Words matter. When we humans think of a word like “apple”, we get many associations. We think about the appearance of the apple, the sound when we bite into the apple, the texture, and the taste. We also understand that an apple can be sliced, but still eaten. This is today impossible for an AI to understand, which Professor Amy Loutfi wants to change.
– The goal is to contribute to the basic research and the building blocks in AI that are needed to build the intelligent machines of the future that work together with humans, says Amy Loutfi.
Seeing, perceiving, and interpreting one’s surroundings is a challenge for an artificial system. In recent years, research in artificial intelligence has succeeded in developing ways to use AI to, for example, add texts to images, answer questions about images, and also generate images based on a text description.
However, today’s robots and intelligent machines do not have a deeper understanding of how an object can be changed and used.
– It is extremely important that the robots of the future understand what we mean with our words and that they can distinguish between possible and impossible instructions, for example: “Pour coffee into the cup” compared to “Pour the cup into coffee”, says Amy Loutfi.
This is exactly what the new research project, led by Amy Loutfi and funded by SEK 3.7 million by the Swedish Research Council, will try to achieve.
The purpose is to solve the challenge with so-called symbol anchoring, i.e., the ability to link language to experiences of the external world. The researchers want to make it possible for artificial systems to learn what functions objects have, but also to be able to reason about the functions themselves. This will be done by the researchers developing new AI models that combine machine learning with logical reasoning in order to provide a deeper understanding of the meaning of words.
– In this project, we also want to understand how we can help a robot to anchor the language to perception. How important is it that the robot itself is allowed to interact with its surroundings? Can it be enough to let the robot observe others interacting or to just simulate an interaction?
– This project gives us opportunities to develop basic theories about artificial intelligence and, at the same time, to have a bearing on how the next generation of robots is developed in, for example, industry, says Amy Loutfi.