AI and BGA

[lefalaf]

True, but most of the time when someone mentions AI in 2026, they mean generative AI. And the origin of this thread was about generative AI. Traditional AI approaches weren't problematic like generative AI is.

There were several threads in recent months about bots that involved LLMs in some fashion. I do not know how the specific TFMBot noted above was created or executed, whether it involved generative AI/LLMs at either point.

[BarnardsStar]

And the origin of this thread was about generative AI.

Mmmmm, no, it was not, not really. The first post was about an EU directive that is not limited to generative AI, and meant to ensure that people know when they are interacting with AIs. That could well apply to game-playing AIs, which as noted, are probably not going to be LLMs.

[Meeplelowda]

Please, do not conflate "LLMs" with "AI". Large Language Models are one specific case of AI, and they happen to be the example that many people are most familiar with, but there are lots of different types of AI that have absolutely nothing to do with LLMs, and whose training doesn't involve using material gathered from outside, let alone stealing content.

If AI was used to create playing bots, they would be unlikely to use LLMs. Their training would involve playing the game many times over and seeing what choices performed well vs. what didn't, not trawling the internet for strategy documents.

Not only are LLMs bad at playing games now, they probably always will be. It's a task ill-suited for them.

Now if we want to complain about BGA using barely edited LLM output for game announcements, I can get behind that.

[nik592]

I would love an intelligent AI system to play games against when none of my friends are available, why would that AI system start by stealing art?

Well, not art in this case, this aspect would have started by being trained, in part, on stolen software code. And take potential work/income from software engineers. You people love to create strawmen to knock down (comparing AI to a hammer, really?) and create disingenuous arguments.

Are you a software engineer? Because (speaking as one), the majority of software development companies are looking to leverage AI in their work, and software engineering is actually a field where it probably has a lot of use. The way it's being implemented at my company is not going to take potential work from engineers as the AI doesn't replace them, but allows them to work faster (and hopefully better, but that's still being analysed),

I get that in creative industries, generative AI is NOT a good thing. I'm also a crocheter and the amount of AI pictures of "crocheted" objects that aren't actually reproducible is infuriating. As are the many comments that inevitably follow either commenting how "amazing" the piece is (because it isn't real) or asking how to make it (which isn't possible). As Suoivax says, AI is a tool, you can use it for good things and bad. The nature of software engineering (specifically the reusability of code) makes it highly suitable to leverage AI for good. Creative industries? Well that's generally a bad one.

As to using AI to create opponents? Sure. But a good AI opponent would need review and maintenance by a human dev. So it's not as easy as telling the AI to create an opponent for a given game and then letting it go. AI can write the base for you, but you need a human to iron out the wrinkles. So I doubt you'll see that at BGA because they don't have the manpower. Devs could certainly use AI to help them code games too, but again, it still needs a human to review, correct and maintain. And the responsibility for what that code does should still rest with the human, not with the AI. Same as shoddy carpentry work rests with the carpenter, not their hammer.

[lefalaf]

And the origin of this thread was about generative AI.

Mmmmm, no, it was not, not really. The first post was about an EU directive that is not limited to generative AI, and meant to ensure that people know when they are interacting with AIs. That could well apply to game-playing AIs, which as noted, are probably not going to be LLMs.

Perhaps the actual act is more broad, but the summary paragraphs at the top of the linked page refer to generative heavily.

[lefalaf]

still needs a human to review, correct and maintain.

Retired software engineer, yes. And any good software engineer would prefer writing code they inherently understand to proofreading and reworking generated slop for accuracy, efficiency, maintainability, security, good architecture practices, etc. I doubt it's a net savings, long term.

From what I read, it's bosses chasing the shiny new object and mandating it. Like blockchain, or agile, or a new programming language.

Just like you cannot rely on it for crochet patterns to be accurate, you cannot rely on it for more complex things.

