Look Very Closely at a Study Claiming Thought Reading of Brains, and You'll Find the Shady Tricks Behind It

Certain web sites tend to fall "hook, line and sinker" for very dubious and poorly designed neuroscience papers (something that are extremely abundant these days). A recent example is a Gizmodo.com story with the very untrue headline "Brain-Reading Device Deciphers Internal Thoughts With Surprising Precision." We have an article hyping a new science paper with the groundless title "Representation of internal speech by single neurons in human supramarginal gyrus."

The EEG is a device attached to a head, a device that reads brain waves. EEG devices that measure brain waves are strongly influenced by muscle movements, and speech is a type of muscle movement. It does nothing to show that brains produce thinking if you have some EEG device that can detect different responses when different things of different length are spoken. That can be mainly picking up muscle movements, not thought. EEG devices are very sensitive to muscle movements, and if you have a medical EEG reading you will be told not to move during the reading.  

To try to back up unfounded boasts of being able to detect thoughts from brain activity, neuroscientists sometimes claim that they are picking up signals during "internal speech." What is the difference between "internal speech" and actual speech? Actual speech is when you say a word, and "internal speech" is when you only think of the word.  Unless a study is carefully designed, what will happen is that faint amounts of actual speech may occur when subjects were supposed to be using only internal speech. No such careful design has occurred in this study. Instead, we wonder whether the authors had chosen a design trying to get lots of traces of actual speech that could be called internal speech. 

Here is how a carefully designed study might be done, in a way that would minimize the chance of subjects producing a slight amount of actual speech (such as silently mouthing a word) when they were supposed to be only producing internal speech:

(1) A subject might first see a screen saying: "Please speak aloud this word: automobile."

(2) After the response was recorded, a device would then be put in the subject's mouth, to prevent any muscle movement in his mouth. 

(3) The next screen might say: "Now we want you to do something very different: to only think a word we give you, without moving your muscles at all. Be very careful not to move any muscle."

(4) Then the next screen might say: "Now please ONLY THINK this word: locomotive."

Nothing like this was done. We read this description of how the subjects or subject was instructed to act:

"During the cue phase, a speaker emitted the sound of one of the eight words (for example, python). Word duration varied between 842 and 1,130 ms. Then, after a delay period (grey circle on screen; 0.5 s), the participant was instructed to internally say the cued word (orange circle on screen; 1.5 s). After a second delay (grey circle on screen; 0.5 s), the participant vocalized the word (green circle on screen, 1.5 s)."

Get the picture? It was a super-hurried affair.  The "ms" in the quote above refers to milliseconds or thousands of a second, and the "s" in the instructions above stand for a second. Within a five-second period, the people tested were supposed to first speak a word, then only think the word, and then again speak the word. What can we expect would happen given so very hurried a situation, in which people were supposed to switch so quickly between speaking a word and only thinking a word, with two vocalizations of a word and one thinking of the word all occurring within five seconds? There would be a high chance of failing to follow the instructions properly. Given such instructions, the reported results are no compelling evidence for any detection of internal speech.  

Here's a test I want you to try right now, one with three easy, safe steps I want you to perform real fast, in under ten seconds:

(1) First give a big smile and say the word "rabbit" loud and clear.

(2) Next, give a big smile and again say the word "rabbit" loud and clear.

(3) Next, give a big smile and think the word "rabbit."

My guess is that at least 30% of my readers trying this test spoke or mouthed the word rabbit" when they came to the third line, even though that line said to merely think that word. Whenever instructions are given very rapidly, and there's a sudden change in the instructions, there's a large chance the instructions will not be followed exactly. 

So we can assume that given the instructions quoted in italics above (in which instructions are given to first speak a word, then think a word, and then speak the word again, all within five seconds), that in a large fraction of the cases some actual speech or mouth muscle movement occurred during the second part when the subjects were supposed to be only thinking the word.  So probably very much of the recorded "internal speech" was something more, including muscle movement. 

How many subjects were used? Only one subject for one of the tests, and two subjects for some other tests.  Were many trials used? No, not very many. Each subject was asked to think a total of only about 100 words. 

