> Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI.

Is this really that different than what humans do?

After watching dozens of AlphaStar commentary videos [1] over the last few months, I was more or less thinking the same thing: the AI has basically evolved a massive ruleset: "Do X. If you see Y, do Z."

Nonetheless, I decided to start playing again myself. Poking around, I saw a recommendation to go through one guy's sort of "training course" [2], and guess what? A lot of it comes down to the same kind of thing. "Send your first overlord to scout their natural. If they haven't expanded to their natural, build one -- ONE -- spine crawler in your natural."

How much of our "intelligence" is really anything more than pattern matching + search? And during the actual dogfight, how much of what the human pilot was doing was anything more than simply pattern matching from their own vast experience racked up in a simulator?

[1] https://www.youtube.com/playlist?list=PLVRQoOk_ltE3Fr1ofRE0Y...

[2] https://www.youtube.com/playlist?list=PLFeZeom2b4Dlt63qmkPO8...

> A lot of it comes down to the same kind of thing. "Send your first overlord to scout their natural. If they haven't expanded to their natural, build one -- ONE -- spine crawler in your natural."

I think your general question is still a good one, but it's worth noting that from a human perspective, Starcraft matches only start off this way. Very quickly, the game state becomes complex enough that decision trees break down and intuition becomes the driving process for high-level human players.

The extent to which AI can begin to compete with this sort of intuitive human processing is most interesting to me. As it relates to this article, I think it matters a great deal if the experiment has constrained the system to the degree that it ceases to operate in that intuitive realm that high-level Starcraft matches operate in.

> Very quickly, the game state becomes complex enough that decision trees break down and intuition becomes the driving process for high-level human players. The extent to which AI can begin to compete with this sort of intuitive human processing is most interesting to me.

As far as I can tell, we're already there. Go is a game where search + explicit rules work really poorly, and intuition is a heavy factor in how humans play. AlphaGo was the first major success where neural nets were able to replicate this "intuition", and actually become far better at it than humans.

You can definitely see this "intuition" factor if you watch the AlphaStar games. Basically, if you have two armies with "middle of the road" units, AlphaStar always knows when to take a fight. You can sometimes see it take its army up to a defensive line, then turn around and go home; but if it decides to attack up that ramp, it always wins, even in a situation where a human would be less sure.

The exception is with unusual unit compositions; for example, it seems in training to have only encountered 1-2 battle cruisers at a time. So there are games where over and over again it throws an expensive army full-throttle into a group of 10 battle cruisers, and over and over gets slaughtered.

But again, I think this "intuition" factor, even in humans, is not much more than a sophisticated pattern matching, perhaps with background search.

> The exception is with unusual unit compositions

Could be wrong, but I believe this is actually a case-in-point moment for the other side. AI as it exists now is flummoxed by unfamiliar input, whereas a human expert is more likely to be able to react in a way that is at least directionally correct.

Put another way, if AlphaStar is confronted with a Terran opponent that can build Zerg units, how likely would it be to win the game? How likely would an expert human be to win that game?

No, what AlphaGo does should never be considered "intuition". It is simply using perfect information to optimize its probability of winning. We have no idea what intuition is or how it works. That's why we have learned literally nothing from AlphaGo. In fact, players who have played against it have often gotten worse. Intuition is very obviously much more than pattern matching. You need thought and probably some form of language to employ it, both of which we have very little understanding of.

The korean fight it beat human For many games afterwards.

Some insight gained like going to three three which is a taboo like old day playing three three alone.

Intuition or not ... learning we from computer go.

What the problem is they are perfectly known game. Air battle is not a perfect information game. That is why you need human decision.

> It is simply using perfect information to optimize its probability of winning.

Perfect information would be the knowledge of all possible outcomes of a given move. That's not even possible in chess, and is fantastically less possible in a game like go. That's why, until AlphaGo, there had never been a go computer program that had ever even come close to beating a professional go player.

Let me emphasize that: As of early 2015, nearly 20 years after Deep Blue beat the world champion Kasparov at chess, there had never been a computer go program that had come close to beating a professional go player. The game starts with literally 361 possible moves, and each move thereafter decreases the possible moves by exactly one. The search space is just so massive that it's impossible, by brute force computation, to do a search with any meaning.

What's needed is an intuition about two things. First, for a given a board position, which color is more likely to win? Secondly, given the hundreds of moves you could make, which ones are the best ones to explore? Go players have some rules, but primarily they develop an intuition for these two things by playing and looking at hundreds and hundreds of games.

