Artificial intelligence for skeptics
By: Luis Amaral
From: Northwestern University, USA
At: Building C1, room 1.4.14
[2022-11-04]
Promises around the capabilities of artificial intelligence (AI) are ever more present, loud, and inescapable.  All this ‘noise’ makes it quite difficult to decide what hype and what is reality. I personally believe that AI, broadly defined, has the potencial to strongly impact that way in which science is conducted.  However, it is critical to review claims by their feasibility and reject those claims that are clearly absurd. In this presentation, I will give a short overview of the history of AI, including its summers and winters, and the ever present financial pulls that reward unfounded claims. I will discuss then how the scientific methods and physical reasoning can help make sense of claims and guide new developments.  In particular, I will discuss how entropy can inform feature selection and how the physics of disordered systems can provide insight into the limitations of optimization methods frequently used.
