An AI managed to beat a retrogame

You know, the video games of the past are not easy at all.
The difficulty of the video games of the XNUMX's cabinets was all due to the fact that it was necessary to collect coins from all those who spent whole afternoons in the arcades. Needless to go around it, despite the undoubted quality, the cabin cruisers were a real money machine.

AI beat Atari's Pitfall

In the following decade, in the nineties, the concept of video game has evolved, becoming ever more homely, lowering the difficulty in order to be more accessible to all.

Over the years, retrogames have been beaten by various industry enthusiasts: speedrunner, record breaker and so on and so forth. However, thanks to science, even artificial intelligence can now beat the old classics.

Scientists fromUber AI Labs they used AI to complete the iconic Atari platformer Pitfall.

The methodology behind this is quite interesting: the researchers have created algorithms that allow theAI to remember previously tried areas of the platform, using that information to push her to new places.

Previous methods, particularly one called intrinsic motivation, rewarded AI for finding new areas, but it tended to forget about areas discovered in the past. This problem is known as "detachment", but thanks to new research in artificial intelligence, the AI ​​has not only been able to complete Pitfall, but also Freeway e Montezuma's Revenge.

An AI that beats classic arcade games may seem seemingly useless, but it has various practical implications. The goal is to create an articular intelligence that is better in navigating real environments (such as, for example, the use of robots in the search for survivors of an earthquake), and to do this it is necessary to “train” it in digital environments.
And the difficult 80s platformers are the right compromise.

The aforementioned case can be a major boost for advances in artificial intelligence and is yet another case in science against the world of video games.
A very interesting result that goes in the long tail of what has been the case with Open AI where a sufficiently developed artificial intelligence was able to face and defeat professional players of Dota 2