AlphaGo's primary architecture was the same. It has two neural nets, which spit out board evaluations and move suggestions. These nets don't have explicitly coded rules, but were developed by playing millions of games. How is that any different than our wetware neural networks?

> In fact, players who have played against it have often gotten worse.

I hadn't heard this, but this is probably not unusual. Take my StarCraft example: Suppose someone had gotten to Diamond league mainly by perfecting their early rush strategies, but then hit a wall. Then they watch that video series which says the foundation of a GrandMaster strategy is having a solid economy first. If they decide to take this advice, it will entail a complete re-building of their skillset; they'll almost certainly drop in their rankings before picking back up again.

You could imagine the same thing happening in go: for hundreds of years, certain principles have been believed to be true. AlphaGo regularly violates these principles. As people are exploring this alternate strategies, they will inevitably get the "new" principles wrong while they're learning.

On the other hand, the early versions of AlphaStar clearly dominated mainly by having inhuman micro capabilities (ability to precisely control individual units). A human trying to replicate a strategy that relied on inhuman micro would inevitably fail.

Similarly, it might be that certain moves in go are good moves if you can do the kind of massively deep search that only AlphaGo can do. If that's the case, then of course humans trying to imitate AlphaGo are going to fail. A similar thing happened in chess: computers historically have been amazing at tactics and weak on strategy. A human playing a computer should focus on a good strategy, because they're never going to beat a computer at tactics.

But none of that changes the fact that what AlphaGo is doing, with two neural nets that spit out answers based not on explicit rules but based on millions of games worth of experience, is indistinguishable from human intuition.

I think you need to look up the definition of "perfect information games". It is not what yoU asume it to be, when you say "knowing all possible outcomes".

Clearly it can surpass peak human skill at StarCraft, but it’s important to understand what Deep Mind is doing differently.

One example is memory. Deep mind doesn’t update it’s strategy when playing the same player repeatedly or even over the same game. It can operate at near peak human performance indefinitely, but it avoids being exploited via deep understanding of the exact rules in play. It was also playing on the ladder under random names to avoid people developing specific counters.

> One example is memory. Deep mind doesn’t update it’s strategy when playing the same player repeatedly or even over the same game. It can operate at near peak human performance indefinitely but it avoids being exploited via deep understanding of the exact rules in play. It was also playing on the ladder under random names to avoid people developing specific counters.

This is because many games without perfect information have "rock-paper-scissor" situations. An AI designed without this in mind, lacking a source of randomness (or memory of the last encounter fed into the next game), is going to pick "rock" every time and start losing to the human who realizes and exploits this.

> Is this really that different than what humans do?

Yes, we have built in mechanism that react to situations we never encountered before. You will never be able to simply sit calmly while something is damaging you for example.

Reacting to damage is something you learned in previous generations. You've encountered that situation billions of times (you just don't remember them).

Imagination plays a significant role for us humans. We also are capable of maintaining incomplete conceptual mental models. Some of us experience epiphanies. Etc.

Pattern matching and search are brute force approximation of actual “intelligence”.

As far as I understand, AlphaZero had three main components:

1. A model of how Go worked, constructed by humans

2. A neural net for generating interesting moves, and a neural net for evaluating game positions.

3. A Markov search algorithm that used 1+2 to do pattern patching and search.

I will say that so far, #1 -- constructing mental models which can then be searched -- is something I haven't seen an AI do yet.

But looking at the actual work that I do -- from coding to deciding what to cook for dinner to deciding how to lay out my extension -- I'm pretty sure that behind the scenes it's exactly the same as 1+2+3 above; and that if I could download my mental model and my utility evaluator into a program, it could (with training) do all of the above faster and more consistently.

After AlphaZero, DeepMind went on to develop MuZero (https://en.wikipedia.org/wiki/MuZero, paper: https://arxiv.org/abs/1911.08265), which is able to learn to play games without being provided an explicit model of their rules. "When evaluated on Go, chess and shogi, without any knowledge of the game rules, MuZero matched the superhuman performance of the AlphaZero algorithm that was supplied with the game rules."

We are quite different as we can evaluate new situations based on our understanding of the principles behind something like a build order in Starcraft.

The “one” spine crawler thing is a bit overstated in the context of an actual game. It’s not like it’s 100% necessary to build it, build it in your natural, or only build one depending on what you’ve scouted, the opponent’s race, and who you’re playing against.

Like the sibling said, the game begins like a decision tree but quickly falls outside that purview.

