SAN FRANCISCO, CA – Quantum computers are set to have a significant impact on the acceleration of AI capabilities, Dario Gil, IBM VP of AI and commercial quantum computing, told the MIT Technology Review’s EmTech Digital conference today in San Francisco.
While classical computing paradigms have led to incredible historical advances in technology over the past century, many commentators warn that Moore’s Law – the principle that chipsets become twice as powerful and twice as small each year – is reaching its end as chips become infinitesimally small. Whereas classical computers store information in bits of 1s and 0s, quantum computers use qubits, which at the sub-atomic level can exist in multiple states of 1 and 0 at the same time.
Crucially, qubits are not limited by physical restrictions in the way that bits are thanks to quantum entanglement, meaning that adding extra qubits to a quantum computer can lead to exponential increases in computing power.
Onstage at EmTech Digital, Gil told the audience that he believes AI and machine learning could significantly benefit from these developments. He presented a simple classification experiment using machine learning to group data (here, dots with complementary colors). IBM’s team ran the task on a quantum machine without entangling the qubits, with an error rate of 5%. In the second iteration of the experiment, the qubits were entangled, producing to an error rate of 2.5%.
Gil suggested that, as quantum computers improve at harnessing and entangling qubits, they’ll also get better at tackling machine learning challenges. “It’s a really great time for the AI community to start exploring this future,” he told the audience, explaining that quantum computers could give certain types of AI problems an edge. Gil called for caution, however, about the prospect of ‘quantum AI networks’, which remain no match for the neural and deep learning networks running on top-end conventional computers today.