Researchers at the University of Geneva have proposed a new approach to running artificial intelligence algorithms using heat instead of electricity. This could significantly reduce the energy consumption of AI technology, which currently consumes vast amounts of electricity.
The Problem with Current AI Technology
Modern AI technology relies on neural networks consisting of billions of interconnected artificial neurons. These networks imitate the function of the human brain. However, unlike the brain, AI technology consumes a large amount of energy.
Using Heat to Mimic Neural Connections
The researchers developed a mathematical model that physically mimics neural connections using qubits and heat. Qubits are quantum bits that can be manipulated to change their quantum states and energies. By turning up the temperature on certain thermal reservoirs, heat flows through the device and changes the states of the qubits, mimicking the flow of electricity in conventional computers.
Building a Heat-Based Perceptron
The team discovered that this type of computer works similarly to a machine learning algorithm called a perceptron. A perceptron is the simplest form of a neural network that can classify objects into specific classes. By building a perceptron using thermal flows, the researchers are taking a conceptually interesting and unusual approach.
Advantages of Heat-Based Computing
According to Marcus Huber at the Austrian Academy of Sciences, building a computer where heat is part of the computation process, rather than a nuisance, could lead to more energy-efficient machines. Heat and entropy are inherent in any computer operation, and leveraging these properties could improve energy efficiency.
Challenges and Future Applications
While the researchers’ conceptual framework could be tested in small-scale laboratory experiments, the challenge lies in adapting it for mass production. If heat-based perceptrons can be manufactured using existing methods, such as chip technology, they could have applications in generative AI and finance, including tasks like derivative pricing.
Source: Science Spotlight
