[FZ1]

Why are they calling it "AI", though? There isn't any AI or ML.

You leave a trail for the enemy to follow, and they follow it.

It's not even path-finding, it's path-following. Which is pretty much an if-then statement.

It's a neat, simple approach, and fun to watch. But there isn't any learning, or knowledge, or other AI.

[tantalor]

AI in the sense of "player vs. AI" is widely understood.

AI does not require ML.

"It's part of the history of the field of artificial intelligence that every time somebody figured out how to make a computer do something—play good checkers, solve simple but relatively informal problems—there was a chorus of critics to say, 'that's not thinking'."

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

[marcinzm]

AI does not mean ML, it is a broad field that is a superset and not a subset of ML. Or as Wikipedia describes it:

>In computer science, artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and animals. Leading AI textbooks define the field as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.

[FZ1]

> AI does not mean ML

Hence the 'or' in my statement. Neither are present here.

[marcinzm]

>any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.

This does exactly that.

[FZ1]

Every program that has ever existed does this. So, you're saying that all programs that have ever existed, then, are all AI. You make no distinction whatsoever.

I would say that the more a program thinks on its own which actions to take to maximize its chances of success, the closer to AI it is.

If it's doing exactly what it's explicitly told, then it's not really intelligent, is it?

[serf]

>>any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.

>Every program that has ever existed does this.

No, not every program is self-tuning, nor do they all take inputs.

[winterismute]

I strongly disagree. For example, in this algorithm, there is "knowledge": the knowledge of the programmer that realizes that following a person moving can mean also follow one of the previous positions of that person (especially if we can revert to follow that person again at some point), and this knowledge is encoded in an algorithm using code, that is, yes, those "if-then statements" you seem to despise.

For years (but even now) Artificial Intelligence meant understanding how intelligent behaviors worked, and then understand what is the sequence of "if-then"s that could express those behaviors in an artificial setting. Use statistical inference is a "hack" (unavoidable, often) to cover the cases in which such behaviors seem to be too complex to be grasped - and then expressed - by one simple and/or comprehensive algorithm, but the (possibly unattainable) ideal would be having everything expressed as "pretty much an if-then statement" indeed.

[FZ1]

> the (possibly unattainable) ideal would be having everything expressed as "pretty much an if-then statement" indeed.

This is flatly incorrect - the point of AI is to have a machine achieve intelligent behaviors without explicit programming.

If an "if-then" must be written by a programmer for every single behavior, then this is called "programming". It is not called "artificial intelligence".

[marcinzm]

No, the point of AI is to have agents achieve goals based on observing an environment. Nothing says they have to be complicated agents or not explicitly programmed. Any book on Computer Science AI will be largely filled with agents that use rather explicit logic and algorithms.

[FZ1]

Still incorrect. And I would urge you to read one of these books you reference - they ALL aim to achieve that agent's action ON ITS OWN - i.e., by learning from its environment, and NOT by being explicitly programmed.

Yes, there are many explicit if-else style programs in Russel & Norvig, & other books - but those are the 'training wheels', until better methods are developed. For actual AI, the training wheels are supposed to come off, and the agent learns and acts on its own.

[marcinzm]

There is a distinction between what is AI and what is state of the art AI. Simple approaches are AI but not state of the art AI. At one point massive rule based systems were considered state of the art for example and those were nothing but explicit if-else statement. Now they're not but they're still AI.

BASIC, for example, would be considered merely a training wheel in any book on programing languages but it is nonetheless a programing language.

[jfkebwjsbx]

Not at all. AI does not need any learning at all.

You are talking about ML, not AI.

[ncallaway]

In games "AI" is the term for the system that controls the behavior of NPCs and other non-controlled entities.

It's not claiming to be "academic AI".

Games have been using this terminology for decades.

[jfkebwjsbx]

It is academic AI.

People simply have no clue what AI actually means and think it is only about sci-fi robots and ML.

[ncallaway]

That's fair. I guess I meant more "the frontier of academic AI", but then again I don't really keep up with academic AI at all, so I might just be wrong about what academic AI is. I can't really make any strong claims about academia.

My stronger claim is that this falls so squarely within the bounds of "game AI" that it's fairly ludicrous to say it's not AI.

[FZ1]

A* search counts as academic AI. If-then statements do not.

[Der_Einzige]

Such a wrong take. Decision trees are nothing more than if statements. Fancier versions of those (random forests) are consistently near SOTA algorithms like neural networks (and beat them on tabular datasets frequently). AI can be just a shit ton of if/else statements.

[jfkebwjsbx]

If-then statements are definitely AI, and A* is also a bunch of if-then statements.

[ncallaway]

It's kinda cool that we get to interact with the final arbiter of the bounds of the term "academic AI".

[dgb23]

AI is much broader than ML. In gaming specifically you typically label behavior of dynamic, life-like entities as AI. Behavior trees, state-machines, path-finding and so on. A typical example: the ghosts in pacman.

[klyrs]

My first game was a homework assignment: write a 1-player pong game (like arkanoid without the bricks). I juiced it up a bit by adding a second "AI" player with a very rudimentary ball-following algorithm.

  if (ai.x - pad > ball.x)
     move_ai(-max_speed);
  else if (ai.x + pad < ball.x)
     move_ai(max_speed);

... or something like that, with some bounds checking and some logic for smaller moves. Bottom line, "just an if/then statement" can be sufficient to make a playable game with a challenging (or even unbeatable) computer player. Tic tac toe is another example of this.

[didibus]

> There isn't any AI or ML.

You seem to acknowledge a distinction between ML and AI, but I'm not able to understand in your categorization, what would AI be?

[FZ1]

A* search, any kind of heuristic estimation, learning, or simulated reasoning. All of those things would count.

We don't need mathematical optimization to call it "AI", but there SHOULD be more than a simple if-then.

At least show me that you're path-finding. That's not even being done here - this is just path-following.

"I leave a trail, you follow it." Explain to me how that qualifies as AI. Simple BFS/DFS achieves a lot more than this - which is considered by most to not even really be AI.

[Eldt]

Because it's not the same kind of "AI" that you seem to be familiar with

[bkovacev]

Could you provide a list of games that have "actual" AI/ML baked in for their NPCs? I'm curious as I haven't heard of them before.

[meheleventyone]

There’s a few, most notably this racing game used neural nets before they were cool: http://www.ai-junkie.com/misc/hannan/hannan.html

A lot of modern work has focused on procedural animation versus behaviour but the two are quite intertwined: https://youtu.be/JZKaqQKcAnw