No, Those "Neural Representations of Experience" Aren't Drifting -- They Simply Never Existed

 Claims by neuroscientists that they have found "representations" in the brain (other than genetic representations) are examples of what very abundantly exists in biology: groundless achievement legends. There is no robust evidence for any such representations. 

Excluding the genetic information stored in DNA and its genes, there are simply no physical signs of learned information stored in a brain in any kind of organized format that resembles some kind of system of representation. If learned information were stored in a brain, it would tend to have an easily detected hallmark: the hallmark of token repetition.  There would be some system of tokens, each of which would represent something, perhaps a sound or a color pixel or a letter. There would be very many repetitions of different types of symbolic tokens.   Some examples of tokens are given below. Other examples of tokens include nucleotide base pairs (which in particular combinations of 3 base pairs represent particular amino acids), and also coins and bills (some particular combination of coins and bills can represent some particular amount of wealth). 

Other than the nucleotide base pair triple combinations that represent mere low-level chemical information such as amino acids, something found in neurons and many other types of cells outside of the brain, there is no sign at all of any repetition of symbolic tokens in the brain. Except for genetic information which is merely low-level chemical information, we can find none of the hallmarks of symbolic information (the repetition of symbolic tokens) inside the brain. No one has ever found anything that looks like traces or remnants of learned information by studying brain tissue. If you cut off some piece of brain tissue when someone dies, and place it under the most powerful electron microscope, you will never find any evidence that such tissue stored information learned during a lifetime, and you will never be able to figure out what a person learned from studying such tissue.  This is one reason why scientists and law enforcement officials never bother to preserve the brains of dead people in hopes of learning something about what such people experienced during their lives, or what they thought or believed, or what deeds they committed.    

But despite their complete failure to find any robust evidence of non-genetic representations in the brain, neuroscientists often make groundless boasts of having discovered representations in the brain. What is going on is pareidolia, people reporting seeing something that is not there, after wishfully analyzing large amounts of ambiguous and hazy and ever-changing data. It's like someone eagerly analyzing his toast every day for years, looking for something that looks like the face of Jesus, and eventually reporting he saw something that looked to him like the face of Jesus.  It's also like someone studying endless photos of clouds, looking for a shape that looks like an animal shape, to try and back up his belief that the ghosts of dead animals live in the sky. 

Claims that there are non-genetic representations in the brain often appeal to the existence of "place cells." Nature does not tell us that there is any such thing as "place cells."  We merely know that there are cells, and that neuroscientists started to use the term "place cells" for some small set of cells, to try to spread ideas that cells help to represent some place where an organism has been in. 

What would a convincing experiment showing representations in a brain look like? It might work like this.  You might have some "blinded" analysts who had no idea of what claims were being made about representations, and no idea of what the goals, procedures or suspicions of the neuroscientists were. Such analysts might be shown some data such as EEG data or brain scan data, and the analysts might be told, "We think that in this data may be representations of something an organism observed or experienced -- can you guess what that was?"  If most of the analysts gave the same answer (such as describing the layout of a particular type of maze that a rat ran through), that might be good evidence that neural representations had been found; for there would be very many thousands of possible answers, so we would not expect most of the answers to coincidentally agree.  

Nothing like that has occurred in any of these experiments claiming evidence for brain representations.  Instead, we typically have some procedure vastly less convincing, in which  a scientist (who knows that some rodents observed some particular thing) attempts to sift through lots of data, looking for something that he can claim is evidence for a representation of that thing, rather like some person walking through a forest of 1000 trees, eagerly looking for some tree that has a face shape on it. 

There is no robust evidence for any spatial representations in the brain. "Place cells" are a social construct of neuroscientists, not something with an objective reality in nature.  What would we expect to find if such claims of representations in the brain are not well-founded? One thing we might expect to find is that there would be great inconsistency in the descriptions of such claimed representations, with little replication of the same results. That is just what seems to have happened. Neuroscientists have invented an inaccurate phrase to describe such inconsistencies. The phrase they are now using is "representational drift." The phrase is an inaccurate one, because there are no actual representations that are drifting. It is simply claims of such representations that are varying, or failures to replicate the original claims of representations. 

The science journal Nature just published an article on so-called "representational drift." We have this title and subtitle:

Below is the text (for those reading this post in a non-English translation:

"The brain’s code seems to be in constant flux. Neuroscientists are baffled

Neurons fire much more erratically than researchers thought. What does that mean for how the brain works?"

