Incredible New AI Outperforms Humans In 57 Atari Games

Incredible New AI Outperforms Humans In 57 Atari Games

Google’s DeepMind is developing a new AI agent that performs at a superhuman expertise level in 57 Atari games.



You Are Reading :Incredible New AI Outperforms Humans In 57 Atari Games

Incredible New AI Outperforms Humans In 57 Atari Games

A new AI is under development by Google’s artificial intelligence company DeepMind, and it’s outperforming humans in 57 Atari games.

DeepMind has been attempting to create an algorithm that learns by itself and demonstrate humanlike performance in a wide variety of tasks. Enter Agent57 – a learning algorithm and AI agent that’s blowing people’s socks off.

As observed by some scientists this year, Agent57 learned how to play 57 Atari games on a superhuman level. It literally played all 57 games at a more superior level than the recorded human benchmark of performance within said games.

Incredible New AI Outperforms Humans In 57 Atari Games

In the the field of AI, an agent is generally deemed intelligent if it performs sufficiently well on a sufficiently wide range of tasks. These 57 Atari games are used widely to test the intelligence of prospective AI agents because they pose a nice wide range of problem-solving, learning, and exploration tasks – the perfect playground for AI.

What’s fascinating is that Agent57 started out performing awfully in the games, showing no signs of intelligence. But after only 120 minutes of playing Breakout, it starts playing like an expert that matches the skill level of an adept human player.

After a further 120 minutes, however, Agent57 reaches superhuman performance levels. It manages to work out that the best way to play and beat the game is to dig a tunnel through the colored bricks and subsequently hit them from behind.

See also  10 Most Thought Provoking Quotes from Saving Private Ryan



But what about the more challenging games? It’s been noted that previous AI agents tend to fail at getting the hang of Atari’s Montezuma’s Revenge, Pitfall, Solaris, and Skiing, because these require more longterm planning. Researchers have managed to enhance the performance of agents such as Agent57 in these games by infusing them with the very human capacity for curiosity. They make it all sound so easy.

Although there is still work to be done, this is being called a true breakthrough in AI research.

Source: YouTube

Link Source : https://www.thegamer.com/incredible-new-ai-outperforms-humans-57-atari-games/


Leave a Reply

Your email address will not be published. Required fields are marked *