The Future Of AI Models And Their Environmental Impact

The Future Of AI Models And Their Environmental Impact

Recent studies have revealed the link between artificial intelligence (AI), machine learning (ML) and rising carbon emissions, sparking concerns about the computational resources and energy needed to drive AI and ML models. One report estimates that training certain AI models entails emissions of approximately 626,000 pounds of carbon dioxide – nearly five times the lifetime emissions of an average US car, including manufacturing, according to the Massachusetts Institute of Technology (MIT). This is forcing developers to rethink how they use AI.

AI and ML models are deployed across various digital solutions to optimize business performance. They are also embedded in devices from smartphones to vehicles to virtual assistant technologies. Now, researchers and advocates are asking why energy efficiency is not at the forefront of this resource-consuming technology, especially considering growing net zero priorities and targets.

Multinational tech firm Google has been directly impacted by this issue after an initial study, co-authored by a Google AI ethicist, highlighted if AI language models may be too big. The paper examined whether tech firms are actively working towards mitigating potential risks – particularly the increasing environmental and financial cost on marginalized communities who are least likely to benefit from the progress achieved by language models.

Since then, Google has commissioned a new paper in collaboration with the University of California to debunk the findings and to illustrate their commitment to energy efficiency. The paper claims that Google engineers work to improve existing models, thus reducing emissions. The research references Google’s Evolved Transformer model, which uses 1.6 times fewer floating-point operations per second (FLOPS) and requires less training time. This and other examples are meant to signal Google’s commitment to carbon efficiency.

In light of the growing emissions footprint of AI, engineers and developers should continue to work on energy efficiency, with initiatives such as running algorithms on hardware powered by green electricity, recycling waste heat, and managing computing resources in an energy-efficient way. Another example of energy efficiency in AI includes recycling models across different applications.

Lastly, AI developers should report on their energy usage and carbon dioxide emissions to show their commitment to net zero. They should also consider carbon neutralization methods and technologies.