In an article at Big Think last month, University of Rochester astrophysicist Adam Frank identifies a deep issue underlying the idea that AI will be equivalent to a human being — one that goes beyond technology and all that. First, he poses a question:

Is life itself, sentient or not, something that can be represented in a machine? In other words, is life computable? Whatever life is, can it be reduced to a series of mathematical operations carried out by a computer?

Adam Frank, “Even the smartest AI likely won’t be ‘alive.’ Here’s why,” Big Think, May 30, 2024

Frank thinks not. He points to the work of theoretical biologist Robert Rosen (1934–1998).

Rosen was called “biology’s Newton” in his obituary at the Elsevier journal Computers & Chemistry. The obit begins, “The history of science is replete with stories of unrecognized brilliance. Bob Rosen’s story is among them.” So what did he say?

The strange loop

Rosen’s argument, as summarized by Frank, focuses on the idea that life forms are self-creating and self-sustaining:

To understand what self-creating and self-maintaining mean, think about the membrane of a single-celled animal. By allowing only certain molecules to enter the cell, the membrane is what allows the organism to metabolize and stay alive. But only by staying alive can the cell create the processes that create and maintain the membrane. The cell creates the processes and products (the membrane) that create the processes and products that create … all in what some people call a “strange loop.”

This closure to efficient cause was central to Rosen’s ideas about life. It was, for him, what set life apart from other physical systems. Most of all, it was what set life apart from machines. Machines always have their efficient cause created by someone else (i.e., humans). Even if you came up with a machine that could fix itself, it would be you who came up with that plan not the machine itself.

Frank, “Here’s why”

So the question here is how we get a “strange loop,” a loop that does seem to have an original beginning in something other than life. In other words, the problem is a version of “How do we get something from nothing?”

In his 1991 book, Life Itself (Columbia University Press), Rosen offered a mathematical argument that this quality — closure to efficient cause — was beyond the reach of a computer. A mathematical argument like his is possible because the problem is somewhat like the NP-complete problem in computer science: “any of a class of computational problems for which no efficient solution algorithm has been found” (Britannica). In other words, an NP-complete problem is not computable.

Not disproven

So far, Frank tells us, Rosen’s argument against computers that behave like living beings has not been disproven, though a “small cottage industry” has tried.

A comment from Rosen’s eulogist, Donald C. Mikulecky, is apt here:

There is one controversial idea around which everything Rosen did was centered: he once said that modern molecular biology is populated by people who insist that something be dead before they study it. In that respect, ‘molecular biology’ is an oxymoron. There is no life at the molecular level. The scientific community did itself a disservice when it allowed that term to become one of its banners. It is, at best, molecular biochemistry and biophysics, if one understands that these two fields have no direct link with the study of life.

Mikulecky, Donald C.. “Robert Rosen (1934-1998): A Snapshot of Biology’s Newton.” Computers & chemistry 25 4 (2001): 317-27.

If so, having adopted an approach to life that interprets it in terms of physical and chemical forces — somewhat like an analysis of the workings of a computer — scientists then assumed that a computer can replicate life. They lost sight of the fact that they themselves have chosen an image of life that does not capture one of its essential features.

If even the smartest AI won’t likely be “alive” in any meaningful sense, what will become of Ray Kurzweil’s quest to beat death by getting uploaded into the cloud? But then, he thinks computers will be human-like within the next five years.

Meanwhile, many programmers would settle for finding a solution to model collapse and other distinctly non-biological ills.

Events over the next five years will tell us a lot about whether Rosen is likely right in principle.

You may also wish to read: Is it sinking in? Chatbots will not soon think like humans. Tech writer Gary Marcus: Even futurist Ray Kurzweil, a fount of optimism on the topic, is sounding less sure now. Problems like model collapse, hallucination and innumeracy may be inherent in chatbots, which would curtail progress toward thinking like humans.