Well you have to remember who he's talking to. You can sort of tell that lots of people, if they think they're about to be rushed, would make loads of spine crawlers as a defense; the emphasis on "ONE spine crawler" to delay an attacking army just long enough that you can build your own army in response (assuming you see it when it starts to cross the map) with the minimal impact on your economy.

And my point is that basically what he does in the series is start with a super simple set of rules, and then slowly add more subtleties and sub-clauses. That's obviously at some level how he works when he's actually playing -- he has a general plan, then a whole load of patterns that he matches on in how to respond. I'd be willing to bet that he rarely encounters a genuinely new situation for which there is absolutely zero precedent -- absolutely nothing in his experience which is similar enough for him to take an existing response and adapt it for that purpose. How is that different to what AlphaStar is doing?

The main difference is likely to be in how they learn afterwards. If he encounters a new situation in a game, afterwards he will use his model of the game to try to come up with new responses. AlphaStar will most likely have to do something much closer to brute force to find an appropriate response.

> "Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI."

Seems like with every breakthrough the goalpost of AI gets moved.

I believe a researcher coined a phrase for this but I can't remember what it was.

It doesn't get moved so much as we discover that a certain mathematical manipulation plus all our previous mathematical manipulations don't amount to an AI.

It also seems like with every breakthrough this complaint is raised without addressing the underlying issues with the new AIs shortcomings.

My hot take: true AI is so far out of reach of our ability it's not even funny, which makes the whole field either a search for the fountain of youth at worst, or at best a search for tools to inform humans or to replace humans in rote tasks. See https://youtu.be/orMtwOz6Db0

Ummm, no? AI means a clear thing: artificial _intelligence_. Depth-first search is not intelligence. Auto-generated algorithms that succeed on most of the training data is not intelligence. And that is obvious from the errors "AI" makes that a human would never do.

For example, would you fool even a child with this? https://cdn.mos.cms.futurecdn.net/s4DuKgTLnS4cngTyiqVkNC-970...

Someday, this line of argument is going to conclude that humans are not intelligent.

> I believe a researcher coined a phrase for this but I can't remember what it was.

There's an article on Wikipedia called "AI effect", I don't know if that's what you had in mind.

https://en.wikipedia.org /wiki/AI_effect

I think there's value in having a huge learning set, but there are also limitations, as you're alluding to.

But I also don't think we're miles away from computers learning some sort of reasoning structure. Some sort of causality-type thinking where you have hierarchies which at some level are reasonably simple, because humans can only fit so much. But when the computers figure it out, they don't have that problem.

At the moment, yes, there's a huge corpus of patterns and you can make some smart decisions just by being able to learn from the huge library, but it's the difference between knowing that one move tends to beat another, without knowing the why. For instance in sports, you have man-marking vs zone-marking. The naive thing to do is just tabulate how often a team did one or the other vs how often they won. Then break it down even more by who they were facing and various stats. But if you don't have a theory of marking, you're a bit lost for explanation, even if the tabulation clearly says zone marking tends to win. A causal explanation might sound something like "man marking allows the other team to pull you out of shape and gives them the choice of which players face which". It might also tell you that sometimes it's actually smarter to man-mark, eg when there's some player you really feel is dominant and needs to be taken out.

I gather that people are working on this causality type AI though, so no doubt we'll see something interesting soon.

As much as I agree with your assessment.

You only get ONE chance to beat AI in real life dogfight. So you better make sure that your strategy to exploit AI is going to work.

That's now how wars work. You get time to learn and adapt to the enemy strategy. One plane down doesn't mean anything if you learn how to exploit the enemy's entire fleet of "AI" with 1 casualty.

"That's not how previous wars worked" would be a more accurate take. Considering how few aircraft there are in modern militaries and how long it takes to manufacture modern fighter aircraft and missiles, one side will be out of high tech aircraft/missiles before the other and then it's game over.

This is one of those things I kind of wonder about the reality of. It's absolutely true that you're unlikely to be able to manufacture a lot of what you'll use during the war.

But simultaneously, it's not like the United States is really trying to fix this problem. The modern generation fighter aircraft are boondoggles with massively inefficient manufacturing designed to win senators and congresspeople votes - not act as the backbone of a hardened supply chain.

They're being built the way they are because the US doesn't seriously expect to face any enemy who can substantially deplete its forces, or is going to have any ability to strike the mainland.

My suspicion is that if there was an actual belief in the need to have reliable supply lines, it could be done and manufacturing could be streamlined a lot. You could definitely imagine a nearly fully automated factory rolling these things off very quickly.

