As AI shifts away from huge datasets and brute-force computing approaches, this will create incentives and opportunities for combining with alternative approaches such as neuromorphic computing (chips mimicking neural networks directly into the hardware; see 1.3) and biocomputing (information processing based on biochemical components such as nerve cells, DNA or metabolic processes in the cell and which takes advantage of naturally occurring stochasticity and evolutionary processes to manipulate information; see 1.4).
Hybrid architecture combining these approaches with traditional machine learning might yield unexpected advantages. Additionally, running machine learning algorithms on quantum computers (see 1.2) might prove useful for problems with small data sets and dealing with quantum objects, such as simulating chemical reactions or new materials.18,19 Quantum computing researchers are already working in collaboration with machine learning experts to assess the potential of leveraging a partnership between these fields, and the first demonstrations of a quantum speedup for machine learning algorithms have increased the hope that the field could have synergies with AI.20
