Our minds aren’t equipped to handle AI

Posts from this topic will be added to your daily email digest and your homepage feed.

Our minds aren’t equipped to handle AI

AI is junk food for the mind: easy, tempting and ultimately very bad for you.

Posts from this author will be added to your daily email digest and your homepage feed.

Image: Pedro Nekoi for The Verge

Norbert Wiener, godfather of cybernetics, once said, “The thought of every age is reflected in its technique.” For the past century, our thought has been reflected in our computers, including by those in the AI industry. Google’s Demis Hassabis calls the brain “a biological approximation to a Turing machine.” Elon Musk puts it more bluntly, declaring that “people should just think of the brain as a biological computer.” (Musk’s brain often worries me.)

But humans are more complex than a straightforward comparison to computers gives us credit for — and far more than most of the AI industry seems to appreciate. And as their products push ever deeper into culture, this helps explain why they’re making such a mess. AI, it turns out, is something like a cognitive version of a hot dog: It appeals to us in the moment, but undermines the health and sustainability of cognitive systems we’ve evolved over several millennia.

To understand the problem, we’ll need a little neuroscience and the entirety of human evolutionary history.

The idea that brains are computers traces back to Alan Turing. It posits thought as a three-step process, functionally an algorithm: Our minds take in information from the world (input), manipulate it in some fashion (computation), and then enact behavior (output). Perception leads to cognition, which in turn leads to action. In slightly more elaborate terms, it looks something like this chart:

In many ways, this model has been productive. Imagining our brains as computers, technologists have pushed actual computing from simple adding machines to artificial neural networks and generative AI, seeking ever more detailed mirrors of our own minds.

But this mirror’s image, while not exactly wrong, is warped and limited. John von Neumann, a seminal figure in developing computers and computer science, doubted that the computational model could possibly capture the “exceptional complexity of the human nervous system.” That is, our nervous systems evolved to help us navigate the large ecological system that we call “the world.” And we act upon the world to exercise control over it (as best we can).

The computational approach looks at the end product of our minds and tries to “reverse engineer” how they function. But instead of working backward, we might instead examine the long arc of evolutionary history to build our model from the ground up. This is the approach favored by Paul Cisek, a neuroscientist at the University of Montreal, who’s patiently developed a biological model of brain and nervous system development spanning millions of years.

Cisek contends that rather than information processors, our brains are better understood as feedback-control systems. Our bodies don’t just receive input, they take action to adjust what that input is, contingent on what options are available. As Cisek himself is quick to note, this is not a new idea — writing at the turn of the 20th century, philosopher John Dewey described the mind as a circuit, “more truly termed organic than reflex, because the motor response determines the stimulus, just as truly as sensory stimulus determines movement.”

What exactly is the difference between the two approaches?

Here’s a classic example from baseball: catching a fly ball in the outfield. According to the computational model, solving this problem must involve some complicated and subconscious mental calculus wherein the outfielder estimates the ball’s velocity, calculates the effect of gravity, and undertakes untold other mathematical procedures to “compute” where the ball will go.

Is language the same as intelligence? The AI industry desperately needs it to be

In contrast, the feedback-control approach suggests a simple heuristic — essentially, “keep the ball in the same position within your visual field, and then move to maintain that situation.” We take action to adjust the stimulus we receive.

This is not only more true to the experiences of anyone who’s played center field, it also avoids separating out the mental process from physical movement, and avoids invoking the use of complex calculations that the computational model requires. Humans are more dynamic than that. .

This approach also neatly maps to the biological architecture of brains as they’ve evolved over time. From ancient fish to amphibians to mammals to primates and eventually modern humans, what we observe is that new behaviors emerge in response to new environmental possibilities. When dinosaurs died off, for example, this meant some nocturnal creatures could move about the world in the daytime with less risk of being eaten, giving rise to a variety of new capacities. The history of our nervous system, Cisek observes, is one of “continuous extension of control further and further into the world.” (His forthcoming book will explore all this in greater detail, and yes, I’m hoping this essay puts subtle pressure on him to finish it.)

This leads to an alternative and very different diagram than the one above. We can both lay out a map of behaviors and abilities as they emerged over time and we can overlay the specific physical components of the brain to these capabilities, like so:

Instead of moving from left to right as in the computational model, this model should be seen as unfolding from top to bottom over evolutionary time. For example, a long time ago, as vertebrate animals developed more mobility, it became useful to have specialized systems for exploration — by using landmarks, say, or navigating at night. This led to the development of what we now call the hippocampus. Importantly, this also led our ancestors to remember important moments of scurrying from one spot to another, leading to the development of “episodic memory” of past experiences.

We can’t do anything remotely like this with the computational model of the mind; it simply does not sync up to observable neuroscientific structures. What’s more, it obscures that so much of what brains are doing involves controlling living organisms’ interactions within their environments. As such, Cisek suggests we need to dramatically shift the paradigm we’re using to understand the relationship between our brains and behavior, moving away from algorithmic input-outputs and toward more dynamic feedback systems.

Thus far, our story of human development has largely centered on feedback from the physical world. But one of humanity’s most important “feedback loops” arises from our profoundly social dispositions. The computational model, it turns out, doesn’t account for this well either — and neither does the AI industry. We’ve spent thousands of years building institutions and norms for learning from and communicating with each other, and over the course of less than a decade, Big Tech companies have systematically worked to dismantle them.

How? Again, we’ll need some historical context.

At some point around many hundreds of thousands of years ago, our distant ancestors did something incredible: They learned to imitate each other. Mimicking gestures and body movements allowed us to pass along successful practices — like chipping away at a stone tool — and coordinate more complex activities through shared ritualistic behaviors, which in turn shaped our cognition. It was the dawn of human culture.

We’ve spent thousands of years building institutions and norms for learning from and communicating with each other, and Big Tech companies are systematically working to dismantle them

Humans soon progressed to imitating sounds and, in turn, to oral language. As we’ve covered previously, language is not the same as thought, but it enables us to communicate our thou