Artificial intelligence (AI) is not only mysterious; it is also a lone performer. The struggle which researchers face to teach AI how to play chess or to drive is not as great as teaching them the art of teamwork. Working collaboratively may be intuitive to human. Most of us do not experience difficulties in forecasting the verbal or physical behaviors of our counterparts or demonstrating empathy, elements that are crucial for working on a team. However, these seemingly effortless gestures, add a whole new level of challenge for AI. 

Researchers are now teaching AI bots to work together as a team, using a video game that mimics Quake III Arena. Teams of AI will have to navigate around a 3D map, to fetch a flag from the opponent’s base and bring it back to theirs. In the meantime, they will also have to fire laser at their enemies, to prevent them from taking back the flag. The team with the greatest number of flag captures after five minutes will be considered as the winner. 

30 different AI bots were trained using neural networks that learn from data by adjusting the strength of connections between artificial neurons. These AI were placed in opposing team in a series of matches against each other on randomly generated maps. The bots do not show any meaningful moves at the beginning of the match. Nevertheless, when their actions do lead to a score, a process known as reinforcement learning, will prime the connections to strengthen that particular behavior. 

AI is able to learn the meaning of joint effort

After 450,000 games, researchers named the best performing bot “For the Win” (FTW). They realized, through playing with non-FTW, human, and the game’s in-built bots, that groups of FTWs will outperform others constantly. Especially in those when they were paired with human. FTW not only learn the meaning of joint effort, they even developed classic cooperative strategies, such as covering for their fellow teammates or exploiting a bug to give teammates a speed boost. 

Researchers were amazed by the emergence of higher-level behaviors, traits which they can spot in human players. Although it’s still far from adapting it to the real World, researchers foresee self-driving vehicles could be safer if the driving systems are able to coordinate with one another to avoid accidents. Likewise, the same kind of partnership between FTWs and human players can also be formed between robotic surgical assistants and surgeons, to ensure medical procedures are safer and more efficient.

In spite of the initial success, researchers noted that more related studies have to be carried out to ascertain AI’s capability of learning or demonstrating teamwork in different settings. Workflow integration has troubled the healthcare community for quite a while, as they are concerned about adopting an AI solution which will hinder them from working efficiently or better care. Knowing that AI may one day collaborate with them to guarantee an enhanced workflow will be great. 

A promising workflow integration? 

However, is it really that simple? According to speakers from the previous AIMed Breakfast Briefings, it may be not. Some of them feel that a true workflow integration needs to be real-time and smart. Presently, human still dominates the workflow orchestration and technologies require human to remember to do the right thing. Hence, if the correct command is not given, collaboration cannot be established. 

On the other hand, infrastructures vary across institutions and departments. Even if a healthcare professional is able to work seamlessly with an AI solution in one setting, it may not be carried forward to other areas, because they don’t happen to use the same platform or software. At the end of the day, it may even be the case that AI can only cooperate with human to a certain level and nothing more. It’s heartening to know AI can become a better team player, but there are still hurdles to overcome before it can team up with a human

Author Bio
synthetic gene empathy chinese artificial intelligence data medicine healthcare ai

Hazel Tang

A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.