AI beats top human player. Yet another AI discussion

[Magma_Dragoon]

Actually the future of AI is kinda scary

Tech singularity super AI is absolutely terrifying. Either we hand complete political control of our societies to such an entity, in which case are pets at its mercy, or the AI answers to a human elite, in which case have created a ruling caste which would likely be impossible to dislodge from power.

 

[Ferocitus]

Tech singularity super AI is absolutely terrifying.

Only if you believe in that kind of nonsense and think it will happen
in your own lifetime.

Either we hand complete political control of our societies to such an entity, in which case are pets at its mercy, or the AI answers to a human elite, in which case have created a ruling caste which would likely be impossible to dislodge from power.

Don't worry, eventually "we" will all win a Darwin Award (or one of
those nice participation certificates).

 

[ztrapon]

Anyway, poker is quite a simple game for an AI. In Limit Hold'em computers reached the level of top pros a long time ago. No limit Hold'em is more complex, but earlier this year someone made an AI that beat several top pros at that over a few hundred thousand hands

I suppose you are right, maybe I'm just desperately clinging on to some sort of hope that "we" have something computers don't. Every day that seems less and less likely though. lol

Tech singularity super AI is absolutely terrifying. Either we hand complete political control of our societies to such an entity, in which case are pets at its mercy, or the AI answers to a human elite, in which case have created a ruling caste which would likely be impossible to dislodge from power.

Wouldn't this be a switch-on kind of the thing as well? I mean the moment someone executes whatever code contains strong AI, it would start improving itself and rapidly exceed us. Assuming it evolves similar to biological life, it would only take seconds before it has repeated millions of generations already -- something that takes living organisms billions of years.

 

[Ferocitus]

I suppose you are right, maybe I'm just desperately clinging on to some sort of hope that "we" have something computers don't. Every day that seems less and less likely though. lol

We do. We decide what games are interesting and fun, we get to make up the
rules, and change them whenever we like.

Wouldn't this be a switch-on kind of the thing as well? I mean the moment someone executes whatever code contains strong AI, it would start improving itself and rapidly exceed us. Assuming it evolves similar to biological life, it would only take seconds before it has repeated millions of generations already -- something that takes living organisms billions of years.

Ask some of those who are scared whether it will happen in their lifetime.
If they do, they probably also believe in some of Elon Musk's Ponzi ideas too,
like colonizing Mars within his own lifetime.

 

[Pythakoreas]

Ask some of those who are scared whether it will happen in their lifetime.
...
like colonizing Mars within his own lifetime.

Civilization 6 taught me that it is possible before 2050. Well I'll be damn alive then and probably that young man as well

 

[Ferocitus]

Civilization 6 taught me that it is possible before 2050.

Yeah, but then you look at your score, sigh, and realize that you're Dan Quayle.

 

[TheMeInTeam]

Ask some of those who are scared whether it will happen in their lifetime.
If they do, they probably also believe in some of Elon Musk's Ponzi ideas too,
like colonizing Mars within his own lifetime.

To me, AI singularity is more credible as a near-horizon thing than "colonizing Mars", at least if we're talking human colonists. I doubt either are all that close, but in principle AI singularity needs a breakthrough + probably not that much in the way of resources. A team of people with sufficient knowledge (which does not exist, yet) and some money would likely be well within their means to create one if they knew how.

Putting people on Mars and having them survive has a tremendous resource & engineering outlay, but even basic stuff like "keep people below dangerous radiation levels" is no trivial task, same with medicine in such an environment (different gravity + long term health effects). Also unless we're talking about contained environments/domes the undertaking required to generate an atmosphere is absolutely freaking massive.

Something like a hurricane is chemically simple, but has tons of energy + size...enough that we can't really stop them. Finding a way to stop those actually sounds easier to me than giving Mars a viable atmosphere + keeping it there, even if we could somehow teleport materials to ignore that massive cost. If we go the dome route on Mars then costs are still enormous and a viable long-term settlement in 50-80 years is pretty laughable. Comparatively tiny earthbound projects can use a good fraction of that timeframe.

If we're talking "a colony of bacteria that can survive it" then something like that could happen?

 

[Ferocitus]

To me, AI singularity is more credible as a near-horizon thing than "colonizing Mars", at least if we're talking human colonists. I doubt either are all that close, but in principle AI singularity needs a breakthrough + probably not that much in the way of resources. A team of people with sufficient knowledge (which does not exist, yet) and some money would likely be well within their means to create one if they knew how.

There are many challenges for AI, but IMO the biggest is to reduce the energy
cost. IIRC (and I'm happy to be corrected!) about 8% of the world's current
total energy production is used by AI applications.

Human brains are about 100,000 more energy efficient than silicon-based
processing, so it might be that huge arrays of "mini-brains" are the way
forward.
Here's a very recent article showing a lovely little mini-brain (I imagine)
bopping to a tune only it can hear.

https://www.quantamagazine.org/mini-brains-go-modular-20170809/

 

[Hammurabbit]

Another interesting ML project explained.

