Engineers, how are you training your AI ?

Deep Learning has a huge carbon footprint. A research paper concretizing this dirty secret dropped from Strubell, et. al, at University of Massachusetts Amherst. A summary :

Reading this sucked the life out of me. Made me feel sick. Threw me into an existential crisis. “Should I quit and do regenerative farming in India for a career ?” The first words I wrote when I started this blog :

1.Make a righteous living.

‘I can only be productive if I consume at the expense of the planet’. ‘I am a coal lobbyist and will continue to push for the burning of the fossil fuels’. ‘I am a pimp and it’s my job to exploit the sex workers’. ‘I benefit from fast fashion and will feature it on my blog’.

I don’t want to be in these situations.

I use AI to help make green tech more efficient, for a living. I choose this over other career paths that are normal for folks in my field : make gadgets cooler, help sell ads, improve usability of certain existing features, help individuals with visual impairment navigate, teaching, …. I chose this for it’s contribution to protecting the environment. This one paper shattered my bubble. But atleast now, I know. A few steps I am taking right out of the bag to reduce the harm.

  1. My team spent the last two years creating an IoT framework for our power systems. They chose AWS back in the day for being the most mature tech out there. Now that we know about the carbon footprint, I have decided to move the development of my module to Google Cloud. Urge your cloud service providers to go renewable. Demand it, if you are in a position of power.
  2. “Markets are usually a good way to organize economic activity, but in some cases, they are prone to malfunctions, which are called market failures. By far the largest example of such a failure is that market prices fail to take environmental costs into account. That means every single economic transaction in the world is based on incorrect information: the environmental cost component is missing from the price. This kind of market failure is called a negative externality.” We the engineers have to start reporting Green House Gasses in our work. We have to put out reports of the impact of mined metals in our products. We have to put out numbers on the amount of recycled content used. We have to design products for 100% recyclability.
  3. Start having conversations about climate crisis and our emissions from every aspect of our lives. The environmental racism should not go unchecked. I especially see it in folks who work in tech. I hear “we will be fine. Tech will come solve all of our problems”. By “we”, they mean themselves. It does not matter that millions of beings are suffering right now because the “We” clan’s privilege distances them from the harm. That’s blatant environmental racism, out in the open ! #climateAparthied
  4. Teachers and leaders, educate your fellow AI practitioners on the resources you consume. For instance, look at this lecture by Kilian Weinberger for Cornell CS4780. The student essentially suggests we consume excess to arrive at a possibly better solution. The response of this professor is worth noting. When we talk about cost, we only seem to talk about time, effort, money, inconvenience, unfulfilled personal whims, …. Start talking about the harm caused by your GHG’s as a cost.
  5. The answer to everything isn’t 42. High tech solutions are not always necessary for every problem. If it can be solved without it, please do so.
  6. Green AI.