More human-like characters appeared in a list of 30 similar things. Although all these people started with the same personality and the same goal—to create a better village and protect the community from being attacked by other game creatures, they simply created special roles in their community without being forced. They held different roles as architects, protectors, traders, and explorers. As the agent began to play professionally, his sporting activities began to reflect his new role. For example, an artist spent a lot of time picking flowers, farmers gathered seeds and guards built many fences.
“We were surprised to see how you put it [in] the right type of brain, they can have a predictable behavior,” says Yang. “This is something we hope people will have, but don’t expect machines to have.”
Yang’s team also tested whether agents would follow community rules. He introduced a state of the art tax system and allowed supporters to vote on changes to the sports tax system. Supporters who were encouraged to be pro- or anti-tax were able to influence the behavior of other supporters around them, so they could vote to lower or raise taxes depending on who they contacted.
The group got bigger and bigger, pushing the number of helpers in each test to the limit that the Minecraft server could handle without glitching, up to 1000 at a time in some cases. In one of Altera’s 500-agent experiments, they observed how agents simply came and spread cultural norms (such as a love of magic, or an interest in things related to nature) among their fellow agents. The group also planted a small group of supporters to try to spread (parody) the religion, Pastafarianism, around the different towns and rural areas that make up the world of the game, and see how these Pastafarian priests convert many of the supporters who joined them. . The converts went on to spread Pastafarianism (the term for the Church of the Flying Spaghetti Monster) to nearby sports towns.
The way the agents act may seem like life, but really all they are doing is repeating what LLMshave learned while training on human-generated content on the Internet. “The bottom line is that LLMs have a fairly advanced model of human behavior [to] show these people’s values,” says Altera co-founder Andrew Ahn.
ALTERNATE
In other words, data makes them more like human behavior, but they are not “alive”.
But Yang has bigger plans. Altera plans to expand into Roblox next, but Yang hopes it will eventually expand into a global game. Ultimately, the goal is a world where people not only play alongside AI characters, but also interact with them in their daily lives. His dream is to create a large number of “digital people” who really care about us and will work with us to help us solve problems, and keep us safe. “We want to create agents that can really love people (like dogs love people, for example),” he says.
This idea – that AI can love us – is very controversial in the field, many experts argue that it is not possible to reproduce the thoughts in machines using modern methods. For example, AI veteran Julian Togelius, who runs the game-testing company Modl.ai, says he likes Altera’s work, especially because it allows us to study human behavior through simulation.
Leave a Reply