Humans have a grand tradition of pitting their wits against artificial intelligence, and it usually doesn't end well: in 1997, IBM's Deep Blue defeated chess great Garry Kasparov, and in 2016, Go champion Lee Sedol lost to a computer named AlphaGo. Even the world of mahjong was rocked when Microsoft introduced Suphx in 2019, an AI capable of beating the world's top players.

Sadly, the AI's winning streak of beating flesh-and-blood opponents seems to continue unabated.

Researchers at the Singapore University of Science and Design (SUTD) have unveiled new software that can understand complex movement designs. The scientists trained the artificial intelligence using reinforcement learning (RL). Reinforcement learning means training through testing on people, rather than leaving them to their own devices on a data set (supervised learning).

The means of this testing was none other than the classic video game of the 90s, Street Fighter II: Championship Edition.

The research team first provided the computer with millions of initial motions in order to create a program to design the movements. The program then made several attempts to improve each motion randomly toward a specific goal. Finally, the team tweaked the character's movements and adjusted its strategy until it had mastered the movements to beat the built-in AI.

After a few days, when the practice wheels came off, the machine comfortably beat the human player and used its complex feedback to make further improvements. It was like a digital version of Veximoku from Dungeons & Dragons.

In addition to humiliating retro game fans, this AI has broader implications for kinetic science and other strategy games like chess. In a nutshell, this AI can observe and analyze human movement patterns and then replicate and ultimately surpass them. This is a relief.

The researchers pointed to a number of outstanding qualities as a measure of how successful the designed system was, including reasonable game etiquette, management of inaccurate information, ability to achieve certain game states, and the short amount of time used to defeat opponents.

"The more effective this technology becomes, the more potential applications it will open up, such as continued advances in competitive tasks that computers can facilitate for top players, such as poker, Starcraft, and Jeopardy," said study director Desmond Rourke, Associate Professor at SUTD, said. 'There could also be high-level, realistic competition to train professional players, discover new tactics, and make video games more interesting.'

Interested parties can read the full research paper here.