to come with more plantlife and warmth,
| maybe dinosaurs will be back in 100 million years if we
| release enough CO2 since they do better in warmer richer
| environments.
| serf wrote:
| The only tried and true metric that HN always falls down to
| question universally :
|
| "Why did you do that if it doesn't make money?"
| btown wrote:
| I doubt that the compute required to ingest game video in real
| time makes it remotely viable for botting. Even if it did, the
| above-human latencies between vision and agentic choice would
| be detectable by much simpler models operating on the more
| data-dense internal MMO server logs.
| maldev wrote:
| The compute and delay isn't worth it at all. Especially when
| you can disable rendering and keep compute pretty minimal.
| We're talking about reading a list of entities, moving towards
| them and casting a few spells vs a whole AI. Exponentially less
| compute, for better performance. Let alone the extra data the
| bot sees from the entity list vs the ai operating on visuals.
| Bots also make money by operating on a scale. And costs from ai
| would outweigh the already slim profit margins for each bot.
| sand500 wrote:
| > disable rendering and keep compute pretty minimal.
|
| So that's what was going on in the Matrix when the humans
| were staring at all that green text.
| TillE wrote:
| While there have been plenty of programming games, the idea of
| a bot-only MMO would be really interesting. Far more
| interesting than actually playing the typical post-WoW MMORPG.
|
| Like, Runescape was already distilled into a surprisingly good
| idle game in Melvor Idle. You could take a slightly different
| path where the "idling" is instead a matter of programming and
| resource allocation.
| nabakin wrote:
| > the idea of a bot-only MMO would be really interesting
|
| Then you may be interested in Screeps: World
| pests wrote:
| Screeps was fun but I really wish they made one simple
| change to the programming model: I want my screeps
| independent and acting on their own knowledge, without a
| global coordinator. The way I remember it scripts processed
| all entitles as a batch so you could "play god" and
| coordinate at a higher level. I really wanted it to be so
| each screep was independent and had to coordinate through
| agent interactions. act(screep) for screep
| in all_screeps // Independent evaluation
| act(all_screeps) // Global coordinator
| mike_hearn wrote:
| It just means MMORPGs would fully migrate to consoles, same as
| other multiplayer gaming. There's nothing fundamentally hard
| about stopping botting if you have good control over the
| hardware platform.
| jedberg wrote:
| If the botting is profitable then it is trivial to build an
| interface for a fake controller and screen grabber to do the
| work.
| FrustratedMonky wrote:
| That is correct.
|
| I think what people are missing, is that by the time you
| build a controller interface, and a screen grabber, and
| have an AI that can interpret the screen grab, understand
| and play the game, that this is super incredible, and
| really the humans are probably already being herded into
| Soylent Green processing centers to feed the remaining
| humans that are kept around for maintenance tasks.
| jedberg wrote:
| I think you're underestimating how easy it is to cut the
| cord on an Xbox controller and hijack the signals and
| just plug the HDMI cable into a capture card.
|
| And this post is showing you an AI that can look at the
| screen grab and play the game.
| FrustratedMonky wrote:
| LOL. Yeah man. Almost there.
|
| Here to offer praise to the AI overloads. Hope they read
| my comments later and know I was a true believer and
| should be included in the maintenance crews they allow to
| live.
| whalesalad wrote:
| I want to see it play helldivers on level 9. That would be so fun
| to watch.
| precompute wrote:
| "3D virtual environments and games" today, IRL tomorrow.
| satvikpendem wrote:
| Perhaps finally we will have competent AI in video games.
| tristor wrote:
| Great, now there will be even more people botting every single
| online game.
| TekMol wrote:
| The link should be changed to:
|
| https://deepmind.google/discover/blog/sima-generalist-ai-age...
| d--b wrote:
| The No Man Sky's "shoot the asteroid" demo is highly disturbing.
|
| They really should not ask any AI agent to shoot at anything.
| Especially when it's not very good at it.
| jamesdwilson wrote:
| > It's about developing embodied AI agents that can translate
| abstract language into useful actions. And using video games as
| sandboxes offer a safe, accessible way of testing them.
|
| not creepy at all.
| cj wrote:
| > safe, accessible way of testing them.
|
| And once validated, sell to the military?
|
| > Ultimately, our research is building towards more general AI
| systems and agents that can understand and safely carry out a
| wide range of tasks in a way that is helpful to people online
| and in the real world.
|
| This makes me nervous.
|
| I hope AI agents that take actions in the real world are
| regulated at least as much as self-driving cars have been over
| the last decade. Or at least AI agents that interact in public
| spaces.
| klabb3 wrote:
| I mean they already are, just look at that announcement
|
| > Ultimately, [..]
|
| I swear this short paragraph style rounding it off with an
| "ultimately", "in conclusion" didn't use to be so common. :
|
| Ai is already strongly influencing how people write. After
| being successfully deployed for a year.
| jamesdwilson wrote:
| DeepMind> kill dissidents
| dist-epoch wrote:
| You don't need AI for that, look at Russia, Saudi Arabia,
| ...