We read this in the article:

"Over a single day, neurons in the parietal cortex, a hub for processing sensory information, fired predictably in response to specific things, such as the position of the mouse in a virtual maze. But over the course of a few weeks, even though the task of navigating the maze remained the same, these activity patterns underwent major reorganization1. Some of the neurons stopped firing in response to stimuli that had previously activated them; others did the reverse....Since then, many more researchers have reported evidence for neurons changing how they respond to certain stimuli or behaviours over time, a phenomenon that neuroscientists have dubbed representational drift (see ‘How neuronal activity drifts over time’). Evidence has been found for it in various brain regions and when using several different techniques....Drift kept turning up in more areas, including some unlikely places, such as the visual cortex and the olfactory cortex. For instance, finding representational drift in the piriform cortex6, a key region for processing olfactory information, was 'completely contrary to how we and all of our colleagues had thought about that cortex', says Fink. Scientists had thought that activity in the neurons in this region would have to be fixed to enable animals to identify smells, but Fink and his colleagues discovered that such activity patterns were almost completely unrecognizable one month after the original recordings."

We have in the Nature article this schematic depiction of this "representational drift." A set of brain cells are tracked on different days when some animal is undergoing the same stimulus or having the same experience. There is no consistency in the responses, with the recorded brain waves varying from day to day. 

We have in the Nature article some goofy-sounding groundless speculations about imagined deep significance of these claimed "representational drifts":

"One hypothesis is that representational drift, at least in the hippocampus, helps the brain to keep track of time. In fact, years before the word ‘drift’ was widely used, Jill Leutgeb, a neuroscientist at the University of California, San Diego, and her colleagues observed shifts in the activity of groups of neurons in the hippocampus over time and suggested that they might encode the passing of time between events7.

Other have posed similar ideas: Yaniv Ziv, a neuroscientist at the Weizmann Institute of Science in Rehovot, Israel, suggests that drift is required for the 'time stamping' of events for long-term memories. And Denise Cai, a neuroscientist at the Icahn School of Medicine at Mount Sinai in New York City, says that drift might enable the brain to link memories that are formed close together in time."

The most likely explanation of what is going on with these claimed "representational drifts" is something much simpler. It is not that these claimed representations are drifting -- it is that they never existed. What happened is that neuroscientists very eager to find some evidence of non-genetic representations in a brain published claims of representations that were not well-founded, and were mere examples of pareidolia, like someone claiming to find an image of Jesus in his toast. Since there never was any sound basis for the original reports of representation, it is no surprise that they fail to be well-replicated. 

Neurons in the brain fire very frequently (between about 1 and 200 times per second), with the firing occurring at unpredictable intervals. Anyone analyzing the firing rates of 500 neurons in some brain location will be able to find some neurons that fired more frequently when the organism having those neurons had some experience.  It is a mistake to ever claim that such higher firing rates are representations of the experience the organism is having, or any part of such an experience. The next time such an experiment is tried, there will be some different set of neurons that have higher firing rates. This is exactly what we would expect under the assumption that the original report of "representations" was groundless, and that all that was going on was noise-mining of random, frequently fluctuating data. Noise mining is when someone analyzes random, rapidly varying data, and tries to suggest it has some deep significance that it does not actually have.  

The visual below is a schematic giving some idea about the variation in the number of times per second that neurons fire in some part of the brain. 

The infographic below also discusses variations in the firing rates of neurons. When it refers to 1 Hertz or 1 Hz , it means a firing rate of 1 time per second. When it refers to 200 Hertz or 200 Hz , it means a firing rate of 200 times per second. 

The infographic below shows some of the factors that cause random variations in the firing rates of neurons.

When firing rates in the brain are varying so often and so randomly because of factors like those depicted above, there are endless opportunities for people wishing to eagerly seek out correlations between firing rates somewhere and what an organism has  experienced,  someone who then wrongly calls this evidence of representations.  But when such correlations are claimed, we should not regard them as something unexpected, even if brains have no representations of anything someone has learned or anything someone is experiencing. 

A paper on so-called "representational drift" states this:

"Cells whose activity was previously correlated with environmental and behavioral variables are most frequently no longer active in response to the same variables weeks later. At the same time, a mostly new pool of neurons develops activity patterns correlated with these variables. Less commonly, cells previously correlated with certain variables become correlated with new variables. In one example of these findings, as mice move through an environment, a subpopulation of hippocampal place cells are informative about the animal's spatial position. Over time, the pool of place cells that make up this spatial code changes, despite the environment staying the same."

There is a simple and plausible explanation for such results:

(1) Brains do  not actually contain representations of anything learned or experienced by humans or animals, and no neuroscientist has ever produced a credible theory of how learned information could be stored as brain states or neural states. 

