Current assistive technologies, including those using artificial intelligence principles, cannot be applied to a large class of problems due to their need for well-understood and fully defined problems. In drug design, for example, scientists must iteratively adjust complex problem specifications until the tools produce the desired output. Instead, we develop an AI-assistant system specifically to handle such problems by reasoning and anticipating interactions with the user in real-time. For this, we develop user models that approximate human reasoning in complex decision-making problems based on cognitive science and, then, design an AI that can utilize these models. As a result, our work does not only attempt to infer the problem from interactions with the user but actively provides suggestions, such as drug designs, in order to maximize collaboration in the long run. The system is applied to drug design and other scientific discovery tasks to truly assist human users.

2024

2026