Beckman Fellow 2007-08

Eyal Amir

Computer Science


Many problems that appear easy for humans are difficult for computers. A human viewing a picture of a chair, for example, can determine easily whether the picture makes sense (e.g., the chair is on the ground and someone is sitting in it vs. the chair is hanging mid-sky). Humans have similar ease with natural-language understanding, spatial scenario understanding, and situations in which common sense is key (e.g., interpreting the intentions of laws and law-makers).

In contrast, research in Artificial Intelligence (AI) struggles with the complexity, amount of knowledge, and techniques required to achieve a general solution to such problems. Thus far, AI research has tried to separate elements of the problems and solve them in isolation. This approach, Professor Amir suggests, fails to address the underlying core computational issue.

Professor Amir aims to develop a foundational theory that relates fundamental AI problems with each other. His research project will modify the traditional model of computation (a Turing Machine) and introduce a human into the model. This will allow his group to classify and compare research problems according to the amount of effort required by both the computer and a human. They will also assign a range of difficulty for a set of traditional AI problems.

An immediate result of this research will be a set of algorithms that can distinguish between humans and machines. More generally, the research will serve to focus theoretical research in AI and lead to larger steps in the direction of human-level AI.