(2) Experimental claims that the brain contains such representation were never robust science, but involved experiments guilty of Questionable Research Practices such as way-too-small study group sizes, a lack of a blinding protocol, a lack of pre-registration, and subjective "see whatever you want to see" cherry-picking analysis by overeager correlation fishers guilty of pareidolia. 

(3) The reported "representational drift" is the kind of random variation and vanishing of reported correlations we would expect given a scientific result that was never robust, but was always based on dubious subjective analysis of randomly varying ambiguous data. and mere pareidolia (people seeing patterns in random data that they were hoping to see).  

Neuroscientists are using extremely misleading language when they claim to see something they call "superior activations" or "higher activations" or "activation patterns," and then pass off such claimed observations as evidence for representations. The concept of a "cell activation" is not even a solid one, and the term "higher cell activation" or "activation pattern" has no precise  meaning. Letting us know exactly when they are active, light switches have an "on" position and an "off" position; but cells have no such things. Representations involve tokens or symbols in which one thing stands for another.  If there were to exist some mere "higher activation" of a cell,  that would not be a representation.  If I see a naked woman on the street, there may be some higher activation of my eyes, with my eyes widening. Such a thing is not a representation of a naked woman.  

When the genetic code was discovered in DNA, that was not some  thing merely involving fleeting "higher cellular activations." That was a discovery of a massive storehouse of symbolic tokens in DNA -- particular combinations of nucleotide base pairs which represented particular amino acids.  Other than such a genetic code, there simply is no evidence of representations in the brain. There is zero robust evidence that brains store representations of any learned knowledge or any place where an organism has visited or any experience an organism is having. 

The latest example of a low-quality science paper making a false claim of observing representations in the brain is the paper here, entitled "Neural representation of action symbols in primate frontal cortex." We have an experiment in which monkeys had electrodes implanted  into their brains, to record brain activity while monkeys were drawing. The authors claim some correlation between brain firing spikes and what the monkeys were drawing, incorrectly claiming this as a "representation of action symbols." What is going on seems to be mainly  noise mining, eager correlation-fishing and pareidolia, the mining of correlations that you would inevitably be able to get even if no such symbols or representations existed in a brain. A halfway-reliable result in an experiment like this would require 15 or 20 animals per study group. The authors used only two monkeys. A sample size calculation would have revealed how very inadequate such a study group size was, but the authors do not report doing any sample size calculation. Much or all of the claimed correlation can be explained as mere EEG spikes caused by muscle movements, which are known to create spikes in EEG readings. Probably a good fraction of the claimed correlations are artifacts of the fancy data-juggling statistical rigmarole that was going on. We don't have here any decent evidence of neural representations, but we do probably have evidence of needless harm to primates being caused by scientists doing poorly-designed experiments. 

Neuroscientists Misspeak Badly When They Refer to "Representations"

Claims by neuroscientists that they have found "representations" in the brain (other than genetic representations) are examples of what very abundantly exists in biology: groundless achievement legends. There is no robust evidence for any such representations. 

Excluding the genetic information stored in DNA and its genes, there are simply no physical signs of learned information stored in a brain in any kind of organized format that resembles some kind of system of representation. If learned information were stored in a brain, it would tend to have an easily detected hallmark: the hallmark of token repetition.  There would be some system of tokens, each of which would represent something, perhaps a sound or a color pixel or a letter. There would be very many repetitions of different types of symbolic tokens.   Some examples of tokens are given below. Other examples of tokens include nucleotide base pairs (which in particular combinations of 3 base pairs represent particular amino acids), and also coins and bills (some particular combination of coins and bills can represent some particular amount of wealth). 

Other than the nucleotide base pair triple combinations that represent mere low-level chemical information such as amino acids, something found in neurons and many other types of cells outside of the brain, there is no sign at all of any repetition of symbolic tokens in the brain. Except for genetic information which is merely low-level chemical information, we can find none of the hallmarks of symbolic information (the repetition of symbolic tokens) inside the brain. No one has ever found anything that looks like traces or remnants of learned information by studying brain tissue. If you cut off some piece of brain tissue when someone dies, and place it under the most powerful electron microscope, you will never find any evidence that such tissue stored information learned during a lifetime, and you will never be able to figure out what a person learned from studying such tissue.  This is one reason why scientists and law enforcement officials never bother to preserve the brains of dead people in hopes of learning something about what such people experienced during their lives, or what they thought or believed, or what deeds they committed.    

