Four reasons to be optimistic about AI’s energy usage
1 min read
Summary
AI development has accelerated rapidly in recent years, but the process is energy-intensive and hugely damaging to the environment, with companies focusing on making bigger, better models rather than improving their energy efficiency.
Ali Farhadi, CEO of the Allen Institute for AI, said ‘dollars are being invested, GPUs are being burned, water is being evaporated – it’s just absolutely the wrong direction’.
However, there are innovations that could improve the efficiency of AI models, including more efficient software, computer chips and data centres.
For example, training AI models on more curated data sets, rather than larger sets, can be quicker and therefore reduce energy usage, while moving computing out of data centres and on to individual devices could reduce energy demand.
There are also more efficient computer chips under development and new cooling systems for data centres that could reduce energy use.
Ultimately, the best way to cut costs in AI is to reduce the energy bill, which should incentivise companies to increase efficiency.