In numerous articles and a book, the neuroscientist Giulio Tononi has advocated the information integration theory of consciousness. He claims that consciousness in humans and other entities is the result of their ability to integrate information of different kinds. This claim seems plausible at first glance, because our conscious experiences do integrate various kinds of information, as when my being aware of shooting a basketball combines sights, sounds, touches, and muscle movements into a unified experience. Consciousness occurs in anything that has more information as a whole than in its parts. 

On closer examination, however, Tononi’s information integration theory (IIT for short) has numerous flaws. I critiqued Tononi’s theory in a recent article, but Tononi has a new publication LINK and archived paper that require additional scrutiny.  IIT is still open to the following objections. 

1.  IIT attributes consciousness to entities such as photodiodes and cell phones that show no behavioral evidence of being conscious. Tononi bites the bullet, but clarifies that he does not make the panpsychist claim that everything has a degree of consciousness. For example, whole countries and some simple computers do not information integration, he says, Smartphones clearly do integrate information, in that the whole phone ties together information gathered by various parts, including WiFi, cellular data, camera, microphone and keyboard. Similarly, I think that countries like Canada have governmental agencies that have more information than their parts such as cities.  So IIT attributes consciousness to entities that display no evidence of consciousness. IIT even implies that individual neurons are conscious, because they integrate information beyond their proteins.  

2.  IIT is extremely vague. Information for Tononi is not the precise mathematical notion of Claude Shannon, but a richer meaning-related phenomenon in systems in which there are “differences that make a difference”. He means causal differences, but never says what these are. Integration occurs when mechanisms are not reducible to independent components, but this is useless without some characterization of reducibility and independence. 

3.   IIT’s mathematical characterization is flawed.  Mathematical formulations are meant to clarify ideas, not obscure them, but it is extremely difficulty to understand how to calculate the quantity PHI that Tononi says is a measure of information.   In my 2014 article, I described problems in computing PHI, and Scott Aaronson  has identified similar problems.  We both conjectured that calculating PHI would be computationally intractable, and Tononi’s 2014 article confirms that “the present analysis is unfeasible for systems of more than a dozen elements”.   Brains have billions of neurons, and modern computers have billions of transistors, so we cannot even begin to calculate their quantity of information integration. 

4.  The axioms of IIT are not self-evident, contrary to Tononi's claims.  The five central axioms of IIT may be true, but they are not self-evident.  The first axiom is that consciousness exists, whose lack of self-evidence is clear from the fact that many smart people have denied it.    Behaviorist psychologists such as B. F. Skinner have tried to expunge consciousness from science, and some philosophers such as Georges Rey have also doubted its existence.   I think that consciousness does indeed exist, but it takes much evidence and theoretical argument to defend this inference.  I argued in by book The Brain and the Meaning of Life that there are no self-evident truths.

5.   IIT is less plausible than the alternative hypothesis that consciousness results from specific neural mechanisms such as semantic pointer competition (Thagard & Stewart), convergence zones (Antonio Damasio), and broadcasting in a neuronal workspace (Stanislas Dehaene).   Neural mechanisms can explain everything that IIT can, and more, without overextending consciousness to cell phones.  Consciousness is an emergent process in large neural networks, not a quantity in individual neurons.  

Tononi deserves credit for trying  to give a scientific explanation of consciousness, without descending into the obfuscation of the so-called hard problem of consciousness, which I made fun of in my last blog post. Thought experiments  do not provide evidence for conclusions, but can be useful for showing incoherence in alternative views.  Hence I compared consciousness to life, whose mechanisms were well understood by the end of the twentieth century.    It may well take decades or even centuries before a similarly extensive understanding of the neural mechanisms of consciousness is achieved.   But the abstractions of information integration theory only get in the way of developing a scientific theory of consciousness.   

You are reading

Hot Thought

Why Is There Evil?

Evil actions result from breakdowns in mechanisms at many levels.

What Causes What?

Good decisions require understanding of causality and inference.

Was It Meant to Be?

Does it make sense when people say that something was just not meant to be?