. Will be interesting to follow the arms race between OpenAI and DeepMind.

 

[elitetroops]

Another interesting AI question is what the AI should be allowed to know about the code. If we assume an AI aims at playing "perfectly", whatever that means, what knowledge of the game is it allowed to use as a basis for it's calculations?

Civ 6 does a notoriously bad job explaining details of the mechanics to the player. Most players play blind, oblivious to what is going on around them. Hardcore players code dive to find out about the mechanics more in detail (or search these forums for what the code divers have found). Then there are also some mechanics that not even the code divers can figure out. Should an AI be told more about the mechanics of the game than the game tells the human player? Or does this even matter? Maybe the AI would find these out on it's own through trial and error during the learning phase...

 

[Siesta Guru]

It's generally way easier to just mimic the code of say a damage calculation than it is to make a human-like damage estimator. So yeah it's likely that AIs are a bit more in tune with the actual code, which I don't think is a big problem as long as it's not predicting random events or something. Usually though, the match won't be perfect since AI devs need to take into account that the game logic will change during AI development. You can't really wait for the rest of the game to be finished to start with AI.

 

[TheMeInTeam]

There are many challenges for AI, but IMO the biggest is to reduce the energy
cost. IIRC (and I'm happy to be corrected!) about 8% of the world's current
total energy production is used by AI applications.

Human brains are about 100,000 more energy efficient than silicon-based
processing, so it might be that huge arrays of "mini-brains" are the way
forward.
Here's a very recent article showing a lovely little mini-brain (I imagine)
bopping to a tune only it can hear.

https://www.quantamagazine.org/mini-brains-go-modular-20170809/

I don't disagree, but improving AI efficiency:

1. Has very high economic incentive even in the short term.
2. At least at a glance appears to require less innovation than sufficiently reducing the relevant costs to setting up a viable Mars colony.

That could eventually trickle into games like Civ, but I wouldn't be holding my breath to see it for Civ 7 or 8...really this franchise is setting itself up to get beat, and not just due to AI. Hopefully competition can manage it.

Civ 6 does a notoriously bad job explaining details of the mechanics to the player. Most players play blind, oblivious to what is going on around them. Hardcore players code dive to find out about the mechanics more in detail (or search these forums for what the code divers have found).

Rather than trying to program the AI against flagrant design flaws that not even the developers could come up with coherent reasoning to justify, it would make more sense to simply fix these broken aspects of the game.

 

[BrokTheFanatic]

Bots like that one get advantages. When the bot can walk up, sit down, hold and control the mouse, look at the screen and react to what it sees, etc., then you have a true bot.

If I could look at the packets and have flags popping up on my screen in DotA or any twitch a game, like some cheat/hack programs provide, I'd have a much easier time of it.

The bots to never make are like the ones in Oblivion, because human target recognition combined with firing accuracy is a deadly machine. Of course, the US army/navy/air force are working really hard on these. A 'bug' will wipe out a school bus of kids in 2 seconds.
----------------------------------------------------------------------------------------------------
As for the question, it should be the best you can make and then the skill level selectable, much like how AI chess programs work. The skill setting is usual based on how many steps ahead it will search. Civ games also have many variables that have long-term implications. Consider Go, Civ is more complex than Go and AI Go cannot yet beat Go masters.

 

[Pythakoreas]

Consider Go, Civ is more complex than Go and AI Go cannot yet beat Go masters.

Look up "AlphaGo". For many years it was thought to be impossible to build a computer that could compete in Go but here we are today.

And I think the priority as of now should be making the Civ AI avoid preventable and obvious mistakes. I won't even ask for an AI that could outsmart a human player when the AI still doesn't have a grasp of how to use its Units.

 

[Acken]

Also the comparison makes little sense.
The purpose of AlphaGo was to have an AI to beat the top human. Go AI that beats mediocre players have been around for a long time. Where we are in Civ is wanting a bonusless AI that at least plays decently. Then the top players will continue to activate the bonuses to get a challenge.

 

[BrokTheFanatic]

AI is all about rules and/or patterns. Some learning with the more modern ones (but this AI 'learning' is extremely structured, it is more a directed search for an optimum)

Basically, if you can make a procedure out of the strategy then you have an AI. Sometimes that takes deep search trees (as in chess) or lots of pattern searching (like in Go) (both use a mix of each), but it always boils down to defining what is 'winning' in a good way and then coming up with structured searches and a useful 'knowledge' database of what 'works' and when.

Civ is more complex so the above is difficult to create. I have suggested over the years so simpler methods for AI unit strategy and tactics. All falls on def ears. Previous creators have also made lame excuses that they don't want the AI to be 'great'. Basically, they don't want to make the effort even though making the AI 'not stupid' 90% of the time would go a long, long way to making Civ a far superior game than it is atm. I bought this latest civ on sale since I knew it would be like this. Unfortunately, most gamers don't seem to want a challenge these days.

 

[Karmah]

Who 's actually working/studying IA here ? I' m pretty sure this forum has more fellow data fanatics than the average forums.