That's a very similar idea to our startup. Except we're not using brute force evolution algorithms. Which I think is a bad idea, for the same reason researchers have found neural networks to be generally a bad approach to machine learning. The problem is the same, while it might be mathematically viable to code a system that will brute force it's way to learn if a great game is. That takes too long to most commercial applications. Specially when it's so easy to "cheat" and pre-input some values you already learned yourself.
In short, why would you write a code that takes 10 years to learn something that you could teach it yourself in a few minutes?
Brute force evolution algorithms are a fun and interesting thing to do research on. But that's it. The only practical application for it, is to research the algorithms itself to better understand how they work! But I'm sure the author of Angelina knows this, as his main goal does seem to be better understand how the algorithms work. That's fine, just don't expect many awesome commercially viable games from it. If that was the goal, I'd recommend "cheating" your own knowledge and the knowledge of other players to train your machine learning algorithms, similar to a recommendation algorithm. That's what we're hoping to achieve ourselves.
Far from my area of expertise, but I always considered the reason to try the brute force approach was to uncover the non-obvious things that "everyone knows won't work".
You're right. That's why these algorithms are often said to be effective, but not efficient. Meaning they do what they're supposed to, but the costs (time and money) might make it not viable for commercial use. Specially when there are known solutions that performs better.
Brute force evolution algorithms are fun to observe and do research on. But at the end of the day, it's the more efficient manually trained algorithms that end up making their way to commercial use.
Evolution is a pretty cool tool. That said, it has its limitations! Inventing brand new mechanics seems to be a particularly tricky problem, for instance. Look at something as left-field as Antichamber (from this year's IGF). That takes some creativity to come up with, I think.
Still, that's why it's research! Wouldn't be fun without feeling a bit impossible.
This is not evolution, it's exactly the opposite. Take the cake sample. You have a goal (the cake) and let the AI system find a way to reach that goal. Evolution has no goals at all.
Evolution is systematic change in allele frequencies.
The combination of natural selection and sexual reproduction tends to produce organisms which "fit to" their environments (hence "fitness"). It does so reliably enough that it might as well be the goal.
"AI systems" are not like people and they don't need people to be around to work, they do not function on the basis of ongoing intelligence from people. If the people die and there is nobody around to care, the same mechanisms do what they always did according to the natural law of the universe. They are just mechanisms which get things done. And so is evolution. And it is irrelevant that no person designed evolution to be that way.
"Evolution" is a touchy word, because there are so many idiots who will deny it exists or who will distort it horribly.
It's a shame that a useful word is less useful now - purists want to make sure that biological evolution is never misrepresented. That's fair enough. It's a simple concept and lots of people (even the ones who accept evolution) sometimes make mistakes.
Personally, I think the term "computational evolution" is clear enough.
Not sure current technology is capable of yielding real creativity. Perhaps as we enter the era of quantum computing.
But fun level design and infinite gameplay generated using AI/ML techniques are a welcome addition to casual games.
Well, 'real creativity' is a tough thing to pin down or talk about at the best of times. I think it can be achieved with conventional computing, but I think many people would reject it as creative even when faced with it.
As you say, though, creative or not I'm just happy to be producing software that can make fun(-ish) games!
In short, why would you write a code that takes 10 years to learn something that you could teach it yourself in a few minutes?
Brute force evolution algorithms are a fun and interesting thing to do research on. But that's it. The only practical application for it, is to research the algorithms itself to better understand how they work! But I'm sure the author of Angelina knows this, as his main goal does seem to be better understand how the algorithms work. That's fine, just don't expect many awesome commercially viable games from it. If that was the goal, I'd recommend "cheating" your own knowledge and the knowledge of other players to train your machine learning algorithms, similar to a recommendation algorithm. That's what we're hoping to achieve ourselves.