But despite their complete failure to find any robust evidence of non-genetic representations in the brain, neuroscientists often make groundless boasts of having discovered representations in brains. What is going on is two types of things:

(1) Misspeaking and language abuse by neuroscientists, in which they misleadingly use the term "representations" to refer to mere fleeting blips of brain activity that may properly be called "neural correlates" but not honestly called "representations."

(2) Pareidolia, people reporting seeing something that is not there, after wishfully analyzing large amounts of ambiguous and hazy data. It's like someone eagerly analyzing his toast every day for years, looking for something that looks like the face of Jesus, and eventually reporting he saw something that looked to him like the face of Jesus.  It's also like someone walking in many different forests, eagerly looking for faces on trees, and occasionally reporting a success. 

Let's consider the first of these two things. You are using language in a misleading way when you refer to a mere correlate as a representation.  I will give an example. Imagine you do a study in which you photograph faces of American football fans when their team scores a touchdown. You might be able to detect a correlation between the times that a touchdown was scored and particular expressions of fans. You may find that football fans are more likely to look like this just after a touchdown was scored. 

But you would be speaking in a misleading way if you called such an expression a "representation" of a touchdown. The expression is merely one that is a correlate of a touchdown: it is an expression more often occurring when a touchdown occurs (and also when any successful play occurs, such as a long run or a long pass). 

There is in American football an actual representation of a scoring event: the sign that a referee makes when he places his two arms up in the air. So the sign below is an actual representation of a football touchdown or a field goal. 

It is a big mistake to be using the term "representation" when referring to something that is merely a correlate. This is mistake that today's neuroscientists keep making. 

Let us consider some EEG readings or brain scans that may be taken of some organism when the organisms see something and then respond to that thing. Visual perception causes greater activity in the visual cortex of that brain. Such an increase is merely a correlate of the seeing of something, not a representation of what was seen.  When presented with some stimulus such as a piece of cheese put in front of it, a mouse may move its muscles to move toward the food. Muscle movements produce spikes in EEG readings. But if you pick up a spike in an EEG reading when a mouse moves toward some cheese, that is not correctly called a representation of the cheese. It is instead merely a neural correlate of responding to the cheese. 

The paper "Brain-wide representations of prior information in mouse decision-making" is an example of a recent science paper misspeaking  when using the term "representation." No representations were actually found. All that was observed were neural correlates. The authors confess, "It remains unclear where and how prior knowledge is represented in the brain."  The truth is that no evidence has ever been found of learned knowledge stored anywhere in the brain of any organism. 

The paper "A brain-wide map of neural activity during complex behaviour" is is another example of a recent science paper misspeaking  when using the term "representation." The paper studied mice with implanted electrodes. The authors have section titles such as "Representation of visual stimulus," "Representation of choice," "Representation of feedback," and "Representation of wheel movement." All of these uses of "representation" are illegitimate. 

Let's look at each of these sections:

"Representation of visual stimulus": It soon becomes clear that all that the authors have picked up is fleeting correlates of something a mouse has seen. We read, "a decoding analysis based on the first 100 ms after stimulus onset revealed correlates of the visual stimulus side in many cortical and subcortical regions." The 100 ms refers to a mere tenth of a second. 

"Representation of choice": All that the authors have picked up is fleeting neural correlates of muscle movements associated with the choice. We have another observation window of 100 milliseconds, merely a tenth of a second.  The analysis going on is correctly described as "neural correlates of muscle movement associated with a choice," but not correctly described as "representation of choice." 

"Representation of feedback": All that the authors have picked up is fleeting neural correlates of mice getting some reward, within an observation window of a fifth of a second. This is no actual "representation of feedback." 

"Representation of wheel movement": All that the authors have picked up is fleeting neural correlates of muscle movements. Such neural correlates are not correctly described as representations. 

How in theory might you get actual evidence of a brain making a choice? It could not be through any research involving mice. It could only be done with humans. It might work rather like this:

Humans would have their brains scanned, or be connected to EEG devices allowing their brain waves to be read. The human subjects would be offered a choice between two different foods. The subjects would be instructed to close their eyes and remain motionless, as a countdown timer ticked down: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0. The subjects would be instructed that when the timer reached 0, they should silently make a choice in their minds as to which of the foods they would select, and then remain motionless for another 30 seconds.  The interval of the countdown could be varied, with it sometimes being 20 seconds, sometimes 20 seconds, and sometimes 10 seconds. 

The data would then be analyzed by blinded analysts, who did not know what moment corresponded to the timer reaching 0, and did not know which choice was made.  The analyst would attempt to figure out which of the two choices was made, purely from the brain data. The analyst would also attempt to figure out at which second the choice was made, purely from the brain data. 

