MAMEWorld Forums - EmuChat

> I as well have thought about this!
> Especially in the typing monkeys field where it takes a person to decide if it's a 1
> or a 0.
> But then I think about if I actually trust AI and I personally don't!
> So

It really depends what you consider to be 'AI' as it's a term that just seems to have been abused of late.

You're definitely not the only one asking this question, something similar was asked in a recent PR in the context of figuring out dipswitches etc.

You could probably 'train' something with a database of common assembly routines (standard C library, functions from common game engines) as to be provided with a better disassembly, and if you had a set of tools that converted things to an intermediate langauge for comparison, maybe apply that to a larger range of game sources, but are tools and pattern matching really 'AI'? Their effectiveness, even then, relies on something alredy being documented.

In terms of code output, most 'AI' is absolute garbage as it really lacks context. In terms of art, it all still depends on real creative input somewhere as it doesn't really create something from nothing, and as 'AI' generated works get fed back into the databases all you're doing is mixing the same paint over and over again. It's bad enough that a lot of game dev (especially Unity) is done today by copy+pasting bad online examples, with no understanding of the problems inherent in a lot of that widely shared code.

I don't think we're going to get to the point where you're going to be able to feed a random unemulated ROM on an unknown platform into an AI engine and have it create an emulation of the system; even the manual process involves a lot of trial and error, leaps of faith, exploring avenues and abandoning them when they start to become too convoluted. There simply isn't enough information/evidence in most cases to draw conclusions without trying to get into the head of whoever designed the chip, right down to considering when it was made.

Better tools are always handy and some of the recent ones that do a decent job of converting assembly into pseudo C code have really helped out with emulation (Windy made extensive use of such when studying Namco System 10 IIRC) but trying to call everything that helps out 'AI' is not really a fair reflection on things.

For the most part the AI we have is just a way of stealing work other people have arleady done, mashing it together, and calling it your own; it's not really anything new, we just have bigger databases and more computational power. Most of it is still as dumb as the online translation engines from the 90s.