AI’s impact on the environment has given rise to a debate over if computing-intensive applications and the chips powering them require regulation. This is as per the view of experts’ oration during the conference – “Advancing Technology for a Sustainable Planet” whose host was the Stanford Institute for Human-Centered Artificial Intelligence as well as the Stanford Woods Institute for the Environment.
Advanced technologies such as artificial intelligence and cloud computing utilize energy that results in the emission of carbon while simultaneously also assisting organizations in meeting sustainability goals. This means that organizations must maintain a balance between quick adoption and scaling of growing technologies and understanding the impact of those technologies on the organization’s overall environment.
The environmental impact of technology like artificial intelligence is scale-dependent, according to Peter Henderson, a Ph.D. student in computer science at Stanford University, during a conference panel discussion.
AI algorithms are frequently enhanced by organizations to address concerns regarding energy consumption and carbon emissions before they use a machine learning model.
We want to ensure that we don’t scale to the level where we are harming the environment when the objective of a lot of machine learning work is AI for social good, where we want to produce more sustainable things, optimize batteries, and energy grids, he explained. However, if all of that optimization results in more negative impact than positive impact, it’s not worth the effort.
Henderson claims that apart from the steps taken by the organization to optimize such technologies, the government should also put in its efforts to provide regulations for AI uses. Though few efforts for regulating AI are ongoing in the European Union, that is not sufficient for completely addressing the environmental impacts of the technology.
Addressing the environmental impact of AI
According to Henderson, the AI regulations of the EU prioritize consumer protection over environmental protection. He went on to say that GPUs, which power AI and machine learning models, contribute significantly to AI’s environmental impact. To address environmental concerns, he said, the regulation would have to comprise chips and other technologies underlying AI use.
Henderson further stated that, in California, according to a recent regulation, few GPUs were prohibited from being sold forthwith due to their lack of efficiency, which could be a step forward in terms of driving innovation and requiring more efficient chipsets.
Though there are incentives for organizations to use more efficient chips, such as lower costs, Henderson says it is still an area that needs to be considered from a regulatory standpoint.
He also stated that since there is no sufficient data on the impact created by technologies like AI, cloud computing, and bitcoin on the environment, creating an effective regulation is still a challenge. Hence the first step is to ensure that there is sufficient reporting, and data, so that efficient regulatory and policy decisions can be taken, according to Henderson.
Salesforce executive discusses sustainability
It is not a simple task to measure the total carbon emissions of an organization, since much of the reporting is an estimate instead of actual measurements, according to Kathy Baxter – principal architect of Salesforce’s ethical AI practice, who was also an orator at the conference.
There is no way to really realize if we’re truly getting better and being able to see what’s correlation, causation, and being able to classify, she explained. It is impossible to control the emissions if their source is unknown.
More data, according to Baxter, is required to help further sustainability efforts.
It is impossible for any single organization or government to solve this problem, she added. We need to merge our data and collaborate to figure out the solutions.
