Artificial Intelligence (AI) suffers from two massive blocks: colossal, costly energy use and a transparency deficit. The technology can be technically feasible but is still too expensive for most organisations to consider it. That was the conclusion of a recent paper from MIT Future Tech, which looked at computer vision tasks as an example of an AI-enabled technology. But the MIT group isn’t alone in highlighting the problem.

The computing cost of Deep Learning is exploding. Sam Altman, CEO of OpenAI, made it clear last year that training ever larger Language Learning Models (LLM) is not the way to advance AI, not least because teaching GPT-4, its latest product, cost over $100m, while by 2027 it is estimated that the AI industry could consume as much power as a country the size of the Netherlands.