The subjects were given one of eight words to speak and then think of: battlefield, cowboy, python, spoon, swimming, telephone, bindip and nifzig.  Using only a time factor based on the length of activity, and the time of task engagement, a computer program could could probably predict with 50% accuracy which word was used. 

The reported results (about 50% prediction accuracy for some computer program) are results that can easily be explained without assuming any pure thoughts in the brain, by assuming:

(1) a significant fraction of the words that are claimed as "internal speech" actually involved traces of actual speech such as softly spoken words or mouthed words, which occurred because of the very rushed time frame in which participants were supposed to speak a word, then think the word, then speak the word all within five seconds (with a sudden change from instructions to speak a word to instructions to think a word). 

(2) the difference in word lengths was leveraged to help predict which word was mouthed or thought of;

(3) simple chance was leveraged,  easy to do in a situation where scientists are free to file away failing results in their file drawers. 

Because of the shady tricks used, we have here no evidence at all of picking up thoughts from some region of the brain.  A properly designed study of this type would have characteristics such as these:

(1) Instead of instructions given so rapidly or such a rapid-pace timescale (as if they were designed to produce a large amount of failure to follow the instructions), the instructions would be delivered slowly with a longer timescale, to maximize the chance that when subjects were supposed to be thinking (and not speaking), they really would be only thinking. 

(2) Additional measures would be taken to prevent mouth muscle movement when the subjects were supposed to be only thinking, measures such as putting something in the subject's mouth to prevent muscle movement. 

(3) All of the test words would be of an identical length, to prevent some computer program from predicting based merely on the length of the subject activity. 

For more on the misleading tricks used by studies claiming thought reading of brains, see my three posts below:

No One Is Actually Doing Thought-Reading by Scanning Brains or Reading Brain Signals

Misleading Tricks of the Latest Claim of Mind-Reading by Brain Scans

Suspect Shenanigans When You Hear Claims of "Mind Reading" Technology

Claims of mind-reading by brain scans or brain signal reading will continue, and they will typically be based on tricks such as these:

(1) Picking up signs of muscle movement, always possible for anyone who can talk, and wrongly passing that off as detection of "inner thought," aided by study designs which maximize the chance of "contrary to the instructions" muscle movement (or allow a possibility of some kind of muscle movement) when only "inner thought" is supposed to be occurring.

(2) Having tests with words of very different lengths, with a claimed detection of which word was thought of, the success being mostly based on longer time intervals corresponding to such longer words, not any real detection of what someone was thinking. 

(3) Various sneaky tricks in which data backdoors are leveraged, so that predictive success (claimed to be from reading brain data) is mostly coming from data grabbed from a source other than brain data (such as the study that showed people images, and that leveraged previously gathered descriptive data that existed for each of the images presented).

No One Is Actually Doing Thought-Reading by Scanning Brains or Reading Brain Signals

Elon Musk's company Neuralink has recently announced that it has implanted a chip in a human brain. The press has many an untrue headline about this, such as headlines talking about mind-reading brain chips. No one has done any such thing as reading thoughts by scanning brains. 

You might expect to get the real scoop on this topic by reading an article on it in the prestigious journal Nature. But a recent article in that journal misinforms us on this topic. The article is entitled  "Elon Musk’s Neuralink brain chip: what scientists think of first human trial." We read this:

"Neuroscientists have long argued that data from individual neurons are needed for sophisticated thought-decoding. But research indicates that averaged signals can allow decoding of complex cognitive processes, such as inner speech."

No, that isn't true. It is impossible to figure out what a person is thinking by scanning his brain signals or brain states.  The quote above has a link to the paper "A high-performance neuroprosthesis for speech decoding and avatar control." The paper describes a device that helps decode what a paralyzed woman is trying to say. But that paper states this:

"For speech decoding, the participant was presented with a sentence as a text prompt on a screen and was instructed to silently attempt to say the sentence after a visual go cue. Specifically, she attempted to silently speak the sentence without vocalizing any sounds. This differs from imagined or inner speech because she was trying to engage her articulators to the best of her ability, although substantial orofacial weakness prevents her from naturally mouthing words."

