AI is here to stay, and ChatGPT is a clear example. Whether you need help with a university assignment, writing a letter to Olentzero, or something else entirely, it’s got you covered. Washing your car? Not yet, but give it time! AI, particularly generative models, can handle almost anything. What began as fairly simplistic models have evolved into systems capable of solving a multitude of problems. By leveraging increased computational resources and extensive, costly training, these models improve daily. But how much of this is truly positive? Do we really need to book a stadium like San Mamés to celebrate a birthday?

In AI, progress has been swift. Companies like Google, Microsoft, and OpenAI compete for dominance by developing increasingly complex models and using more powerful computers to accelerate tasks. However, this progress comes with a high energy cost, consuming growing amounts of resources like water and electricity and raising concerns about sustainability. Faced with this clear limitation, a key question emerges: instead of seeking more resources, what if we explored new ways to tackle these problems?

THE CROSSROADS OF AI: PROGRESS AND HIDDEN COSTS

Given the current challenges of AI, such as the complexity of models and their energy demands, it may be time to look to other fields for inspiration. Think of it like preparing dinner at a txoko: you may be an expert at making cod, but listening to others might teach you a new trick for improving your marmitako. In condensed matter physics, for example, scientists have long grappled with similar issues when attempting to simplify quantum systems.

To understand this better, imagine trying to simulate how particles interact in a quantum world. Everything seems fine until you add another particle, and suddenly, the required calculations double. It’s like organizing a party: every additional guest not only increases the fun but also the chaos and the number of pizzas you need to order. In the quantum world, this happens because particles are interconnected in incredibly intricate ways (thanks to phenomena like superposition and entanglement), making the system unmanageable faster than your post-vacation to-do list.

However, the scientific community discovered a clever trick. This runaway need for resources mostly occurs when particles are extremely entangled, making the system too complex to handle classically. But if the entanglement is more "local", where each particle only interacts closely with its neighbors, special mathematical tools can simplify the system without overloading your computer. It’s like the earlier party, but instead of everyone shouting, they play a quantum version of “telephone.” Best of all, many natural systems work this way, making this approach incredibly useful.

TENSOR NETWORKS: QUANTUM INSPIRATION TO SIMPLIFY COMPLEXITY

These tools are called tensor networks. Essentially, they’re a smart shortcut for breaking down complex quantum systems into something more manageable. For decades, tensor networks have been vital in advancing fields like condensed matter physics and quantum chemistry.

But what does one thing have to do with the other? Much more than you might think. Tensor network techniques can decompose quantum systems by capturing how one particle is entangled with the rest of the system. This concept has quickly found parallels in AI models, which also involve correlations between different nodes.

As a result, tensor networks can simplify highly complex models by breaking them into manageable parts or combining simple models to represent intricate systems. A useful analogy is DNA analysis: while a person may have thousands of unique characteristics (such as hair color, eye color, or height) their DNA condenses this information into a compact sequence. This compact representation can even enable precise reconstructions, as seen with Dolly the sheep. Similarly, tensor networks unravel complicated structures, transforming them into more comprehensible and efficient representations.

A SUSTAINABLE FUTURE: INNOVATION AND EFFICIENCY IN AI DEVELOPMENT

These quantum-inspired techniques, originally created to understand complex quantum systems, are now emerging as an exciting alternative for designing AI models. They function much like lateral thinking: instead of following the conventional path, they find creative, unconventional solutions that might otherwise go unnoticed.

In a world where AI development consumes ever-increasing resources, these strategies offer a more efficient and sustainable option. It’s like viewing a problem from a completely different angle, uncovering smart shortcuts that make the process faster and less costly.

With the planet’s resources limited, creating more efficient and sustainable models isn’t just an option, it’s a responsibility. Quantum-inspired techniques like tensor networks represent an alternative for making AI more powerful while respecting the environment. This becomes especially relevant as the energy demands of large language models (LLMs) rise, prompting some tech companies to consider buying and reopening power plants or even building new ones to meet the energy needs of training these models. By adopting innovative approaches that optimize resource use, we can solve complex problems while ensuring that technological progress aligns with preserving the planet for future generations.