Stephen Hawking on model-dependent realism

bubble chamber trace

Traces left in a bubble chamber by interacting subatomic particles. From Hawking and Mlodinow 2010 The Grand Design.

With Stephen Hawking’s recent passing, it made me think about the influence that he has had not only as a physicist but also as a writer for the general scientific audience.  In particular, there is a wonderful discussion of models and their role in the epistemology of science in his book The Grand Design (2010, with Leonard Mlodinow).

In this book, Hawking goes through theories, or models, of the physics of the universe from the ancients, to Newton, to Einstein, to quantum mechanics, to present-day searches for unified theories in physics.  The book is really more about epistemology than physics.   The epistemology that Hawkins adopts and describes as being at the foundation of science is “model-dependent realism.”   To understand the significance of this view, it helps to understand the various positions that philosophers of epistemology have taken over the centuries.  Historically, and from a very coarse perspective, they have fallen into two main schools of thought:  Idealism and realism.  Idealists argue that humans can have no real knowledge of the external world; we can only have direct experience of our ideas (even our perceptions are just sensations in our mind); thus we can never be sure if an external reality really exists.  Realists take the position that an external reality does exist and that we can have knowledge of it.

In Hawking’s model-dependent realism, our sense organs provide input, and we build a model or models of the world.  An external reality does exist, but we do not have direct access to it.  We work through our models.  There are no model-independent tests of reality.  When I first read this book a few years ago, this idea immediately appealed to me because I have long held the view that models are the way we think (everyone, not just scientists, although scientists are more explicit about it).  When you think of a hydrogen atom, what do you think of?  You probably picture the Bohr model of the atom in your mind.  A physicist may visualize a probability distribution of the electron’s position, described by a 3-d Bessel function.  Are atoms real in the world, external to human thought?  Yes.  But when you think of an atom, what you think of is a model.  If you designed experiments to better understand what atoms are, they would be experiments based on your model and designed to test or improve your model.  I argue that this is also true when you mentally combine various sensations of light and shade and touch and call it a brown table (Bertrand Russel’s example), and it’s also true when you think of an ecosystem.  The object of your reasoning is a model — you have built it from learning and observation, but nonetheless it is a model.

Hawking thus places models at the center of our understanding in the sciences.  Interestingly, John Holland, the University of Michigan professor who was a pioneer of complexity science did so as well.  In Holland’s book Emergence (1998), he wrote this:  “Although model building is not usually considered critical in the construction of scientific theory, I would claim that it is.  Every time a scientist constructs a set of equations to describe the world, such as Newton’s or Maxwell’s equations, he or she is constructing a model.”  And later, “In one sense, all of science is based on model construction … To understand emergence, we must understand the way in which models in science and elsewhere allow us to transcend the knowledge that went into their construction.”

Also interesting is the fact that with rapid advances now occurring in the development of artificial intelligence, the question of models and model-dependent realism comes to the fore in a very practical way.  Russel and Norvig, in their widely-used textbook on AI, from the very start outline the key factors needed to design an AI agent that has sensors and actuators and can perform actions in the world in order to achieve goals.  They describe the central importance of the AI agent’s internal model of the world.  In a tangible example, the Mars Curiosity Rover uses its stereoscopic visual inputs to build a 3-d model of the topography of the surrounding surface of Mars (including remembering the part of the surface that it outside of its immediate vision), which it uses to autonomously plan its route.

It is fascinating to consider how modern-day scientists are taking what used to be a branch of philosophy and bringing it squarely into the sciences, catalyzed by the ability that we have now to develop computer models as emphasized by both Hawking and Holland.