The paper describes a woman paralyzed so badly that she is unable to make intelligible speech. What the device was doing was responding to muscle signals that occur when a person is trying to speak. Picking up such signals is not an example of reading thoughts from brain signals. As the quote above confesses, it is not an example of "inner speech."

The Nature article has misled us by claiming that some decoding of brain signals can "allow decoding of complex cognitive processes, such as inner speech." The first paper it cites in support of this claim denies that claim, as the quote above shows. 

Besides having a link to the paper above, the Nature quote above also had a reference to the paper "Generalizable spelling using a speech neuroprosthesis in an individual with severe limb and vocal paralysis." That paper is also a paper that relies on attempted speech in a severely paralyzed person, rather than pure thought that makes no attempt at muscle movement. We read this in the paper: 

"Here, we demonstrate that real-time decoding of silent attempts to say 26 alphabetic code words from the NATO phonetic alphabet can enable highly accurate and rapid spelling in a clinical-trial participant (ClinicalTrials.gov; NCT03698149) with paralysis and anarthria. During training sessions, we cued the participant to attempt to produce individual code words and a hand-motor movement, and we used the simultaneously recorded cortical activity from an implanted 128-channel electrocorticography (ECoG) array to train classification and detection models. After training, the participant performed spelling tasks in which he spelled out sentences in real time with a 1152-word vocabulary using attempts to silently say the corresponding alphabetic code words. A beam-search algorithm used predicted code-word probabilities from a classification model to find the most likely sentence given the neural activity while automatically inserting spaces between decoded words. To initiate spelling, the participant silently attempted to speak, and a speech-detection model identified this start signal directly from ECoG activity.".

The second paper cited by the Nature article also gives no support for the article's claim that some decoding of brain signals can "allow decoding of complex cognitive processes, such as inner speech." To the contrary, what we have is another case of picking up attempted muscle movements. 

The ideologically neutral diagram below may help you understand the situation better:

What the two papers above are doing are picking up brain signals from the motor cortex. Such papers do nothing to show that the Mind Core (thought, recall, will, imagination, self-hood, etc.) comes from the brain. Virtually no one disputes that there can be increased activity in the motor cortex area of the brain when someone is engaging in muscle activity or attempting to move his muscles. 

Real thought-reading using neural devices would require picking up something when no one was attempting to move their muscles. Such a thing has never been done. There have been some misleading "mashup"-technique studies that have included attempts to read brain signals and also the use of computer databases and AI (artificial intelligence). Such a mixture can be an extremely confusing "smoke and mirrors" affair in which results coming not from thought-reading by brain scans are mixed up with some brain scanning activity, with the results misleadingly passed off as some breakthrough in "neural mind reading." For a discussion of some of the resulting misleading studies see my two posts below:  

Misleading Tricks of the Latest Claim of Mind-Reading by Brain Scans

Suspect Shenanigans When You Hear Claims of "Mind Reading" Technology

Elon Musk's company Neuralink may produce something of benefit to paralyzed people, but brain implants will never produce any great benefit to the general public. The brain is not the source of your thoughts or your imagination, and is not the storage place of your memories. The idea that people will be able to enhance their intelligence or learn more quickly by brain implants is an idea based on the incorrect idea that your brain is the source of your mind, and the incorrect idea that the brain is the storage place of human memories. To discover the reasons why those ideas cannot be correct, read the posts on this blog. 

I read that Neuralink is now claiming that a human with a Neuralink chip implant is controlling a mouse "by thoughts alone." To the contrary, all that is occurring is such a chip picking up intended muscle movements, which are not thoughts. Previously we got from Neuralink or one of its chief investors the inaccurate claim that no monkey had died from one of Neuralink's chip implants. A story in Wired showed that the claim was not correct, and details the specifics of how monkeys had died from Neuralink implants. In November the Reuters news agency reported that "Four U.S. lawmakers have asked the Securities and Exchange Commission to investigate whether Elon Musk committed securities fraud by allegedly misleading investors about the safety of a brain implant being developed by the billionaire’s firm Neuralink, according to a letter to the regulator."

