There are two questions here. First there is the question of whether, if we accept the Bayesian brain model, we are all psychic. The second, and more interesting, question concerns whether the Bayesian brain model works as a model of human understanding.

Question 1: Does New Scientist Article, “The Future, Predicted by Your Brain,” Provide Evidence for Everyone Being Psychic?

The God Example. In his answer to my dark matter-dark energy thought experiment, Skrivener pointed out the act of naming something god does not make it god. This, the nominal fallacy, is common. Just because something is named or categorized is not an explanation. By naming the gap “god,” “nature,” or some undetectable “dark matter” or “dark energy” nothing is gained unless the name itself conveys specific meaning. Astrophysicists have chosen dark to represent the (directly) undetectable attribute of both dark matter and dark energy: Neither can be “seen” at any wavelength of the electro-magnetic spectrum. They have chosen matter and energy to convey the various theorized effects dark matter and dark energy have upon normal matter, gravity, the electromagnetic spectrum, and accelerated spatial expansion and the very brief period of cosmological inflation (exponential expansion). There is good reason to choose these terms, as their hypothesized existence does begin to explain observed phenomena; nevertheless, the terminology does not completely define the is of “what is dark matter?” and “what is dark energy?”. In science incomplete definitions can lead to more research, and ultimately deeper understanding. In religion this same incompleteness (gaps in our understanding) can lead to the invocation of “god” as explanation. Unlike “dark matter” and “dark energy”, though, “god” offers no insight into the observed phenomena.

The Meaning of Psychic. Both as noun and adjective psychic has the peculiarity of being a catch-all word for persons who possess supernatural senses or powers (a psychic) or the powers themselves (psychic abilities). Like “god” in the example above, “psychic” is loaded with meaning, yet is meaningless. When another says of someone else she/he is psychic we immediately, although only superficially, understand what is meant: She/he has an uncommon numinous connection to the universe that is beyond the ordinary state of consciousness. Yet, when pressed to define exactly what the connection is for a given case (a specific psychic)—when pressed to define how their psychic-ness manifests itself—we are at a loss. Why? Because—as with Skrivener’s god, psychic can mean whatever in god’s name one chooses.

What psychic does not mean is anything to do with normal or ordinary sensation, cognition, experience, consciousness and brain function: Which is precisely what the New Scientist article is describing. The article is short, and so I include it here:

The Future, Predicted by Your Brain [title] IT'S like remembering the future. Our brain generates predictions of likely visual inputs so it can focus on dealing with the unexpected. Predictable sights trigger less brain activity than unfamiliar stimuli, bolstering the view that the brain is not merely reactive, but generates predictions based on the recent past. "The brain expects to see things and really just wants to confirm it now and again," says Lars Muckli at the University of Glasgow, UK. He and Arjen Alink at the Max Planck Institute for Brain Research in Frankfurt, Germany, asked 12 volunteers to focus on a cross on a screen, above and below which bars flashed on and off to create the illusion of movement. To test a predictable stimulus, a third bar would appear in a position timed to fit in with the illusion of smooth movement. For the unpredictable stimulus it would appear out of sync. fMRI scans showed that the unpredictable stimulus increased the activity in parts of the brain which deal with the earliest stages of visual processing (Journal of Neuroscience, vol 30, p 2960). The finding supports the "Bayesian brain" theory, which sees the brain as making predictions about the world which it updates when new information comes in.

Notwithstanding the questionable title, this is basic neuroscience and neuropsychology; nothing numinous at all. No explanation of, or even allusion to, the extra-ordinary. The authors of the original research are studying the neuronal activity associated with the most banal of visual sensory inputs. The findings: Predictable stimuli produce less neuronal activity than unpredictable stimuli. Why? Well, if every stimulus triggered the same, heightened level of neural activity, the brain would be rapidly overwhelmed and unable to process new inputs. That novel, unpredicted stimuli receive more neuronal attention (activity) seems an analog to human behavior in general. We pay more attention to novel situations while commonplace situations sort themselves out with little or no conscious attention. We could easily reverse this statement: A person’s behavior toward novel and banal stimuli is analogous to that of our sensory neurons to novel or banal visual inputs. Indeed, the results suggest human behavior and human sensory neural processes follow similar rules. There is nothing psychic about this. We certainly do not get this level of understanding by throwing about the catch-all term “psychic.”

One could argue “Our brain generates predictions of likely visual inputs so it can focus on dealing with the unexpected” demonstrates a psychic ability by predicting likely future inputs. However, there are two problems with this reasoning. First, the term “predictions” was chosen by the author(s) to represent the neural activity. By choosing this term they have primed the reader to see the research through a coloured lens—a specific model. They could have used another model, one that does not describe visual neuronal activity in terms of prediction. This is certainly not necessarily an issue. Good scientific research usually sets out to test a hypothesis based upon an existing model; often in an attempt to test the model. However, while the results may support the model, it may not exclusively support the model. Other models may also be supported by the data. Second, their research shows the predictions are of likely future inputs. This is interesting because psychic generally refers to the ability to divine or predict the unlikely. A psychic claims to foresee what a non-psychic cannot normally anticipate. In contrast, this research shows a tendency for our brain to predict (and ignore) what is likely. Again, calling this psychic is an example of not only the nominal fallacy, but of confirmation bias. To call this evidence for psychic ability requires one to twist the results to fit his/her biased belief in precognition. Ironically, in doing so, the numinous meaning of “psychic” (what makes “psychic” meaningful) is lost and the psychic now predicts the same mundane events that everyone else predicts. By accepting this research as evidence for psychic ability, psychic becomes self-contradictory, internally inconsistent, self-nullifying, and as meaningless as my nominating god the placeholder for dark energy and dark matter.

