Difficulty: Frontier

Acting to minimize surprise or free energy.

Perceived actions an environment allows.

AI capable of performing most intellectual tasks humans can.

System-level design for general intelligence.

Internal representation of environment layout.

Human-like understanding of physical behavior.

Interpreting human gestures.

Humans assist or override autonomous behavior.

Inferring human goals from behavior.

Modeling environment evolution in latent space.

Imagined future trajectories.

Awareness and regulation of internal processes.

AI limited to specific domains.

Controlling robots via language.

Understanding objects exist when unseen.

Ensuring robots do not harm humans.

Learning by minimizing prediction error.

Hard constraints preventing unsafe actions.

Internal representation of the agent itself.

Control shared between human and agent.

Human controlling robot remotely.

Learned model of environment dynamics.