In the history of neuroscience there have been endless misstatements. They have included the following:

• The very massive current occurrence of misleading university press releases, very often announcing new research and making claims about such research not matching anything shown by the research (with the press release claims very often not matching anything even claimed in the corresponding scientific paper). 
• The extensive use of deceptive brain scan visuals, which "lie with colors" by using misleading coloring effects in which very tiny brain activity differences are depicted in bright colors, leading people to think that particular regions of the brain "light up" and are much more active during certain cognitive activities, when in reality the difference is only about 1 part in 200 (the type of difference we might expect from random fluctuations, even if brains do not produce minds and do not store memories). 
• The massive occurrence in scientific papers of inaccurate citations, claiming that some paper showed or supported some claim that it never showed or supported, typically made by scientists who never read the paper they are citing (a scientific paper estimated that only 20% of people citing a scientific paper actually read the paper they are citing). 
• Inaccurate descriptions of what was stated by people who were brain zapped to try to produce an out-of-body experience, in which the subjects hesitating and ambiguous responses (often in response to "leading" questions) are described as reports of an out-of-body experience, when such a report was not given by the subject. 
• Deceptive papers in which purely software implementations are passed off as things that help to explain human memory, by means of outrageous language abuses in which sections or layers of software code are improperly given anatomical names corresponding to parts of the human brain, and in which tricky equivocation occurs involving mixing up the human definition of memory (involving mental experiences) and the computer definition of memory (not involving mental experiences). 
• A very large number of misstatements and misrepresentations by psychiatrists, very carefully documented in Peter Gøtzsche’s "Critical Psychiatry Textbook" that can be read here, with the misstatements often occurring to try to bolster weakly supported or unfounded claims that various types of mental illness are caused by brain states rather than life histories or socioeconomic conditions or personal life conditions. 
• The massive current occurrence of both misleading titles in scientific papers and misleading claims in scientific paper abstracts, with paper titles very commonly making claims not matching anything established by the research in the main body of the paper, and abstracts also frequently claiming the research showed something it failed to show. 
• Extremely misleading statements about the quality of evidence for spiritual and psychical phenomena that tend to contradict neuroscientist dogmas, typically made by people who have never seriously studied such evidence, often claiming very large bodies of solid evidence gathered over decades or centuries are "no evidence," combined with misleading stereotypical, mudslinging or gaslighting characterizations of the people who have reported such phenomena.
• Many decades of poor research practices in neuroscience, such as the use of way-too-small sample sizes of fewer than ten subjects, failure to follow blinding protocols, lack of pre-registration, and the use of an unreliable "see-whatever-you-want-to-see" method for trying to measure animal recollection ("freezing behavior" estimation) rather than the reliable measurement techniques discussed here.  
• The repetition by neuroscientists of utterly absurd claims that it takes hours for a human to form a long-lasting memory, claims that are contrary to every person's experience, which is that permanent memories can form instantly, with the claims being made because the people making such claims want us to believe that memories are formed through synapse strengthening known to take at least hours.  
• The frequent appearance of highly speculative "brain functional map" charts suggesting some knowledge that particular parts of the brain produce cognitive functions, suggestions that are unwarranted (see here for evidence against one of the standard elements of such charts). 
• The extremely severe lie by materialists that a split-brain operation (severing the nerve fibers connecting the two hemispheres of the brain) results in two separate human minds, contrary to the facts that no such thing occurs, and that such an operation leaves people with a single self.   
• The use in neuroscience experiments of tortuous extremely convoluted analysis pathways in which brain scan data or biological data is passed through a series of programming iterations, often involving poorly documented gobbledygook code that no one but the original programmer could have ever understood, with the effect of the rigmarole iterations being some "black box" manipulation that not even the original programmer can now understand, and the resulting mess (perhaps having some desired pattern) being passed off as some enhancement of the original data, even though there is every season to suspect the result is a corruption, distortion or contortion of the original data. 