> "My suspicion is that if there was an actual belief in the need to have reliable supply lines, it could be done and manufacturing could be streamlined a lot. You could definitely imagine a nearly fully automated factory rolling these things off very quickly."

Unlike previous wars, cruise missiles are a mature technology, making it much easier to precisely zap an opponent's manufacturing facility and infrastructure than before. High tech gizmos also have a lot more dependencies that go into them, making it easier to disrupt by knocking out a few or even just one of them.

In case of a prolonged conflict you need to consider the AI and it's programmers as a joint system. The AI by itself may be exploitable, but the programmers will adjust it to counter anti-ai strategies.

We don't program what are called AI's nowadays. We curate data, sometimes supervise the learning process, and do our best to detect when a model is bad.

If you think there's any intention or well defined predictability to these things, I've got a bridge to sell ya.

This is why people train. Once AI piloted planes are prevalent on the battlefield, there will be a huge effort to capture enemy systems and develop tactics against them. The USA has been doing it for decades (https://en.m.wikipedia.org/wiki/Have_Doughnut), I’m sure that China and Russia have as well.

But how do you know which version of AI are you fighting?

Its not really all that simple, as you have limited time/resources to determine what exactly you are up against.

And you get to test your hypothesis once. You win or lose after that.

The biggest thing you seem to be missing is that none of your arguments are unique to fighting AI. You still only get one chance against a human pilot. You don't know how good they are individually, you don't know if they have any special tricks, and you don't know whether they've got all the available upgrades on their jet, or if they have some old or broken equipment. You do know the average skill level of the enemy pilots, and what equipment they might have, so you make your decisions by assuming that they are a competent pilot with the best equipment that is likely to be on that jet. Warfare still happens even when you don't have time to gather up perfect information and make a perfect decision.

One other thing to note is that there will not be a unique AI in every individual jet; that would be impossible to maintain. There will be one version rolled out to each type of jet, maybe with a few customization options that can be changed during mission planning. When a new AI version comes out of testing it will get rolled out to every plane in the fleet that has the required computing power- some jets used for training/reserve/national guard duties won't have the latest hardware upgrades and won't get the newest version. So a pilot will be able to ID a jet, know where it is generally strong and weak, and then try to target the enemy's weakness while avoiding the strengths. Sometimes they'll lose, but once enough people have survived encounters with AI jets (either from exploiting stolen jets or from plain old combat) the best practices will get pushed to the other squadrons. Just like what already happens.

Except an AI tweak which changes behavior enough can be rolled out - let's face it - probably via a wi-fi update within minutes (aviation is, ironically, pretty simply compared to self-driving cars - we're not talking about big algorithms in a lot of cases).

The side with AI planes is just looking up K/D ratios on their telemetry from the planes. The other side is losing pilots.

Ejection is a thing. At this point it’s also about who has better missiles and more of them.

The USAF missiles will make guns obsolete doctrine of the 1960’s was premature not completely wrong.

It's almost always about who has better missiles and longer range radar/sensors. No point in getting into a dogfight if you can shoot the opponent out of the sky from 100 miles away.

In general yes you can make that distinction but more importantly this is a problem for peace time.

No one would authorize a non-visual shoot down of a target in a non-active AO and not during an actual conflict.

If you are at the stage where you have factions duking it out in the skies especially these days were talking about a major conflict in which case there won’t be any civilian jets flying there anyhow.

There is probably a way they know his, but can you really fire a missile at something on a computer screen miles away and be fully confident the thing you are shooting at is enemy military aircraft? And not say a civilian aircraft ?

I guess if its going fast enough its almost certianly a military aircraft. Can the radars in the military jets identity the type of the aircraft before it gets blown out of the sky?

Radars on the F-15 can determine the type of turbine on an aircraft...

Another way is to announce a no-fly zone and then assume anything within the zone is hostile. (Yes, civilian planes can still fly there, but they've literally been warned at that point)

I assume this would happen pretty quickly when a war went hot.

That lets you distinguish friend from not-friend, but not enemy from airliner.

Until it isn't, because the enemy has already captured and compromised your government back home.

> It cannot reason on the spot and will fall for the most ridiculous of traps.

> My theory is that AI will not exist until we reach AGI.

You are confusing the definition of AI for AGI so of course you think that. AI doesn’t need to have true understanding to be considered AI, it just has to have the appearance of intelligence.

I'm not an expert but I think in the future you will have swarm of drones against another swarm of drones. No human involved (in the air at least).

The best algorithm will win the war.

People stop bullying your nerds at school if you don't want to loose your next war ;)

> I'm not an expert but I think in the future you will have swarm of drones against another swarm of drones.