| jamesdwilson wrote:
| touche. the real power is the ability to blame the
| computer, isn't it?
| bogwog wrote:
| > And once validated, sell to the military?
|
| Can't wait to see the leaked footage of war crimes showing
| robots murdering civilians and teabagging their corpses
| belter wrote:
| "A generalist AI agent for 3D virtual environments" -
| https://deepmind.google/discover/blog/sima-generalist-ai-age...
| 101008 wrote:
| I understand that this is a great leap in AI and it sounds
| amazing, looks amazing, almost unbeliable. I wonder if it is
| needed, though.
|
| I can't find a good reason for computers playing videogames. I
| read another comment saying that they could be your buddies in an
| adventure game... what's the point? The fun is to play with other
| people. We already are able to play with bots (different
| algorithms rule them), so I can't see why someone would prefer
| this over them.
|
| About traslating this from a virtual world to the real world... I
| can't imagine who would think it's a good idea to give this type
| of freedom to machines in a physical world, were consequences are
| way riskier than something digital (and yes, digitally they could
| empty your bank account, physically they could kill someone. One
| is much worser than the other).
| kapperchino wrote:
| It's to replace qa testers for video games
| chankstein38 wrote:
| >About traslating this from a virtual world to the real
| world... I can't imagine who would think it's a good idea to
| give this type of freedom to machines in a physical world, were
| consequences are way riskier than something digital (and yes,
| digitally they could empty your bank account, physically they
| could kill someone. One is much worser than the other).
|
| Right now there are robots in many factories around the world,
| some are discrete machines that aren't tethered down and have
| movement capabilities. You don't think that there are factory
| managers/etc out there drooling about the idea of getting those
| or something similar to be able to do general factory tasks?
|
| Imagine your employee who tapes up boxes before shipment quits
| one day out of the blue. "Hey, package carrying bot 9000, can
| you go tape those boxes? I'll have someone show you what to do"
|
| Not necessarily a good idea still but just because we don't
| want it doesn't mean there aren't a million beneficial uses of
| this kind of generalizing.
| WFHRenaissance wrote:
| My thinking is that it enables a "more full" virtual world.
| solarpunk wrote:
| stadia buildout paying off more and more.
| qgin wrote:
| Honestly did not expect the physical side of robotics to be the
| bottleneck for fully autonomous robots doing tasks out in the
| world.
| 317070 wrote:
| Moravec's paradox hard at work. [0]
|
| [0]https://en.wikipedia.org/wiki/Moravec's_paradox
| qgin wrote:
| Wow, I guess lots of people saw this coming a mile away
| nsagent wrote:
| Sometimes I wish I had the reach of Google Deepmind. I created a
| sandbox environment for the text-heavy RPG 'Disco Elysium' [1].
| The current research I'm focused on is having an agent use a
| natural language interface (via text generation) to solve quests
| in the game.
|
| The project required lots of reverse engineering on my part to
| make a web-based facsimile of the game such that it's possible to
| conduct controlled experiments on the language capabilities of
| current agents.
|
| Hopefully what I've created will be useful for others, because
| unlike big tech, I've released all my code under the AGPL [2].
|
| [1]: https://pl.aiwright.dev [2]:
| https://git.sr.ht/~dojoteef/pl.aiwright
| doctorpangloss wrote:
| Isn't an essential part of what they are doing, and why they
| have results, that they are tackling all games at the same
| time, rather than focusing on one? Is Disco Elysium a good
| choice?
| nsagent wrote:
| Good point, they are quite different objectives.
|
| Their approach is one that works for simple directives: "Go
| to ship" or "Pick up iron ore" which lends itself well to
| sandbox-like games (which seems to be a major focus looking
| at Deepmind's tech report). Similar research has been done in
| Minecraft [1].
|
| These instruction following agents are more an RL achievement
| than a language understanding achievement. On the other hand,
| Disco Elysium has over a million words of dialogue, and
| solving the quests requires an agent to understand and reason
| about language much more extensively. People have looked at
| text-based game agents, like Microsoft's TextWorld [2], but
| these are much smaller in scope and not easily adapted for
| humans-in-the-loop.
|
| My work bridges that gap, focusing on the language aspect,
| rather than navigating a 3D world. Again, they are definitely
| different objectives, but as a sole researcher there's no way
| I can compete with Deepmind's budget and manpower anyway.
| Just look at the extensive author list in the tech report. So
| it doesn't make sense to necessarily focus on outcompeting
| them in producing a better generalized RL agent (in fact I
| merely use GPT-4). Instead, I made a publicly available
| experimentation platform that allows others to be able to
| build upon this work, which is valuable for the community at
| large.
|
| At least, that's my take.
|
| [1]: https://sites.google.com/view/steve-1
|
| [2]: https://www.microsoft.com/en-
| us/research/project/textworld/
| simpletone wrote:
| > Introducing SIMA: the first generalist AI agent to follow
| natural-language instructions in a broad range of 3D virtual
| environments and video games.