In general, you should be suspicious about some neuroscientist trumpeting his own research, because that person is a vested interest who stands to gain financially and career-wise the more you believe his or her claims. And when corporations issue  neuroscience-related claims that are controlled by the desire of such corporations to maintain high stock prices and the desire to attract more investors, then perhaps you should be suspicious to the second power. 

The erring Nature article I quoted above is not the only erroneous article Nature has published making the false claim that minds are being read by brain devices.  A recent Nature article has the misleading title "Mind-reading devices are revealing the brain’s secrets," a claim it provides no evidence to support.  The article starts out like this:

"Moving a prosthetic arm. Controlling a speaking avatar. Typing at speed. These are all things that people with paralysis have learnt to do using brain–computer interfaces (BCIs) — implanted devices that are powered by thought alone."

No, none of these things is "powered by thought alone." They are all devices reading brain signals to detect muscle movement attempts rather than pure thoughts. 

We hear that the Broca's region supposedly involved in speech contains no actual information about words:

"They also found that Broca’s area, a brain region thought to have a role in speech production and articulation, contained little to no information about words, facial movements or units of sound called phonemes. 'It seems surprising that it’s not really involved in speech production per se,' says Willett."

The article makes this claim, which is not backed up any good research:

"His team developed a BCI implant capable of generating speech in real time when participants either whisper or imagine speaking without moving their lips or making a sound. The brain signals picked up by the BCI device in both whispered and imagined speech were similar to those for spoken speech. They share areas and patterns of activity, but are not the same, explains Herff."

In the quote above we have a reference to the paper here which has the misleading title "Real-time synthesis of imagined speech processes from minimally invasive recordings of neural activity." We have a study involving a single subject who was not paralyzed. The study has got brain wave readings of the subject while she was told to speak, whisper, and just imagine. From the combination of all three readings, some better-than-chance prediction is made about the words spoken or whispered or imagined.  But since data is being gathered in all three ways, no claim can be made to have read pure thoughts. And we don't know whether the subject kind of halfway-mouthed some of the words during the cases of supposedly "just imagining."  The authors refer to "the inherent difficulty of knowing how the participant internally performed imagined speaking,"  Exactly: we don't know whether the claimed "imagined speaking" really was pure thought.  

The Nature article's claim that the "BCI implant" was "capable of generating speech in real time when  participants either whisper or imagine speaking without moving their lips" is not correct. The study merely guessed about the words its single subject had spoken, whispered or imagined, and made guesses with a success rate slightly above chance,  a result we might expect to get by chance given 100 experiments trying such a thing.  You can plausibly explain the reported results easily without supposing any actual reading of thoughts, by simply imagining that the single subject faintly vocalized a few of the small number of words she was asked to imagine.  

Since the experiment was a rapid-fire affair, with very quick responses required, an experiment requiring responses to 100 random words in a period of only five minutes, it would have been very plausible that a subject might have done such a thing, despite being told to only imagine the displayed word. Reliable neuroscience experiments involving correlation effects require at least 15 subjects, and an experiment based on data from a single subject should not be taken very seriously. The authors claim that their one-subject experiment "demonstrates real-time speech synthesis from imagined neural activity" is unfounded.

What kind of design would a study like this need to have to provide credible evidence of thought-reading by a neural device? For one thing, you would need to test at least 15 subjects. Also, steps could be taken to prevent subjects from doing something like attempts at muscle movements in tests in which pure thought was required. You could have the subjects wear a mouthpiece that would prevent muscle movement in the mouth. Or you could have subjects wear in their mouth a device that would detect faint muscle movements or attempted muscle movements when a subject was supposed to be engaging in pure thought. Whenever such a detection was made, that response would be removed from the data, with a warning flashed not to whisper the displayed word. You would also make sure to have a decent time interval between each word test, such as 30 seconds. You might have the display of each word preceded by a warning such as "Now make sure not to speak the next word -- just THINK it."

