The Trouble with Brain Metaphors in Science

The human brain has been likened to many things over the course of history: a hydraulic pump, a telegraph network, a telephone switchboard, and now, a computer. Each metaphor has been a reflection of its time's technological zeitgeist, promising clarity but often leading us astray. The YouTube channel Machine Learning Street Talk dives into this metaphorical quagmire, asking: what if our current understanding of the brain as a computer is just another chapter in a long history of simplifications that miss the mark?

Professor Karl Friston's free energy principle, which attempts to reduce all behavior to a single equation, is one such simplification. Friston himself likens it to the 'spherical cow' joke in physics—a model so simplified it becomes almost meaningless. Yet, as Friston admits, “The free energy principle is not meant to be complicated or difficult to understand. It’s almost logically simple.” This raises a crucial question: has Friston uncovered a universal truth about the mind, or is he merely offering another oversimplification?

Professor Mazviita Chirimuuta, in her influential book The Brain Abstracted, wrestles with the consequences of these simplifications. She argues that while simplifications are necessary—after all, our brains are limited in their capacity to hold complex systems—they often lead us to mistake our models for reality itself. Chirimuuta's work suggests that the metaphor of the brain as a computer may obscure more than it illuminates.

Francois Chollet's 'kaleidoscope hypothesis' posits that beneath the universe's apparent chaos lie simple, repeating patterns. It's a beautifully optimistic view, reminiscent of Platonic ideals, suggesting that reality is a neatly structured code. Yet, as Chirimuuta might argue, embracing such ideas risks falling into the trap of believing our metaphors a bit too much.

Then there's Joscha Bach's provocative claim that software is not just metaphorically but literally spirit. He argues that the same algorithm produces the same effects, regardless of the substrate it runs on. This idea, reminiscent of 17th-century Cartesian dualism, posits software as a causal mechanism independent of its physical form. But as Chirimuuta would likely ask, isn't this 'sameness' imposed by us rather than existing in nature? It's the distinction between seeing patterns as truths versus useful fictions.

The allure of metaphors is their simplicity, but as history has shown, simplicity often comes at the cost of accuracy. We've seen it before: Descartes likened nerves to tubes filled with fluids, later scientists compared them to telegraphs, and now we see them as computer circuits. Each metaphor was enlightening in its time but ultimately limited in scope.

Nobel laureate John Jumper offers a pragmatic view: "AI can predict and control, but understanding requires a human in the loop." This sentiment underscores the limitations of our models. When we lean on abstractions like the mind-as-computer, we risk losing the depth of understanding that comes from acknowledging complexity. The challenge lies in balancing the utility of simplifications with the richness of reality they often obscure.

So where does that leave us? Perhaps, like the spherical cow, our metaphors are useful starting points but poor endpoints. As we continue to develop AI and explore the mysteries of consciousness, we might do well to keep one eye on the history of our metaphors, recognizing them for what they are: tools, not truths.

By Mike Sullivan