Nowadays, artificial intelligence is in use in variety of fields. However, it needs to have some fun as well. AI entered gamer society years ago with different types of board games. I would say that it made a solid statement in 1997 when Deep Blue (chess playing computer designed by IBM) defeated world champion Garry Kasparov.
As gaming trends change, computers change their gaming preferences with it. AI wants to hit video games now. Traditional approach for designing AI is to set a goal and define some rules. Nevertheless, Google’s DeepMind and Open AI found another way how it could work. Scientists gave a machine one of the most human like features you can come up with – Curiosity.
According to Common Sense Media “Curiosity is having a strong desire to learn or know something. Folks who are curious often don’t “need” the information they inquire about. They seek answers to their questions for the sake of gaining knowledge.” Not a character for a machine, right?!
In the research of DeepMind, they had to create something comparable to brains “pleasure chemicals”, what drives curiosity in humans, so scientists added some hidden rewards, what was motivating AI to explore system. For experiment they used Atari game Montezuma’s Revenge. Unlike traditional bots, used in video games, AI didn’t have any information regarding the game and was exploring it as human would do, by looking at screen, pushing buttons and changing actions based on results. According to research “After 50 million frames, the agent using exploration bonuses has seen a total of 15 rooms, while the no-bonus agent has seen two.”
In the research of Open AI, scientists tried several different games with the same approach. They wanted to know how far can curiosity lead and if machine can suffer with the same problems that curious people do. Scientists paired two curious Pong player machines against one another. They quickly abandoned the game and just got curious how many volleys they could achieve. In addition, Scientists tried to add virtual TV that changed channels on an action to one of the video games, that also was a huge distraction for AI.
References:
Larry Greenemeier. (2017, June 01). 20 Years after Deep Blue: How AI Has Advanced Since Conquering Chess. Retrieved from https://www.scientificamerican.com/article/20-years-after-deep-blue-how-ai-has-advanced-since-conquering-chess/
Common Sense. (2018). What is curiosity?. Retrieved from https://www.commonsensemedia.org/character-strengths-and-life-skills/what-is-curiosity
James Vincent. (2016, June 09). Watch Google’s AI master the infamously difficult Atari game Montezuma’s Revenge. Retrieved from https://www.theverge.com/2016/6/9/11893002/google-ai-deepmind-atari-montezumas-revenge
Tristan Greene. (2018, August 23). Researchers gave AI curiosity and it played video games all day. Retrieved from https://thenextweb.com/artificial-intelligence/2018/08/23/researchers-gave-ai-curiosity-and-it-played-video-games-all-day/
