• In a Nutshell - Quantum Computers
    60 replies, posted
Since we're talking about AIs, lemme take a chance to show off Learnfun & Playfun, whom the creator of has made a video series which is both semi-entertaining and demystifies how real AIs work. Learnfun & Playfun are two AIs that work together to "learn" how to play NES games. The basic idea is that Learnfun "observes" a human playing the game, then decides what the objective of the game is (it determines "winning" by certain data values in memory going up - e.g. advancing to another level, or your score going up). It passes on this information to playfun, which then plays the game by running a simulation of what would happen in the next frame for every combination of button presses it could try in the current frame, then deciding the best button presses to use. It then repeats this process for an arbitrary amount of gameplay time. It also takes advantage of an evolutionary algorithm; playfun generally gets better at a game every time it plays (however, after a while, it seems to get "as good as it can be"). [video=youtube;xOCurBYI_gY]https://www.youtube.com/watch?v=xOCurBYI_gY[/video] [URL="https://www.youtube.com/watch?v=YGJHR9Ovszs"]Part 2[/URL], [URL="https://www.youtube.com/watch?v=Q-WgQcnessA"]part 3[/URL], the [URL="https://www.cs.cmu.edu/~tom7/mario/mario.pdf"]research paper[/URL], and the [URL="http://sourceforge.net/p/tom7misc/svn/HEAD/tree/trunk/tasbot/"]source code[/URL] if you're interested in more. Notice how even though Playfun uses an evolutionary algorithm, and its game-playing behavior changes from one play-through to the next, it is still confined to its original instructions - press buttons in a way that makes values go up. Playfun can never do anything other than this, because that's all he was programmed to do. Could a program that changes its own program to better adapt to its environment be made? That's within the realm of possibilities, yes. Is it feasible for such a thing to exist in an even semi-coherent way? Not really. Is it a sound strategy for making a coherent Artificial Intelligence meant to emulate humans? Again, not really. To relate this back to quantum computing, could Learnfun & Playfun benefit from being on a quantum computer (assuming it's properly ported & optimized)? Currently, the playfun algorithm is too slow to run in real time ("real time" being the framerate the game was intended to run at) - the videos you see are recorded by playfun and played back at a much higher framerate than it actually runs at. With a fast enough quantum computer, Playfun could potentially simulate its future button presses and decide on the best combination much faster than a conventional computer - perhaps even allowing it to run at real-time speeds, or close. However, this does not make Playfun more competent at playing games - since Playfun has infinite time to simulate button presses anyway, it gains no real advantage from quantum computing. Other AIs that [I]do[/I] need to run in real-time may benefit - e.g. you'd want an AI that flies a plane or drives a car to respond to data almost immediately. A quantum computer may allow more data to be processed, analyzed, and responded to much more quickly, thus allowing the AI to account for more variables in its simulation than it would otherwise.
Sorry, you need to Log In to post a reply to this thread.