Convergence: Artificial Intelligence and Quantum Computing

CHAPTER 5The Challenge of Quantum Noise

Philip Johnson, Associate Professor in the Department of Physics, American University

Quantum computing and artificial intelligence share an important characteristic—the ability to produce results not available by other methods and that are difficult to independently verify. Explainability and verification aren't always hard. For example, factoring numbers into their composite primes is a problem with both explainable algorithms and where we don't have to trust candidate solutions because we can easily check if they work. But complex learning algorithms trained and optimized over large datasets and used to solve problems of very high dimensionality are inherently hard to verify and explain. Yet for many problems, it is important to not only efficiently obtain a result but to understand why the result was produced and to verify its reliability. It is dangerous to simply trust complex solutions to complex problems.

The tendency to be black boxes arises separately for both artificial intelligence and quantum computers, but achieving explainability and verification will be significantly harder when they are combined as quantum AI systems for at least three reasons. First, the states of quantum AI systems live in exponentially large Hilbert spaces with fantastically high dimensionality such that, for even modest-sized quantum AI systems, we will simply never be able to directly sample more than an infinitesimal fraction of the system's state ...