Artificial intelligence and its environmental impact

370 Views

Artificial intelligence is often considered a concrete solution to accompany the climate transition. However, it is paradoxically doomed to aggravate it – in particular because of its excessive consumption of energy necessary for the storage and processing of data in terms of algorithm training. The pollution caused by AI is sometimes forgotten, because it is ‘invisible’ – but its impact on the environment is very real.

The carbon footprint of digital technology represents nearly 4% of global greenhouse gas emissions, and the amount of energy and resources necessary for its operation today raises major concerns This includes systems such as ChatGPT, Dalle-E and Diffusion accounting for a large proportion of this. Here, Arthur Llau, Head of AI at Flowlity, an innovative AI-based supply chain planning and optimisation solution, discusses how greener artificial intelligence can be achieved, whilst also continuing to satisfy its growing demand.

Concrete solutions do exist – it is crucial to put them into practice

How to do as well, and even better, with less is the question that artificial intelligence must now address – with all the seriousness and investment that requires. Steps have already been taken in the search for an AI that is more efficient in terms of data, energy and resources, without compromising its performance or its scalability. And the solutions are found both in artificial intelligence itself, but also in all the external structures that bring it to life.

First, it is necessary to better understand the AI ​​in order to adjust the resources necessary for its proper functioning, and thus reduce its energy consumption as much as possible. It also requires better targeting of the data, which it is possible to isolate and activate the right mechanisms for fewer iterations.This also requires a change in research, to focus more on how to do well for a lower cost as opposed to continuing to go for more and more in terms of performance.

In addition, data is a central issue, due to its size and storage methods. Today an incalculable amount of data is stored and is increasing every day. It is estimated that the volume of data stored in the world will increase from 33 Zo in 2018 to 175 Zo in 2025. The challenge is therefore to better analyse them to sort them, and ultimately only store clean data that is reliable, usable and useful.

Raising awareness – educating and constraining to make things happen

If the first solutions are already in place, the effort is far from sufficient. While the ‘end of abundance’ has pushed many sectors to act and transform, tech does not seem to have decided to make it a priority. Far from it, it is almost a non-event. It is true that engineers are not really encouraged to direct research and practice towards greener notions. It is also true that it is complex to perceive and quantify the impact of an algorithm, in particular due to its immateriality. Finally, it is true that nothing obliges a tech player to measure its carbon footprint and try to reduce it.

However, it is up to the players in the sector to take responsibility, to act and to participate in the collective effort. First, by taking a step back: asking yourself the right questions, reframing your needs, recalculating the relationship between the real interest of performance and the cost of sometimes useless models.

Then by educating all of its stakeholders: guiding AI users towards more moderate and better organised use, encouraging research laboratories to move forward on these issues, developing new learning and processes for engineers. A greener AI will also depend on the ability of its players to make compromises in order to better redefine the place of the cursor between technological prowess and reductions in energy and environmental costs. Changing the approach to research will also be crucial for this – placing focus on how to do well for a lower cost.

Artificial intelligence has interfered in all sectors and has become a real development issue, leading to gigantic progress in a few years. However, these developments have required the mobilisation of a considerable number of resources, which need to be reduced in a context where they are increasingly precious. After years of racing for performance, it is urgent to change the paradigm and embark on a race for intelligence. The road to a frugal artificial intelligence will be long – but it has to start somewhere.

For more information about Flowlity’s AI-based stock optimisation and forecasting solution, visit: https://flowlity.com