This reminded me of something I read earlier. In a domain you aren't an expert in, it can seem amazing, but if it is your area of expertise, you know better and can see the issues. It's predictive text at its core, with limited memory depth and persistence last I checked. Not any sort of true intelligence.

[FSKFSK]

Here's an AI task that would impress me: An AI that could play all the games on BGA at an expert level, including new ones, without being trained specifically for that game. I.e., Stockfish would be the top rated chess player, which doesn't count, because Stockfish can only play one game.

[nik592]

Retired software engineer, yes. And any good software engineer would prefer writing code they inherently understand to proofreading and reworking generated slop for accuracy, efficiency, maintainability, security, good architecture practices, etc. I doubt it's a net savings, long term.

That's not what people are finding. I've talked to people in other (software dev) companies too and those embracing AI are finding it useful. Not all AI is built the same, and some of it is extremely good at writing code. And if it's writing slop, then it's probably not an agent worth pursuing - no one says once you start with one you have to keep using it if it isn't useful.

More than that, if you think that the job of a software developer doesn't involve reading and reworking less-than-good code, then I question how long you worked in software engineering. Even before the advent of AI, that's most of what we were doing. Any large piece of software is not written and maintained by a single person (even in small teams, people eventually move on from jobs and the team makeup changes) and you have to read and make sense of other people's code (and fix it, or adapt it to handle new tasks and requirements).

I will say that due to my experiences with generative AI in the creative space, I was skeptical about what AI could do in my work and how useful it would be. But the (free) AI agents I'd used outside of my job tend to be less specialised and less, well, good. I've had to adjust my opinions and admit that good AI agents can probably do some useful things for us and I'm already seeing the benefits in my work. So given that you're apparently no longer in the industry, I'm not sure you're an expert voice on this.

That said, that you want to lump "blockchain", "agile", "new programming languages" in with "shiny new objects" (which is clearly meant to suggest they're not worthwhile "chasing") does make me wonder if you're open to new things generally. There are pros and cons to all of those things and the point is about considering those, not simply running on, well, what we do now works so let's just avoid anything new.

[kostinen]

As a software engineer myself, I can confirm that AI has completely changed the way I work. In a nutshell, it allows me to achieve a functional result much more quickly, leaving me much more time to work on improving the code I’ve produced. Until now, there was always a trade-off between speed and quality, which often led to either frustration (code produced quickly but not clean) or delays (clean code but slow to produce). Today, AI allows me to quickly produce the code I’ve always dreamed of writing.

However, I’m not as confident as Nik when it comes to AI replacing our profession. A task can be completed at least twice as fast (and we’re only just getting started), which means companies will need fewer and fewer developers to do the same amount of work.

Let me give an example. Suppose I’m a farmer and I harvest my wheat with a scythe or a sickle. If tomorrow I use a combine harvester, I’ll work much faster and be able to produce much more. The result is increased competition with other farmers, and ultimately the disappearance of some of us.

One solution is to pair productivity gains with a reduction in working hours.

To conclude on a different but important note, some productivity gains come at the expense of—or to the benefit of—other aspects, such as job satisfaction, the physical demands of work, or the resulting environmental costs (which may be borne in the long term by humanity as a whole).

[tbhp]

Here's an AI task that would impress me: An AI that could play all the games on BGA at an expert level, including new ones, without being trained specifically for that game. I.e., Stockfish would be the top rated chess player, which doesn't count, because Stockfish can only play one game.

I don't think that would be too hard to do. After all, this isn't some surgical task where you can't afford any mistake. The AI's understanding of the game could be broken in many ways, it wouldn't matter as long as it still manages to win often enough to reach expert level.

There doesn't even need to be human maintenance like people said: if it reaches the desired ELO, then you know it works.

There might be some games where the algorithm wouldn't be able to do that by itself, in which case you may want to refine it. But the majority of games are not all that complex to play. A self learning AI would probably pick up winning patterns and get good at the game that way.