If you hang out on the web you will probably know QWOP , a very popular 2008 browser game inspired by the much loved Track & Field of the 80s, featuring however from a key element, that is the very high difficulty .
The purpose is very simple: to make the athlete run using the Q, W, O and P buttons, each of which moves the runner's calves and thighs. The problem is that the physics engine animates it as if it were some kind of limp rag doll, which makes it extremely complicated, as mastering the timing and cadence of keystrokes is frustrating to say the least. To say, taking even one step is not easy at all. What if an artificial intelligence tries to play?
Liao's second attempt , on the other hand, involved recording video of the games of some of the strongest users, so that the AI could improve its performance. The result has had positive effects, because in fact the times have improved, but the computer has proved unable to assimilate a special technique used by the best QWOP players, which involves an upward swing of the legs to generate additional momentum.
Eventually, Liao contacted a veteran gamer known as Kurodo ( @ cld_el on Twitter), one of the best QWOP speed runners in the world, to record 50 videos of himself playing, so as to feed them to the artificial intelligence. Despite all the efforts and hours spent training artificial intelligence, the computer failed to reach the top 3, but only the top 10, with a time of 1 minute and 8 seconds, against only 48 seconds of the first position. .
However, the programmer has not lost hope and, on the contrary, is confident of the fact that an even more in-depth training and a different reward system (it is necessary to study a more effective way in making people understand AI of having performed a move correctly) can cause the computer to set a world record. What then, since it is an AI, the record may never be officially recognized, but you want to put the pleasure of having succeeded?