Cleaning the AI(r)

AI is not that silver bullet that will help us achieve all our targets. What it can do is help us reduce greenhouse gas emissions in various ways. But is AI entirely green and clean?

The Paris Accord had 190+ countries as signatories and a stated goal – by the middle of this century, the average global temperature should not be allowed to rise more than 1.5 degrees C from pre-industrial levels. If it touches 2, the climatic damage caused will be irreversible. Millions will die or fall sick due to displacement, wildfires will burn down habitation, icebergs will melt, and the sea level can increase by as much as 3 inches. Without going into the gory details, I think we all understand and agree that we must do something big, really big to avert a catastrophe.

At the outset, it must be clarified that AI is not that silver bullet that will help us achieve all our targets. What it can do is help us reduce greenhouse gas emissions in various ways. By the way, AI is not all green, it is red too and we will see shortly how AI also adds to the carbon footprints. But let that be for later.

With data, we can plug leakages in power generation and distribution. Capgemini says that the improvement through load-balancing and effective grid management in real-time can be 15% in the next 3 to 5 years. At present, data & AI are not being used to the extent possible and energy companies can attack this angle to minimize wastage and leakage. It can start with large industrial outlets and subsequently be adopted by households too.

January last year, Larry Fink, the Chairman & CEO of Blackrock, the world’s largest asset management company, reached out to the companies in his portfolio, emphasising the need for full disclosure on the ESG front. But the problem is that these initiatives are costly and not so easily measured. That is why, leaders often end up paying lip service and not doing something concrete. AI is the proverbial game-changer, here. It can provide deep insights by monitoring emission levels from multiple sources and every part of the value chain including downstream suppliers & SCM providers. The next step is of course a predictive approach that helps to cut down emissions. If done right, a BCG analysis says that value-added through corporate sustainability can be in the range of 1 – 3 trillion dollars in the next ten years. The challenge is when do we start? If in year 1, it is revealed that the company is doing badly on CO2 emissions, would it be able to make a full disclosure and risk its share prices from tanking? And, in a year when companies are limping back to pre-Covid growth but the threat of the second wave also looming large?

Let us look at the math.

Global GHG emissions currently are about 53 gigatons of CO2 (Carbon Disclosure Project). To meet the Paris Accord, we must reduce it by 50%, and AI can help us by about 5 – 10% or 5 gigatons.   

Red AI is a tricky one. From a cited research paper these words stare back menacingly: “AI seems destined to play a dual role. On the one hand, it can help reduce the effects of the climate crisis, such as in smart grid design, developing low-emission infrastructure, and modelling climate change predictions. On the other hand, AI is itself a significant emitter of carbon. This message reached the attention of a general audience in the latter half of 2019 when researchers at the University of Massachusetts Amherst analysed various natural language processing (NLP) training models available online to estimate the energy cost in kilowatts required to train them.”

The study showed that training a single massive-sized language model releases 300,000 lakh kilos of CO2. For a layperson to understand, that is equivalent to 125 round-trip flights between NY & Beijing. But this is not the only problem. Is Big Tech consciously avoiding working with companies who are not committed to the cause of saving the planet? We do not know yet, and to what extent. And, of course, every training model is not as big as to release vast amounts of CO2. Training servers consume the bulk of the energy and the alternative is to shift large-scale computing to locations that are reliant on renewable energy. A group has developed an emissions calculator which gives a fairly accurate indication of the level of emissions while considering various aspects such as the neural network architecture, location of the training server, the hardware used, and the energy source.    

The WEF study has gone in-depth on how to save the planet. Clean Air is one of the parameters and it has been broken down into 5 parts: clean power, smart transport systems, sustainable production & consumption, sustainable land-use; and smart cities.

Saving the planet will require a concerted effort. Piece-meal approaches will not cut it. 


Leave a Reply

Your email address will not be published.

AI ML Machine Advocacy Council (c) 2022