The research showed that an agent trained on many games performed better than one trained on a single game, and an agent trained in all but one game performed nearly as well on the unseen game. This ability to function in new environments highlights SIMA’s ability to generalize beyond its training. However, more research is needed for SIMA to perform at human levels in both seen and unseen games. The ultimate goal is to develop more general AI systems and agents that can understand and safely carry out a wide range of tasks in a way that is helpful to people online and in the real world.
Key takeaways:
- Google DeepMind has introduced a new AI agent called Scalable Instructable Multiworld Agent (SIMA) that can follow natural-language instructions to perform tasks in various video game settings.
- SIMA was trained on a variety of video games in collaboration with game developers, marking the first time an agent has demonstrated understanding of a broad range of gaming worlds and can follow instructions within them.
- The AI agent requires only two inputs: the images on screen and simple, natural-language instructions provided by the user, and can potentially interact with any virtual environment.
- SIMA's performance in unseen games and its ability to generalize beyond its training shows promise for the development of more general AI systems that can understand and safely carry out a wide range of tasks.