Question 2: Does the Bayesian Brain Explain the Brain?

The Bayesian Brain is relatively new; Bayesian statistics is not. Bayesian statistics is a powerful method of hypothesis testing. The key to this statistical approach is prior knowledge (evidence). Over time evidence related to a given phenomenon or set of phenomena will accumulate. Some of the evidence will support a given hypothesis, some will not. Each piece of evidence must be taken into account. There can be no picking and choosing. From this matrix of evidence a prediction (hypothesis) is made and measured against objective reality. Over time a vast array of evidence from wide range of sources will be gathered. The more variables this array of evidence contains (the more robust the research) the more predictive the derived hypothesis/hypotheses will be.

An example from Karl Friston, a leading proponent of the Bayesian brain model:

Let’s take a simple example to clarify how this works. Let’s say climatic conditions have recently been dry and we are interested in the hypothesis that it will rain tomorrow. We know that the sky has been clear all day today and that this usually gives some weight to supposing it will not rain tomorrow. However this evening we get some new data, a weather forecast predicting rain tomorrow. What effect should this new data have on our confidence level on our hypothesis that it will rain tomorrow? Obviously the new data should boost our confidence in this hypothesis but exactly what adjustment to our confidence should be made?

Using Bayesian Probability we can calculate the exact adjustment we should make. First we need to quantify three related probabilities that can be estimated from historical data:

Probability it will rain tomorrow given that it was clear all day today. We can estimate this by looking through previous historical weather forecasts and data. Let’s say we look at the last 100 clear days and then at each of the following day's weather and we find on 32 of those 100 subsequent days it rained. We can estimate this probability at .32. Before we consider the new data (in the form of the weather forecast) we should assign a .32 probability as our confidence level of it raining tomorrow.

Probability of a weather forecast for rain tomorrow given that it was clear today. Again we can estimate this probability by looking through historical weather data and forecasts. Let’s say we count that the weatherman predicted rain for days following a clear day on 40 of those 100 days giving us a probability of .4.

Probability of a weather forecast for rain tomorrow given that the sky is clear today and that it rains tomorrow. This is an estimation of the probability of the weatherman making the right forecast in this situation (when there is a preceding clear day). Let’s say the weatherman is pretty good and we see from the historical data that he called it correctly on 24 of those 32 days yielding a probability of 24/32 or .75.

Now we have all the numbers we need to calculate the revised probability of our hypothesis being true in light of this the new data. To do the calculation let’s follow the terminology of Bayesian Probability and analyze the problem by identifying three variables: H, X and D.

H is our hypothesis; it will rain tomorrow

X is our relevant prior knowledge; the sky was clear today

D is our new data; a weather report predicting rain tomorrow.

We form the following conditional probabilities:

P(H│DX) Probability of it raining tomorrow given that the sky is clear today and the forecast is for rain. This is the confidence we should have in our hypothesis and it is the item we are going to calculate.

P(H|X) Probability it will rain tomorrow given the sky is clear today =.32

P(D|HX) Probability of a weather forecast for rain tomorrow given that the sky is clear today and that it does rain tomorrow. = .75

P(D|X) Probability of a weather forecast for rain tomorrow given that the sky is clear today = .4
The Bayesian formula that relates our correct confidence level to our hypothesis, our prior data and our new data is:

P(H│DX)=P(H|X)(P(D|HX))/(P(D|X))

Substituting in our numbers: P(H│DX) = .32*.75/.4 =.6 Given the new data we should revise our confidence level in the hypothesis that it will rain tomorrow from .32 to .6.

It is great to put some numbers into a formula, do the calculation, and get an answer but does our answer make sense? Can we check it? First let's remember that the probabilities used in this formula were created by counting historical data for three values and then using these three values to calculate our answer. There are mathematical theorems proving that this calculation will always give the right answer but it is often possible to check this answer by directly counting the pertinent cases in our data. In this case in our sample of 100 clear days we can count 40 them where it rained the following day and 24 of those where the rain was forecasted. Our answer, the correct probability estimate that rain actually occurs when it is forecasted for days following clear days is obviously 24/40 or .6, confirming our answer.

This example illustrates how prior knowledge (data) of probabilities can result in a prediction that may or not correspond to reality. In this case more data resulted in a probability that it will rain tomorrow of .6 only after additional data was added. The next step is to test this prediction against reality. If after a number of years, the probability is indeed .6, then our model is sound. If the reality is statistically different, then we must look for more variables (in other data) which will further increase the predictive power of our model. Additionally, over time new evidence will arise which must be accounted for by our model. Take again our dark matter example. When dwarf galaxies were shown to lack the mass to exist, new models of gravity and matter (cold dark matter and warm dark matter) were developed.