Don't be fooled by claims of mind-reading technology in the science news. Today's "Science News" pages and sites are a boastful cesspool of glitzy goofy semi-true clickbait-infected razzle-dazzle razzmatazz that largely serve to promote the vested interests of corporations, elitist institutions, click miners and glory  seekers. 

Postscript: In an article we learn about the person who has received the first Neuralink chip. He is a person paralyzed from the shoulders down, who can speak and move his head normally (as we can see from the video here).  We read that he could play Civilization VI (a PC strategy game) for hours before having the brain chip implanted. There are a variety of ways in which technology could pick up intended game movements from such a patient.  Without even having a chip inside the brain, you could have technology that could translate head muscle movements into the equivalent of game controller movements.  We should not be impressed by the ability of such a patient to play a video game after having a brain chip implant. What is probably going on is that some system is receiving a variety of inputs, only some of which are signals from the motor cortex.  Calling such a thing a case of "reading thoughts from the brain" is very misleading. 

What would be impressive would be if someone were to have a brain chip implant, and were then to silently imagine some novel scene, such as a blue sphere sitting on top of a green cube, resting on a red table, with a computer then creating exactly such a scene.  You'll never see that, because brains are not the source of our thinking and imagination.  

The latest untrue news on this topic comes in the form of an untrue headline at www.thedebrief.org, a headline of "ROUNDBREAKING NON-INVASIVE UNIVERSAL BRAIN-COMPUTER INTERFACE LETS PEOPLE PLAY VIDEO GAMES USING ONLY THEIR MIND." The story recycles an untrue press release issued by the University of Texas, one claiming "Universal Brain-Computer Interface Lets People Play Games With Just Their Thoughts."  A look at the scientific paper  tells us that "Subjects were instructed to mentally rehearse the kinesthetics, not the visualization, of a movement without overtly causing any contraction of their muscles."  Given such instructions, people will make small muscle movements corresponding to how they want to move in a video game, and it is those small muscle movements that are being mined, rather than "just thoughts."  The subjects were wearing EEG caps, and it is well known that EEG signals are very sensitive to muscle movements.  Lacking controls and laden with statistical fog, the paper does not even give clear intelligible evidence the subjects performed above chance.  

Post-postscript: The paper "Long-term independent use of an intracortical brain-computer interface for speech and cursor control" is one that was promoted by news stories claiming progress in so-called "brain-computer interfaces." A look at the paper reveals that what is going on is a lot more than just analysis of brain waves.  The paper confesses, "The system independently using neural decoding, supplemented by eye gaze tracking to facilitate user-interface control." The next sentence confesses, "The speech decoder software ran continuously in the background and displayed decoded words on-screen in real time (Fig. 1C, left)." This involved specialized software for decoding garbled speech, something much different from brain wave analysis. There was also hand movement analysis, because we read this of the disabled person using the system: "He could also control his computer by attempting to move or squeeze his hand." So we have here at least three different ways in which the system was getting data from the user other than picking up his brain waves. It is misleading to use the term "brain-computer interfaces" for systems like this that involve so much more than just inputs from brains. 

Carefully designed devices can pick up a wealth of information from scanning human eyes. For example, given the right type of computer screen viewed by a paralyzed person or a person not using his hands, a carefully designed device can pick up direction indicators from eye movements alone.  Imagine a computer screen showing a scene like the one below:

Through the right technical setup (such as something with a camera looking from this screen to the viewer), you can get a "drive without touching anything" effect displaying on the screen, in which a viewer's gaze at position A causes the screen car to turn to the left, a viewer's gaze at position B causes the screen car to drive straight ahead, and a gaze at position C causes the screen car to turn to the right. If such a system is combined with some EEG apparatus reading brain waves, someone can make a claim that "thoughts in the brain are being read to move the car in 3D space," even though the brain waves are superfluous, and the system is really working based on eye movements. 

There are endless similar variations that might leverage eye movements or eye focus. For example, a screen might show three rows of choices: Choice 1, Choice 2 and Choice 3.  A system might be able to pick up a person focusing his eyes on any of these choices, with a corresponding action occurring. If combined with some EEG reading of brain waves, a neuroscientist might pass this system off as something that is "reading your brain's decisions," even though all that is happening is that eye focus on a particular screen position is being picked up, with the reading of brain waves being superfluous. 