The experiment would fail. Excluding the neural correlates of muscle movements, brains do not show any sign of being involved in decisions. When a person makes a decision without moving his muscles or changing his facial expression, there is no neural correlate of this action. Brains have no actual representations of decisions, and no actual representations of beliefs, or anything someone has learned. 

"Consciousness Theory Showdown" Shows Mainly Shady Neuroscientist Sleight-of-Hand

When talking about the problem of explaining human minds, those in academia love to use the term "problem of consciousness." But is is a huge fallacy to think there is merely some "problem of consciousness" when there is a trillion times bigger "problem of explaining human minds, human mental capabilities and human mental phenomena."  Once you realize this, you may realize that presenting some "theory of consciousness" can never do much to solve the explanatory problems in the philosophy of mind, which are huge and "all over the place."  

The trick of posing a mere "problem of consciousness" is a ridiculous ruse. A human being is not merely "some consciousness." A human being is an enormously complex reality, and the mental reality is as complex as the physical reality.  You dehumanize and degrade human beings when you refer to their minds as mere "consciousness." The problem of human mentality is the problem of credibly explaining the forty or fifty most interesting types of human mental experiences, human mental characteristics and human mental capabilities.

It is always a silly, stupid trick when someone tries to reduce so complex a reality to try and make it sound like the faintest shadow of what it is, by speaking as if there is a mere "problem of consciousness," and talking as if humans are just "some consciousness" that needs to be explained.  Such a shabby, pathetic trick (which can be called consciousness shadow-speaking) is as silly as ignoring the vast complexity of the organization of the human body, and speaking as if explaining the origin of human bodies is just a task of explaining how there might occur "some carbon concentrations." 

The person attempting so pathetic a trick is acting as silly as a person who stands at the seashore, fills a glass with seawater, and says, "Oceans are easy to explain -- they're just water."  Just as the ocean includes trillions of deep, baffling complexities such as all of the organization and biochemistry of sea creatures -- something infinitely more complex than mere water -- the human mind and human mental experiences involve trillions of complexities, and such a reality is something almost infinitely more complex than mere "consciousness."  The reductionist who engages in consciousness shadow-speaking is someone engaging in a trick as misleading as someone who says, "Mathematics is real simple -- it's just counting." 

The majority of people who try to reduce the mountain-sized problem of explaining human minds and human mental experiences in all their variety into the mouse-sized problem of explaining some mere dry abstraction of "consciousness"  are people who were too lazy to very deeply study minds and brains, and who used this stupid trick of consciousness shadow-speaking to try to make their explanation job a million times easier.  People who lack credible explanations for very complex realities (whether physical or mental) love to use poorly descriptive language in which they try to make the complex realities sound a million times simpler than they are. 

The dialog below illustrates the stupidity of trying to explain human minds by describing a human mind as mere "consciousness" and then trying to create a "theory of consciousness" that applies to everything conscious. 

James: John, I've made great progress in explaining how the human body arises during a mother's pregnancy. John: Great, tell me about it. James: I call my explanation a “theory of solidity.” John: A theory of solidity? James: Yes, because that's the essential nature of human bodies, that they are solid. So my theory attempts to explain how solidity arises. John: I think you've gone in the wrong direction, and made a big mistake. James: Why? John: Because a human body is something gigantically greater than mere “solidity.” A human body is a state of vast hierarchical organization, with a oceanic level of functional complexity. For example, in our bodies are 20,000 different types of protein inventions, most very special arrangements of many thousands of atoms. And we have 200 types of cells, each so complex they are compared to factories. You would do nothing to explain so impressive a reality of physical organization by merely explaining “solidity.” Your body is something gigantically more than mere “solidity.”

James: John, I've made great progress in explaining how the human mind arises. John: Great, tell me about it. James: I call my explanation a “theory of consciousness.” John: A theory of consciousness? James: Yes, because that's the essential nature of human minds, that they are conscious. So my theory attempts to explain how consciousness arises. John: I think you've gone in the wrong direction, and made a big mistake. James: Why? John: Because a human mind is something gigantically greater than mere “consciousness.” You and I are not merely “some consciousness.” We are thinking, believing, seeing, reading, hearing, loving imagining minds with insight, emotions, viewpoints, and a great variety of mental powers such as instant learning ability, the ability to hold memories for decades, and the ability to instantly recall knowledge when only hearing a word or seeing a face. Human minds and human mental experiences are a reality of oceanic depth, so much more than mere “consciousness.”