|
| If it can be done in 3D virtual environments and video games, it
| shouldn't be much of a leap to do it in the real world. After all
| we have cameras, voice recorders, sensors, etc that can map the
| real world into 3D virtual environments already. Have they tried
| linking this generalist AI to a robot to see how the robot does
| in the real world?
| actionfromafar wrote:
| Reality has a surprising amount of detail, though.
| acover wrote:
| And variation. Video games cheat at everything.
| isx726552 wrote:
| Like this?
|
| https://news.ycombinator.com/item?id=39691886
| wdroz wrote:
| Soon the 2017 April joke "GeForce GTX G-Assist" from Nvidia will
| be doable.
|
| [0] -- https://www.youtube.com/watch?v=smM-Wdk2RLQ
| ChrisArchitect wrote:
| [dupe]
|
| Some more discussion on the official post:
| https://news.ycombinator.com/item?id=39691783
| karmakaze wrote:
| I was just thinking how well could it play StarCraft II vs say
| DeepMind's AlphaStar, if I'm giving the high-level directives and
| SIMA is executing them.
|
| _Then I got the creepy feeling that this is likely the kind of
| wargames that are already being tested._ We 'll probably also
| need reverse safeguards where the AI raises concerns and requires
| confirmation before carrying out some requests.
| pdimitar wrote:
| Great, maybe we'll finally have NPC tanks and healers in dungeon
| queues so we the DPS players don't have to wait for 25 minutes.
| coddle-hark wrote:
| This got me thinking of Ender's Game, where they basically
| tricked a kid into committing xenocide by telling him he was
| playing a computer game.
| educaysean wrote:
| I immediately made this association too. Although thinking back
| on it, the connection is rather strenuous.
|
| Maybe we simply keyword matched on "video games" and
| "simulations". Or, perhaps more cynically, we're foreseeing a
| future in which AI agents don't care to differentiate between
| shooting at the enemy combatant in Call of Duty verses shooting
| at us in real life.
| tintor wrote:
| combat training agent? [0]
|
| this is direct violation of google ai principles on autonomous
| weapon development: [1]
|
| [0] Screenshot from SIMA Technical Report: https://ibb.co/qM7KBTK
|
| [1] https://ai.google/responsibility/principles/
| cboswel1 wrote:
| Dawg, we both know the moment there is any share holder value
| to be found in the tech, the TOS changes real quick. Look at
| Open AI.
| smallest-number wrote:
| > Weapons or other technologies whose principal purpose or
| implementation is to cause or directly facilitate injury to
| people.
|
| I dont think an agent fighting in a video game really counts?
| There is quite a significant gap between an FPS and a missile
| launcher, and it would be a waste not to explore how these
| agents learn in FPS environments.
| tintor wrote:
| What counts then?
|
| They intentionally included combat training in the dataset.
| It is in their Technical Report.
|
| How can combat training not be interpreted as "principal
| purpose or implementation is to cause or directly facilitate
| injury to people"?
|
| Do you believe the agent was trained to distinguish game from
| reality, and refuse to operate when not in game environment?
| No safety mechanisms were mentioned in the technical report.
|
| This agent could be deployed on a weaponized quad-copter, or
| on Figure 01 [0] / Tesla Optimus [1] / Boston Dynamic Atlas.
|
| [0] https://twitter.com/Figure_robot/status/17679136612539844
| 74?... [1] https://www.youtube.com/watch?v=cpraXaw7dyc
| bogwog wrote:
| This thing + Vtuber thing + chat bot that interacts with audience
| + text to speech == gaming influencer automation
| 10xDev wrote:
| So it can't generalise to other tasks? Then how is this much more
| than overfitting text/image input to controller output?
| TaylorAlexander wrote:
| Can we link to the blog post instead of the tweet?
| https://deepmind.google/discover/blog/sima-generalist-ai-age...
| gsuuon wrote:
| This would be cool to see as a game dev CI tool, an end-to-end
| playthrough test which would validate not only UI but the writing
| and game flow. Imagine getting a report that says something like
| "time to chop first tree: +20%"
| Jensson wrote:
| That isn't super hard to code, many of the large studios has
| such tools already. The main thing this could test that other
| things couldn't is the UX, that the UI actually is easy enough
| to understand for the AI to be able to navigate it etc.
|
| So if I could get reports like "+10% failed to understand how
| to chop their first tree" that would be good.
| gsuuon wrote:
| And for release builds, running the tests across a cohort of
| gaming profiles so you could get detailed results for various
| personas instead of just a guess and check in production.
| It'd be great for indie and hobby projects as well - getting
| playtest feedback cheaply and quickly would be awesome. Hope
| this turn this into a service.
| 2OEH8eoCRo0 wrote:
| Valheim provides a server that you can run yourself. It would be
| cool to populate the massive world with AI's.
| aussieguy1234 wrote:
| Is this AI simply pressing buttons as a human player would, or is
| it simulating physics to control a body in a 3D virtual space?
| margorczynski wrote:
| Pressing buttons, basically it operates the game just as a
| human would (image as input and mouse + keyboard as
| output/actions)
___________________________________________________________________
(page generated 2024-03-13 23:00 UTC)