In a series of sophisticated 30-agent simulations, researchers observed the spontaneous emergence of distinctly human-like behaviors. Despite sharing a common starting point and objective – establishing an eco-friendly village and protecting their community from external threats – the brokers naturally evolved distinct specializations within the group, without any external encouragement or guidance. Once an agent began to specialize, its in-game actions began to reflect its evolving role. As a testament to dedication, an artist invested hours in meticulously choosing flowers, while farmers went the extra mile to collect seeds with care, and guards reinforced their vigilance by constructing additional fences.
According to Yang, when you approach a situation with the right mindset, individuals can develop surprisingly habitual patterns of behavior. “That’s what we expect from humans, but don’t rely on machines to possess.”
Researchers also explored whether brokers could adhere to community-wide regulations. With the introduction of a comprehensive tax framework, they empowered brokerages to shape the in-game taxation landscape by casting votes on proposed alterations. Brokers’ professionalism or anti-tax sentiments can influence their interactions with peers, ultimately shaping votes to adjust taxes based on those relationships.
The team successfully expanded their setup, accommodating a vast array of brokers in each simulation, effectively maxing out the Minecraft server’s capacity without encountering any glitches, reaching up to 1,000 simultaneous connections in certain instances? Within one of Altera’s 500 agent-based simulations, researchers observed how brokers spontaneously developed and propagated cultural memes – such as a fascination with prankster culture or environmental issues – among their peers. In an innovative move, the group dispatched a small contingent of brokers to spread the (parody) faith of Pastafarianism across various urban and rural locations within the in-game world, observing with interest as these monk-like figures converted the majority of brokers they encountered. The enthusiasts spread the doctrine of Pastafarianism, as coined by the Church of the Flying Spaghetti Monster, to nearby urban centers within the virtual realm.
While the AI-powered brokers’ behavior may seem strikingly realistic, their habits are actually rooted in patterns identified by large language models from human-generated knowledge, which are then translated by Altera’s system into contextually aware actions – think picking up an instrument or engaging with another entity. According to Andrew Ahn, co-founder of Altera, the key takeaway is that large language models possess a sophisticated understanding of human social dynamics, enabling them to accurately mirror complex human behaviors.
ALTERA
While artificial intelligence systems excel in emulating human behavior, they remain inanimate entities, devoid of life.
However Yang has grander plans. Altera aims to expand its presence on Roblox, but Yang envisions a future beyond virtual gaming realms entirely? Ultimately, his vision is to create a world where humans and artificial intelligence collaborate seamlessly in their daily routines. He aspires to develop a diverse array of indistinguishable digital entities that not only cater to our needs but also collaborate with us to resolve problems, while providing endless entertainment. “We should create artificial intelligence agents that genuinely connect with humans – much like dogs do when they form strong bonds with their owners.”
The notion that AI might harbour affection for humans is a contentious topic among experts, with numerous specialists disputing the feasibility of replicating emotions in machines using current technologies. Julian Togelius, a seasoned AI expert and founder of video game testing firm Modl.ai, commends Altera’s efforts for allowing researchers to study human behavior in simulated environments, he notes specifically the significant impact this has on our understanding of human habits.