This is exactly what Bayesians propose happens in the brain. The same probability calculations occur. This, though, is where things get blurry. At the extreme, Friston and others argue this is (or at least will one day become) the unified theory of the brain. They argue our prior experiences are encoded as probabilities and combined via a Bayesian statistical algorithm to make predictions about the world. The neurological goal is to eliminate “free energy” (a term borrowed from physics) or in psychological terms, “surprise.” If not enough data is available for us to assess a situation, we attend to data collection for this relatively novel situation. For example: A motion occurs to my side. I turn my head to assess the motion so I can apply prior knowledge (encoded probabilities) and predict the “future” path this dynamic situation will develop along. And here is the rub. This is the most complicated example one can give of research done thus far to test this grand theory.

The majority of the research done has involved mostly simple visual sensory and sensory-motor tasks. Friston and colleagues (see references) have conducted experiments that show changes in cortical activity in response to novel stimuli emanating from lower (more primitive) regions and reduced sensitivity in these more primitive sensory regions when presented familiar stimuli usually experienced within a specific perceptual context. Friston’s explanation is the cortex predicts specific sensory stimuli will occur within a given situation and if those predictions are accurate, the sensory-motor brain regions will be less active (less responsive). In his terms, the more probable (predictable) a stimulus is, the less surprise the sensory system will experience and the less active the motor cortex will become. Bayesian statistics are the brain’s mechanism of action. The neocortex computes the probability of an upcoming stimulus. If a novel, unpredicted, stimulus occurs, then the probability of the expected stimuli occurring within the same context must be adjusted. Variables must be added, complicating the algorithm. If the predicted stimulus occurs, then the original probability increases (the prediction is strengthened). This seems to work well for simple stimulus-response scenarios; however, complex daily tasks have not been tested. A Bayesian explanation complex disease such as dementia is not obvious. How does Friston account for morality, rationality, the arts, the sciences? What is the probability of mystical experience? What does such a question even mean? Can the Bayesian brain explain consciousness?

Long before supporters of the Bayesian brain theory can tackle these more abstract questions, they will have to do a great deal more research and experimentation moving beyond simple sensory-motor (specifically visual) paradigms to demonstrating the explanatory power in areas of memory, learning, cognition, disease, and other complex psychological processes. In addition for humans, olfaction, gustation, tactition, audition, and Equilibrioception are more primitive and neurologically simpler senses in contrast to vision. These senses are less nuanced. Human vision involves a greater proportion of the brain, greater fidelity, and offers more detail about the world than any other sense. Where vision may be mediated by top-down (neocortical-sensory) processes, the Bayesian hypothesis must not only show the other sense modalities are so mediated, but also Bayesian probability is the mechanism by which they are modulated. Are these modulated by the neocortex in the same way as vision? Any unified theory of the brain, including the Bayesian brain theory, must broaden its area of study to all sensory modalities before it can hope to claim the status of “grand theory.”

Until it can effectively address all of these questions the Bayesian brain is not a unified theory (a grand theory) of the brain. Additionally, one is left with the uneasy feeling, Friston and supporters of the Bayesian brain are merely expressing old ideas in new terms. Often neuroscientists talk of neural connections becoming “stronger” as commonly repeated sensory, perceptual, cognitive, and emotional events recur. When a rare event occurs, the neuronal connections are fewer, and it is only repeated similar events which produce a greater number of connections. Usually this is referred to as neural plasticity. Over time neural adaptation occurs for events that do not change, or are repeated regularly. For instance, we usually do not attend to our legs when standing, or bottoms when sitting: Our nervous system adapts to the ongoing or repeats regularly. Is the Bayesian probability model just a new name for these same phenomena? Is this the nominal fallacy again? Possibly—and if so—the Bayesian brain may not reveal the mechanism by which the brain works, but another analogy that helps us understand aspects of brain function, and not the actual underlying rules or mechanism.

This is not to suggest the Bayesian model must be thrown out. Indeed, Bayesian brain theory offers an interesting model for brain function…a metaphor. Future research may show the Bayesian brain to actually be a functional model, describing the actual mechanism(s) of brain function as opposed to a loose metaphor. It may be the Bayesian position works for some neural systems, but not others. There is some evidence Bayesian brain may or may not work in comparative neuropsychology (animal models), though more research is required. Bayesian probability may prove itself in the realm of artificial intelligence. Who knows, algorithms developed from Bayesian statistics could finally lead to an artificial intelligence capable to pass the Turing Test. Regardless, Bayesian brain theory offers an interesting perspective on the brain. My suspicion, however, is it is unlikely to become a unified theory of the brain any time soon, if at all. Other, simpler models that explain the same phenomena will likely win out over this interesting, yet awkward model. The law of parsimony (Ockham’s razor) should be the lens through which such theories are judged. Or, as so often happens, I could be wrong and Friston and the Bayesian brain theory could carry the day. In the end, time and well-executed scientific research will determine the Bayesian brain’s fate.