Lance Nizami’s Paper Proposal

The Homunculus Rides Again: why “information transmitted” in neuroscience tells us nothing

Presenting a stimulus to an organism may evoke firing of voltage spikes in afferent neurons connected to peripheral sensory cells. A neuron’s spike production may be predictable on average but be variable in the instance, hence appearing inherently “confused”. Neuroscientists attempt to quantify the “confusion” through “Information Transmitted”, their interpretation of Shannon Information Theory’s information transmitted.

Shannon’s theory’s conceptual basis was his “general communication system”. There, a recipient examines a received message, a string of received symbols (“outcomes”), at the destination. Computing the information transmitted also requires knowing what symbols were sent (“events”) – known by the sender at the source, but not by the recipient, unless both are the same person. In neuroscience, that “same person” is the experimenter, being both the source of the “events” and the destination of the “outcomes”. The experimenter is hence the person who computes “Information Transmitted”, the observer.

Note importantly that the “reception” side of Shannon’s system is a mirror-image of the “transmission” side. But neurons lack the “reception” side, which (otherwise) would perform decoding. The brain, however, need not “decode”, and shows no such action; nonetheless, neuroscientists held to the whole of Shannon’s system, and so (perhaps unconsciously) revived an old concept: Homunculus.

Homunculus mimics the experimenter by observing spikes, hence allowing spike-statistics-based computations of discriminability (whose precision is supposedly indicated by “Information Transmitted”). Note that Homunculus hence needs legs, in order to roam the brain; arms, in order to isolate neurons; and eyes, with which to observe spike-firing. Of course, Homunculus needs a brain, in order to coordinate those actions, and to consequently “discriminate”. Two questions thus arise. First, if Homunculus discriminates for us, is Free Will an illusion? In fact, some of our “discriminations” are consciously performed and some are reflexes, but all form a continuous stream – which would require an entire population of Homunculi, each “discriminating” among different things. This provokes Question Two: how does Homunculus make decisions? Homunculus has legs and arms and eyes and a brain, etc. – ultimately a fully-formed human being. Consequently, the head of Homunculus must contain a sub-Homunculus who performs computations on behalf of Homunculus – and so on and so forth, an infinity of Homunculi. This is plainly absurd, not least because any “discrimination” would potentially require an infinity of time and resources. Hence, Homunculus must be beyond human. The obvious candidate is a human invention called “God”, revealing Homunculus as a superstition.

Homunculus is unreal. Functions attributed to Homunculus, such as spike-observing and the computation of “Information Transmitted” which it allows, are done by the experimenter/observer. As such, a change in the experimenter/observer’s interaction with the organism – say, by changing “events”, or by redefining “outcomes” – could change the magnitude of “Information Transmitted”, and its maximum, “Channel Capacity”. And this, indeed, is what happens. Altogether, “Information Transmitted” expresses, at best, how well neuroscientists can communicate to themselves through [another species’] neurons. Neuroscientists fail to recognize this, instead continuing to falsely infer “the neural code” from comparisons of “Information Transmitted”.

Cybernetic traditions:

  • 8) Neurobiology; consciousness studies
  • 1) Computer science; AI; robotics

1 thought on “Lance Nizami’s Paper Proposal

  1. Peter Cariani

    I have worked on the neural coding of pitch now for over 2 decades and let me assure you that nobody involved in this kind of research believes in homunculi in the brain. It would be like accusing physicists of believing in demons because they construct thought experiments like Maxwell’s Demon. They may have very vague ideas about how distinctions are made, but most would subscribe to some sort of explanation based on distributed neural networks.

    I do think that assessments of informational channel capacity and arguments about informational optimality are premature if the nature of the neural codes is unknown. In the vast majority of systems this is presentlly the case. Information rate depends on the nature of the code and the presence of a decoding mechanism in the receiver. Also rate of information transmission is relatively uninteresting — what we want to know is how different aspects of stimuli are distinguished and why they have the perceptual structure and qualities that they do (e.g. why do we hear octave similarities?).

    WHere arguments involving optimal use of information CAN play a useful role is in eliminating candidate codes — if the organism can make much finer distinctions (say for pitch) than the system can support under a given putative coding scheme (such as rate-place coding at high sound levels), then that putative coding scheme can be falsified — thrown out of serious consideration. But it does not work the other way, that simply because a coding scheme is more efficient than another, that this proves that this must be the way that brain encodes that kind of information.

    The only way to identify a neural code is to examine which aspects of neuronal activity covary systematically with perception — for example, which aspects of neuronal activity stay the same when the stimulus is changed, but the percept remains the same? In the case of pitch, many different stimuli with different spectra evoke the same low pitch (at their fundamental frequency).

    I do agree with you that information transmission capacity per se has nothing to say about whether the changes in neuronal activity have any functional significance.

    But this should not prompt anyone to give up on looking to identify neural codes — to do so is to give up on trying to understand how brains work.


Leave a Reply

Your email address will not be published.