Unlikely. You will rather a swarm of drones against an enemy that has no such technology.

A country that can have a swarm of drones would very likely also have nuclear weapons and deterrence is a thing.

Canada isn't going to be building nuclear weapons ever again[0], but we're going to have drone swarms. They don't melt the face off of a bunch of civilians. They can selectively target. They're cheap.

[0] Most people don't know this but Canada is one of the few countries that built nukes (in conjunction with America and the UK) and then gave them up. I think we still technically own the seven nukes, but America is in possession of them.

IIRC Daech is using swarms of drones against the Russian air base in Syria (they probably don't have AI inside though).

I think AI powered swarm of drones is certainly feasible for a small team of motivated engineers and modest funding.

The thing that would let them at a disadvantage is the quality or lack thereof of their missiles guidance system.

Why do people talk about swarms of drones but never of swarms of fighter jets? That's incredibly weird terminology considering drones are never deployed in this manner.

> I think in the future you will have swarm of drones against another swarm of drones

Unless the swarms are easily exploitable like the current "AI"s.

Yes, only thing is... usually it does not work that way in reality, and when they breech enemy defense drones, they will continue with a quick genocide.

>We've seen this before, even for a lot more limited and controlled environment like the game Dota the "AI" can beat the humans a few times before the humans learns to exploit it's many weaknesses.

I recall that there was AI beating top Dota players. I'm not familiar with humans figuring out the AI and exploiting weaknesses. Every search I've done just shows articles about the AI winning. By chance do you know where I can read up on humans figuring out the weaknesses of the AI?

After AI beat 1v1 the top players they gave it to casuals at the international if I remember correctly: https://www.reddit.com/r/DotA2/comments/6t8qvs/openai_bots_w...

> The general strategy is to win by claiming first tower. At 0:00, you aggro the enemy creep wave so that they start following you. Then you walk around in a circle around the jungle, and the enemy wave will start to form a congo line that will follow you around. You then path around the jungle so that on the next wave spawn, you can aggro the wave again and continue to walk around in circles. The AI will burn glyph when your creep wave hits the tower, and for some reason it can't really decide between chasing you or defending the tower. So after about 5 minutes of doing this, your creep waves will eventually destroy the tower and you win the 1v1.

This reminds me of how 2b2t PvP has evolved. At first everyone was just wearing fully enchanted diamond armor and had golden apples that regenerate HP faster than you can lose them. Hacking is allowed on 2b2t. Because of dupe glitches getting large quantities of specific items is possible. Ender crystals that are normally used to resurrect the ender dragon can be smashed to create an explosion which kills players in full diamond armor instantly. Suddenly it was all about who could deploy his crystals first. Then totem duplication + auto equip allowed you to survive instant deaths until your inventory no longer contains totems.

No matter how impressive each of these techniques were, each of them had a counter that was highly effective.

https://openai.com/blog/more-on-dota-2/ mentions some of the strategies to beat the bots.

You can also check the ml subreddit for discussions of exploits:

https://www.reddit.com/r/MachineLearning/comments/bfq8v9/d_o...

https://www.reddit.com/r/MachineLearning/comments/6t58ks/n_o...

https://www.reddit.com/r/MachineLearning/comments/bcumrs/d_o...

https://www.reddit.com/r/MachineLearning/comments/6u304t/n_m...

I also recall a post from one of the pros talking about his strategies to try and beat them but I can't find it.

>Dota the "AI" can beat the humans a few times before the humans learns to exploit it's many weaknesses

Lets retry this experiment but the loser of any games get shot in the head, and the next player only gets basic telemetry while also shitting themselves. A human Dota player with a gun pointed at him will likely perform differently too.

Note that the AI player doesn't get killed if it loses, because it is software. It just doesn't get full telemetry and diagnostics.

Yes, they are all specialised AIs but we make them so fast nowadays that it doesn't matter they don't generalise yet, we can have as many as we want and they can be really useful in many fields. ImageNet can be trained in 30 seconds from scratch, and fine tuning takes just a few extra epochs. We don't need to wait for AGI, most low hanging fruit have not been picked yet.

Not sayin they are not useful. They are. But they are bots. Not AI.

Without a clear definition of AI that everyone agrees on, we will never reach AI. If an AI is considered intelligent only if it is "General", then are we, as human, even intelligent? I would argue strongly that we are missing the 'G'.

OpenAI Five lost just 42 out of 7,257 matches. It's true that it has exploitable weaknesses, but it did not simply fall apart.