AI optimizes resources, but also devours them: can the paradox be overcome?

While AI has shown boost precision agriculture, improving medical diagnoses and even – as contradictory as it may seem – optimize the use of water and energy, the paradox remains.

old architecture

One of the main reasons AI consumes so many resources is that it is built on computer systems designed decades ago. The volume of data is such that it is physically impossible for processors to continue managing it efficiently.

In recent years, technology was able to circumvent the von Neumann bottleneck—the incessant data traffic between memory and processor—thanks to Moore’s Law, which doubled the number of transistors on a chip every two years. But that limit has been reached: Stacking more components reduces the surface area available to cool them.

To function optimally, these chips require refrigeration processes which not only demand a large amount of energy, but also enormous volumes of water. An average data center can use about 9 liters of clean water for every kWh consumed.

Added to this is the accumulation of waste. The increase in demand for AI forces the manufacturing of large quantities of hardware that must be constantly renewed, generating increasing volumes of electronic waste.

New paradigm

There is little point in anchoring ourselves in the discussion about whether we should adopt artificial intelligence: its development is inevitable and necessary to improve resource management. Efforts should focus on optimize it.

And although progress is already being made in lighter and less demanding AI models, the true revolution will come not from software optimization, but from hardware.

It is necessary to continue exploring new paradigms computing, such as in-memory computingwhich seeks to solve the von Neumann bottleneck by designing chips where processing and memory are integrated into a single component: ‘memristors’ are an example of this technique.

The revolution photonics is another great bet: it proposes to stop using electrons and replace them with light particles (photons). By not having mass or generating heat through friction, a photonic processor could be thousands of times more efficient. There are still challenges to be resolved to achieve mass adoption.

Finally, we explore the analog computing: Unlike digital chips (which operate with 0 and 1), these systems are inspired by the physics of natural systems to process information more fluidly, similar to our brain.

More sustainable AI is not only a technical issue. Innovation must be accompanied by policies that promote its development. We are facing a technology capable of transforming our world for the better, as long as its progress moves in the right direction.