A recent study attempted to do a "showdown" between two different theories called "theories of consciousness," in an attempt to reveal a winner and a loser. The experimental study should be regarded with the greatest of suspicion, because of all the suspicions we should have about any thing at all calling itself a "theory of consciousness." An experimental showdown between two different theories calling themselves a "theory of consciousness" is rather like trying to do an experimental showdown between the theory of palm-reading and the theory of astrology. The paper is entitled "Adversarial testing of global neuronal workspace and integrated information theories of consciousness."

The paper makes quite a few dubious claims about predictions made by one or the other of these theories.  We should treat with suspicion claims made about what is predicted by either of these theories (the global workspace theory and the integrated information theory). Scientists often make unwarranted claims that this or that theory predicts something. Often such claims are made to try to achieve some aura of predictive success for some theory. It works like this:

(1) A scientist claims that some theory he favors predicts the observation of X. 

(2) The scientist then tries to show that X was observed. 

(3) The scientist then says we should have confidence in the theory because it made a successful prediction. 

Very often this is misleading in one way or another. The claim that the theory predicted the observation of X may be untrue. The claim that X was observed may be untrue. And just because some theory predicts something does not mean the theory is true or likely to be true. There are all kinds of false theories that may predict 1001 things, and some of those things may be true. 

We read in the paper claims about predictions of two rival theories of consciousness:

"We tested three preregistered, peer-reviewed predictions of IIT and GNWT for how the brain enables conscious experience (Fig. 1a). Prediction 1 addresses the cortical areas holding information about different aspects of conscious content. IIT predicts that conscious content is maximal in posterior brain areas, whereas GNWT predicts a necessary role for PFC. Prediction 2 pertains to the maintenance of conscious percepts over time. IIT predicts that conscious content is actively maintained by neural activity in the posterior ‘hot zone’ throughout the duration of a conscious experience, whereas GNWT predicts ignition events in PFC at stimulus onset and offset, updating the global workspace, with activity-silent information maintenance in between. Prediction 3 examines interareal connectivity during conscious perception. IIT predicts sustained short-range connectivity within the posterior cortex, linking low-level sensory (V1/V2) with high-level category-selective areas (for example, fusiform face area and lateral occipital cortex), whereas GNWT predicts long-range connectivity between high-level category-selective areas and PFC."

We should treat with skepticism all of these claims that such statements are actually predictions of such theories, and we should note that none of the claimed "predictions" qualify as precise predictions or exact numerical predictions. The claimed "predictions" are wooly kind of statements that are vague enough to be claimed as true no matter what is observed. Also the claimed "predictions" are not clearly at odds with each other, meaning you don't actually have a situation which is suitable for doing observations and announcing that one of the theories is the winner and the other the loser. 

To perform this dubious "showdown" of these two theories of consciousness, a large number of subjects had their brains scanned in fMRI machines, and another group had their eyes scanned while their brain waves were read using invasive brain-implanted electrodes.  Different images were shown to these observers, with each sight appearing for only about a second. We read this:

"To test critical predictions of the theories, five experimental manipulations were included in the experimental design: (1) four stimulus categories (faces, objects, letters and false fonts), (2) 20 stimulus identities (20 different exemplars per stimulus category), (3) three stimulus orientations (front, left and right view), (4) three stimulus durations (0.5 s, 1.0 s and 1.5 s), and (5) task relevance (relevant targets, relevant non-targets and irrelevant)."

So apparently subjects were shown pictures for a tiny instant, ranging from between half a second and 1.5 seconds. The pictures might have been a picture of a face, an object, a letter such as A or B, or a "false font."  The authors claim to have done "decoding of conscious content" from analyzing data obtained from these subjects: fMRI brain scan data, EEG brain wave data, eye movement data using an eye movement tracker, and magnetoencephalography  brain scan data. The claim is misleading. No robust evidence of any "decoding of conscious content" has occurred. 

To try to back up this claim of "decoding of conscious content," we have a Figure 2 that shows us some result obtained by an AI-type pattern recognizer after analyzing both EEG brain wave data and eye movement data gathered using an eye tracker device (the Eyelink 1000 Plus system shown in a photo below).  We see a "decoding accuracy" graph in which accuracy above 50% completely dies off after 1 second of someone seeing the visual stimulus. This is for 29 subjects who had intracranial electrodes inserted into their brains. The analytics are black-box analytics, and it is hard to unravel what flaws or tricks may have gone on to get these results. A look at the programming code used shows a byzantine maze of spaghetti code. No evidence is provided of being able to predict from brain data alone what a person is thinking or imagining. All we have is some attempt to show that by analyzing brain wave data and eye movement data taken at the instant someone was seeing something, you can predict the category of what the person was seeing. 

