Recent AI advancements may come at a huge cost to the environment, according to a new report by the Stanford Institute for Human-Centered Artificial Intelligence.
The report revealed that ChatGPT consumed the most energy of all AI models it reviewed, recording the highest carbon emissions and associated cost of training.
Models assessed included Gopher from DeepMind, Meta’s OPT, OpenAI’s GPT-3, and BigScience initiative’s BLOOM. Of all four, GPT-3 training released 502 metric tons of carbon, 20.4x higher than Bloom’s carbon release and 1.4x that of Gopher.
Besides the carbon emissions, GPT-3 consumed the most energy for its training, with 1,287 MWh.
In order to determine these metrics, researchers considered factors such as the energy efficiency of data centers used and the parameters on which AI models were trained. Apart from Gopher, all models in the study were trained on approximately 175 billion parameters.
Despite such results, the Stanford report noted that existing data is insufficient to say that further AI developments will definitively harm the environment. Indeed, it suggested that AI could even help optimize energy consumption.
AI can help tackle environmental challenges
Several organizations have researched how AI could play a role in addressing environmental challenges. The UN Environmental Program published a post on this very subject, noting that AI could make accessing and interpreting climate data easier, thereby aiding any climate efforts.
According to the coordinator of the UN Environment Program’s (UNEP’s) Digital Transformation sub-program David Jensen, AI’s contribution to tackling environmental issues “can be on a large scale – such as satellite monitoring of global emissions, or a more granular scale – such as a smart house automatically turning off lights or heat after a certain time.”
The European Commission also published a report as far back as 2019, acknowledging the various ways AI could help fight climate change. But it also added that there are environmental impacts.
Should we be concerned?
The focus on the environmental impact of AI is unsurprising given the attention paid to the sector of late. Many have criticized the pace of development, with at least one group calling for a pause on further advancement. Critics will likely point to the environmental costs as yet more reason for a moratorium.
Chat GPT came so fast we forgot to run the societal playbook on it, one step of which is criticizing its carbon footprint. I marvel at our collective ability to demonize blockchains on this and then stay mum on AI. So I asked Chat GPT about its impact. This is what I learned:
— professorstam.eth (@ProfessorStam) April 2, 2023
However, AI is not the first innovative technology criticized for its associated environmental costs. Cryptocurrencies, especially Bitcoin, attracted a fair share of criticism over their energy consumption.
Just as these early criticisms of crypto and Web3 led to more energy efficiency in the industry, many expect this to happen for the AI space too. Such efforts have already been christened “green AI” and “sustainable AI.” One study by PricewaterhouseCoopers suggests AI has the potential to deliver a global reduction in emissions of up to 4% by 2030 compared to business as usual.