Computation is at a crossroads
Exciting new developments around the Internet of Things, Big Data, autonomous vehicles, and Artificial Intelligence highlight how our approach to computation must fundamentally change.
Present-day computers are enormously successful at a huge range of tasks that require fast number-crunching. They are excellent, and getting better, at complex things like image recognition, autonomous decision-making, and interacting with humans in an intuitive way. We have seen remarkable achievements from modern Machine Learning systems. However, these achievements come with a very significant energy cost. While the human brain works happily on 20 Watts of power, modern Machine Learning systems can consume tens of kilowatts and only perform a fraction of the tasks our brains are capable of. This is largely because Machine Learning algorithms use conventional digital computer architectures to simulate the analogue behaviour of biological systems.
In order to properly exploit the potential of machine learning to improve our lives we need a step change in computer architectures to radically reduce their power consumption for these new tasks.