It is well known that EEG readings are extremely sensitive to muscle movements, which cause blips in the lines picked up electrodes. So imagine some experiment in which you get EEG readings while someone is shown a picture that may be a recognizable face, a picture of a cute or scary animal, or something neutral like the letter "X" or "Y."  A person might more often make muscle movements when seeing certain types of images. He might give a smile of recognition or appreciation when seeing a celebrity's face or a kitten, or he might squint when seeing some puzzling image, or he might raise his eyebrow when seeing some scary image; or he might grimace when seeing an offensive image. No such muscle movements might occur when the person sees something like the letter X or the letter Y. From such muscle movements alone, an AI pattern classifier might be able to guess higher than 50% what the category was of the thing the person saw. But that would not be "decoding of conscious content." 

Then there's the fact that the eye movement data gathered by some hi-tech eye movement device could have tended to pick up eye movement differences when different categories of images were shown.  Show a human a picture of face, and his eye will tend to focus on the face, or his eye may widen if he is surprised. Show a human a picture of a mere character (such as X) or some meaningless symbol, and the person's eye will not tend to focus as strongly, and it will not widen. 

So the evidence presented for "decoding of conscious content" in this paper is not robust evidence of being able to detect the type of thing someone is thinking about or seeing by analyzing brain data. The evidence the paper presents is all data based on "the moment of perception," when different type of muscle movements or eye movements may have occurred when different type of things are seen. 

Figure 3 in the paper is very similar to Figure 2. We are shown a line graph which seems to indicate some above-average predictive success coming from analyzing iEEG brain wave data (and also eye movement data) coming from 31 patients with implanted electrodes. The claimed success is purely in predicting the category of a type of image someone saw.  But the predictive success only occurs at the half-second mark, vanishing at the one second mark.  The result is consistent with the idea that the claimed predictive success comes purely from picking up different types of muscle movements (such as eye movements) which occur differently when a person has different types  of facial expressions in reacting to things he sees. 

The paper is one of many neuroscience papers which makes false claims about neural representations. There is no evidence that the brain contains any representations of anything anyone learns, recalls or sees. But neuroscientists love to claim that this or that thing they see in the brain is a "representation" of something. In this case the authors again and again refer to "representations" in the brain, without producing any good evidence for any such thing. 

An example of the paper's misrepresentations about representations is its statement "In posterior cortex ROIs, cross-temporal RSA revealed sustained face–object categorical representation." The evidence given for this claim is Figure 3D, which shows no sign of anything beyond the 1.5 second mark after someone saw something.  Whatever is being graphed is some momentary response to a stimulus, and it is  misleading to refer to that as either "sustained" or a "representation."  A similar misstatement would occur if I showed you a picture of something disgusting, and then claimed that your momentary facial expression was a representation of what I showed you. Momentary responses are not representations. 

The authors of this study have failed to produce any robust evidence for either the global workspace theory or the integrated information theory, and the authors give a kind of "it's a draw" verdict about their results, without saying that either theory was the winner. The global workspace theory is not a credible theory of consciousness, for reasons discussed here. The integrated information theory theory is not a credible theory of consciousness, for reasons discussed here and here. 

We have in this study a classic example of how neuroscientists resort to "something else" kind of cheats. Here's how it works:

(1) A neuroscientist will produce a study claiming to have determined something or predicted something based on brain data. 

(2) Sneaked into the study design will be some other source of data other than brain data.  That "something else" may be some software facility that the study is using, such as a database that has text annotations corresponding to images subjects were shown. Or the "something else" may be an eye-tracking system, which allows the study to make predictions not merely on brain data, but on how a person's eyes are behaving. Or the "something else" may be any number of other things, such as some AI system that predicts words someone is about to state, based on historical tendencies of people to say one word after saying a previous word. 

(3) Misleadingly it will be claimed or insinuated (in either the paper itself or the paper's press release) that the study predicted successfully based only on brain data, when any predictive success was crucially dependent on something other than just brain data. 

In this study the sleazy "something else" was eye movement data gathered by some high-tech eye tracker in addition to the EEG data  being taken to detect brain waves. The paper tells us that the Eye Link 1000 Plus system was used. Below is how that system looks (from a page promoting that system). 

Figure 4 of the paper shows the same defects as Figure 3 and Figure 2, as no predictive success beyond the 1.5 second mark is shown, and there is the same reliance on a combination of EEG brain wave data and eye-scanning data, which cannot be called a prediction from brain states alone. Figure 4 is even less reliable as evidence that Figure 2 and Figure 3, because the sample size used is much less than 31. 

