For a long time, games have been a way to measure the progress of AI technology, and show the “intelligence” of algorithms by competing with top human players.
Julian Togelius, Associate Professor of New York University’s Tandon School of Engineering, said: “Video games provide the best intelligence test we have. Video games are an excellent way to train AI algorithms because they gradually make the human mind experience increasingly difficult challenges.”
Video games are designed to challenge human thinking by allowing players to focus on elements such as overcoming obstacles and following subsequent stories. This makes video games an ideal environment for training AI. The structural framework, repetitive training and ability enhancement provided by digital games contributes to the learning of AI algorithms.
In fact, training AI through games is not new. OpenAI has trained AI through games a few years ago. For this reason, they have launched a massively multi-player online game called Neural MMO. All “players” in this game are AI.
In the game setting, the resources on the entire map are limited, and there is food. AI is divided into different races. According to the race, there are three modes of melee, ranged, and spell attack.
Their only goal in this game is to live to the end. AI must reasonably choose an environment suitable for its development and compete with other AIs for limited food resources.
Unlike human players, in the “cognition” of AI, they do not know that they can be “reborn” after a game is over. AI believes that it has only one life, and will firmly execute the instruction of live to the end according to its own logic.
When the number of AI players entering the game increases, AIs will also make decisions, and spontaneously open up the corners of the map to avoid disputes, and every decision made by AI will affect the decision-making choices of other AIs to a certain extent. So the situation in the game is changing rapidly.
Simply put, this game is an AI “Colosseum”, and the AI that survived to the end must have the strongest adaptability and learning ability. OpenAI is such a continuous game to enhance the adaptability and learning ability of AI.
In the long process of biological evolution, the surviving animals have evolved their own characteristics. Human beings are among the best, and they have evolved the characteristics and lifestyles that are most suitable for living in this world. And with the ultimate goal of “human beings”, the direction of AI evolution is naturally the same, and it is more of an evolution in “thinking way”.
Putting AI in the game world for training is a logical choice for research, and it seems to be a suitable choice. Whether it is an AI world or a real world, such activities have a fundamental point: prepare AI to enter the biggest game — Reality.
Founded in 2011, Datatang is a professional artificial intelligence data service provider and committed to providing high-quality training data and data services for global AI companies. Relying on own data resources, technical advantages and intensive data processing experiences, Datatang provides data services to 1000+ companies and institutions worldwide. Datatang entered Chinese stock market (NEEQ: 831428) in 2014 and became the first listed company in China’s artificial intelligence data service industry.
If you need data services, please feel free to contact us: firstname.lastname@example.org