Continuing our discussion from Part 1, generative A.I. can be a remarkably helpful yet controversial tool. Aside from the potential negative cognitive impacts of generative A.I., there are also significant environmental impacts caused by the massive data processing centers required to operate generative A.I. systems such as ChatGPT. MIT reports that training a generative AI model may require up to 7 to 8 times as much power as training typical computing systems. Additionally it reports estimates that every ChatGPT query consumes about 5 times as much electricity as a regular web search. Many users are unaware that these generative A.I. data processing centers also consume an enormous amount of water for cooling purposes. This environmental impact is significant and devastating to a world that is already choking from the effects of climate change which include severe weather such as wildfires and droughts.
Image of wildfires below a hillside and smoke filled sky
Photo by Matt Palmer on Unsplash
This all sounds very doom and gloom, but what can we do about it? First, we can be mindful about our usage of generative A.I. Do you really need ChatGPT to answer this question, or is it something a simple Google search can accomplish? Check here for information on how to turn off automatic Google A.I. overviews. Do you need an A.I. image generator to create a copyright-friendly image for your post, or can you use a website such as Unsplash which is committed to only accepting A.I.-free photos? Or better yet, could you design an original image using your creativity? Next, we can get the word out about these environmental impacts and make sure that journalists and media companies are reporting on it. According to tech consulting firm Capgemini, only 12% of A.I. companies report tracking the environmental impacts of their products. But if humans can switch to electric cars and solar panels, we can certainly come up with a way to harness generative A.I. in a more sustainable, environmentally-friendly way. Generative A.I. companies just need the right incentive to do so, such as public awareness and media pressure.
Ending on some good news–there are already researchers working on sustainable A.I. Scientists at UCLA have recently developed generative A.I. models that utilize photonics to generate images rather than digital computation. This essentially means that instead of the 1000 steps required for traditional A.I. image generation, this new method only uses 1 step. So while this new method still uses electricity, it uses much much less.
Generative A.I. has massive environmental impacts, and those impacts are only projected to grow over the next few years. Therefore we should make sure we are talking about these issues and pressuring tech companies to invest in environmental preservation and sustainable A.I. development. And of course, we should all be mindful of our personal A.I. usage, in the same way that we may take shorter showers or turn off all the lights when we leave the house. The steps you take don’t have to be huge, as long as they’re taking you in the right direction.