In the Supplementary Notes, we read about this funny business going on:

"In the preregistration document, it is stated that iEEG patients with poor behavioral performance, defined as <70% hits or >30% FAs, were to be excluded (Data quality checks and exclusion of subjects, page 15). This threshold was considered based on a target recruitment of 50 patients. However, due to the coronavirus pandemic and despite our best efforts, only 34 patients were collected at the time of manuscript completion. To weigh the pros and cons of data inclusion and to increase sample size and coverage to better test the theories, it was decided to include in the analysis three iEEG patients whose behavior fell marginally short of the predefined behavioral criteria (i.e., hits < 70%, FA > 30%) to compensate for the lower number of participants."

So the authors set a standard for subjects that would be included, and found that they did not have enough subjects if that standard were to be followed. So the standard was then lowered.  But even with that  bit of malfeasance, was the study group size adequate for a good statistical power? We don't know, because no sample size calculation was done. 

The results in this study were highly dependent upon the patients with implanted electrodes, referred to in the paper as iEEG subjects. These were very sick patients with treatment-resistant epilepsy, who were being evaluated for surgery, through a method in which electrodes were implanted to try and find suitable spots for surgery. We are told, "A total of 4,057 electrodes (892 grids, 346 strips and 2,819 depths) were implanted across 32 patients with drug-resistant focal epilepsy undergoing clinically motivated invasive monitoring."  So each of these patients had an average of about 126 electrodes implanted in their brains. Most of the times people have electrodes implanted for epilepsy surgery evaluation, it is a much smaller number of electrodes such as only 20.  

A key question is: were all these electrode implantations medically necessary? Or was there only a medical need to implant a much smaller number of electrodes?  Were many of the risky electrode implants into the brains of these sick patients done purely for the sake of this poorly designed study? We do not know the answer to these questions, because the authors have not told us. They have not made any claim that all of the electrode implants were medically necessary. 

The abuse of very sick epilepsy patients is one of the most appalling scandals of modern experimental neuroscience. Neuroscientists hungry for brain data are luring very sick epilepsy patients into agreeing to implants inside their brains of more electrodes than are needed for surgical evaluation. When this happens, the patient undergoes very serious risks that are not medically necessary, for the sake of the research needs of the neuroscientist and not the needs of the patient. A paper tells us this:

"A recent meta-analysis reviewed complication rates and types of complications in patients undergoing subdural grid implantation for seizure mapping [41]. The most common complication which was reported was intracranial haemorrhage with a mean rate of 4% closely followed by other complications such as neurologic infections, superficial infections and elevated intracranial pressure. They also found that an increased number of electrodes (>67 electrodes) was independently associated with complications."

Another paper tells us this:

"There are definite medical risks associated with the use of intracranial electrodes. The complication rate of subdural electrodes has been reported to range between 6% and 26%.19–23) Relatively common adverse events associated with subdural electrodes are fever, headache, and nausea.24) Another group reported transient cerebrospinal fluid (CSF) leakage (13–31%), infection (6–8%), intracranial bleeding (8%), and cerebral edema in addition to an intracranial mass effect.25) Nair et al. reported that complications included (in the order of their frequency) infection, transient neurological deficit, epidural hematoma, increased intracranial pressure, and infarction.2) An increase in the complication rate was associated with (a) a greater number of grids/electrodes, (b) longer duration of monitoring, (c) older age of the patient, (d) left-sided grid insertion, (e) the use of burr holes in addition to craniotomy, and (f) an earlier year of monitoring (most likely a reflection of the aforementioned surgeon’s experience)."

The authors of any paper that reports on readings of electrodes implanted in the brains of epilepsy patients  have a duty to fully inform us about whether epilepsy patients were endangered by the implantation of additional electrodes that were not medically necessary, and which were implanted mainly for the research purposes of the paper authors.  Any such paper authors that fail to do that are authors we should tend to distrust. 

Postscript: A paper on pre-registration in psychology research finds some dismal results. We read this:

"In this paper, we have investigated adherence and  disclosure of deviations for all articles published with the Preregistered badge in Psychological Science between February 2015 and November 2017 and shared our findings with the corresponding authors for feedback. Two out of 27 preregistered studies contained no deviations from the

preregistration plan. In one study, all deviations were disclosed.

Nine studies disclosed none of the deviations."

What is supposed to go with a pre-registered paper is that scientists publish an exact plan for how they will gather and analyze data, before gathering data, and that they adhere to such a plan. The paper suggests that almost all pre-registered studies actually deviate from their pre-registered plan, and that a large fraction (maybe one third) of studies that deviate from their pre-registered plan fail to report that they deviated from their plan.