The creation of AI-designed chips raises concerns about security, repair, and technological sustainability

A research project developed an artificial intelligence system capable of creating a series of chips that are not comprehensible to human beings—in other words, no scientist currently can fully understand how they function.

A striking discovery, considering that humanity has reached a point where virtually everything around us depends on chips. These electronic components, which power everything from our phones to cars and the systems that allow us to track animals, are the backbone of modern technology.

How Artificial Intelligence Created Incomprehensible Chips:
Chip development has always been a long and complex process. Engineers, with years of experience and extensive training, have been responsible for designing the circuits that power the devices we use constantly. To do so, they follow a set of established rules based on decades of research.

In this context, artificial intelligence has begun to play an increasingly important role. Through convolutional neural networks (CNNs), researchers are achieving chip designs far more complex than humans could conceive in a reasonable amount of time.

Kaushik Sengupta, an electrical engineer at Princeton University and the project’s leader, has been one of the main drivers of this research. In his studies, he uses AI to design more efficient wireless chips— a key area for the future of global connectivity.

The most surprising aspect of this research is that the chips designed by AI cannot be fully understood by humans. According to Sengupta, current engineers could not—nor likely will in the future—deeply understand how each of these chips works, which raises a series of questions about their repair, modification, or even safe usage.

In his own words, AI-designed chips could be “disposable” if we fail to understand how to repair or improve them.

The process of designing a chip with artificial intelligence deviates from traditional methods. Instead of starting from an already established blueprint, AI uses what is known as “bottom-up design” or “inverse design.”

This approach starts with the desired outcomes and works backward to create the components that will make up the final hardware. Unlike humans, who must follow a logical and structured pattern to design components, AI algorithms do not require such linearity. This allows the algorithms to find nonlinear, even unexpected solutions.

Human designers have always had to work within a set of constraints, such as pre-existing templates that define how chips should be. However, AI is capable of going beyond these restrictions, suggesting new configurations and paradigms that might have once been considered unthinkable.

The speed at which these algorithms can generate new designs is impressive. According to the researchers, what might take a human years to accomplish, an AI algorithm can suggest in minutes.

Sengupta and his team have explained that traditional chips are the result of meticulous interaction between components that are assembled piece by piece to ensure signals flow as intended. However, with AI, the resulting configurations are far more complex.

AI-Generated Chips:
“Classic designs carefully assemble these circuits and electromagnetic elements, piece by piece, so that the signal flows the way we want it to flow within the chip. By changing those structures, we incorporate new properties. Before, we had a finite way of doing it, but now the options are much greater,” stated Sengupta.

The Risks of This AI Creation:
This breakthrough could have both positive and negative implications. On the one hand, AI’s ability to create more complex and efficient designs could drive the development of far more advanced technologies. The chips AI designs can, for example, be faster, smaller, and capable of performing complex tasks beyond those of their predecessors.

If human engineers cannot understand how to repair or modify a chip, dependence on AI could lead to a situation where devices simply have to be replaced in the event of failure, rather than repaired. This could make these chips more “disposable” and less durable, which, in addition to creating an economic issue, could also contribute to the growth of electronic waste—one of the major environmental challenges of the future.

The CEO of NVIDIA, one of the leading companies in the development of advanced chips, has already warned at CES in Las Vegas that its AI chips are advancing faster than Moore’s Law (which states that the number of transistors on a chip roughly doubles every two years) could predict. This suggests that we are entering an era of change so rapid it may become difficult for humans to keep up.