Misleading Tricks of the Latest Claim of Mind-Reading by Brain Scans

In a previous post entitled "Suspect Shenanigans When You Hear Claims of 'Mind Reading' Technology" I discussed some of the tricks used by people claiming that brain scans can reveal mind activity. I discussed one example. It was the case of a researcher who had used some incredibly elaborate analysis pipeline that included brain scans to create movies. I pointed out that brain scans were only one element in the extremely elaborate set of inputs, and that it was misleading to claim that the output movies were generated from brain scans. I stated this:

"This bizarre and very complicated rigmarole is some very elaborate scheme in which brain activity is only one of the inputs, and the main inputs are lots of footage from Youtube videos.  It is very misleading to identify the videos as 'clips reconstructed from brain activity,' as the clips are mainly constructed from data other than brain activity. No actual evidence has been produced that someone detected anything like 'movies in the brain.' It seems like merely smoke and mirrors under which some output from a variety of sources (produced by a ridiculously complicated process) is being passed off as something like 'movies in the brain.' "

Recently we had another case of the press fooling us with untrue claims about mind-reading brain scans.  The Daily Mail gave us this bogus headline: "Scientists can now read your MIND: AI turns people's thoughts into images with 80% accuracy." Vice.com gave us this equally untrue headline: "Researchers Use AI to Generate Images Based on People's Brain." Upon analyzing the scientific paper that inspired these stories, I was able to figure out how the misleading "sleight of hand" is being done. It's a "fool you" mashup methodology. 

The paper is the one you can here. It has the very misleading title "High-resolution image reconstruction with latent diffusion models from human brain activity." What is going on is that the researchers used an analysis methodology in which the actual brain scans are a superfluous input. They got AI outputs using a technique in which there was no need at all to use brain scans.  The paper title is misleading because it implies that such brain scans were a crucial input, when such brain scans were an unnecessary input. 

Below is an explanation of how it worked:

(1) There is a Natural Scenes Dataset that was created by some morally dubious excessive-seeming fMRI scanning in which subjects were brain-scanned with high-intensity 7T scanners for about 40 hours each while looking at natural scenes (a medically unnecessary risk to these subjects). That dataset is described in the paper here, entitled "A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence."

That dataset was created using images from the Microsoft Common Objects in Context (COCO) image dataset. The authors of the previously mentioned paper mentioned above say, "We obtained 73,000 color natural scenes from the richly annotated Microsoft Common Objects in Context (COCO) image dataset." These authors then brain-scanned people while they were looking at these images, using 7T fMRI scanners. 

(2) The authors of the new paper ("High-resolution image reconstruction with latent diffusion models from human brain activity") traced back the images of the Natural Scenes Dataset to their COCO source, as they admit by saying, "The images used in the NSD experiments were retrieved from MS COCO and cropped to 425 x 425 (if needed)." This COCO database includes text annotations for each image, which is one or more words identifying each image. The authors of the new paper ("High-resolution image reconstruction with latent diffusion models from human brain activity") clearly indicate that they grabbed these text annotations from the COCO database. They say they used an "average of five text annotations associated to each MS COCO image." 

(3) Having a text phrase associated with each image they used from the Natural Scenes Dataset, a phrase identifying what each image was, the authors used such text phrases as inputs to the Stable Diffusion generative AI, which can generate multiple images from text phrases.  In case you have not tried the Stable Diffusion AI, which you can try using this link,  it works like the example below. I typed in "spooky snowy castle" as the prompt, and the AI generated four images of spooky snowy castles:

(4) Some additional use was made of the actual brain scans from the Natural Scenes Dataset, but that was not necessary, and was probably just a little "icing on the cake," apparently as a way for the authors to kind of "cover their tracks" by some convoluted rigmarole making it harder for people to track down the main way their images were generated. 

(5) The authors then incorrectly claimed that they had done "image reconstruction ... from human brain activity." In fact, the human brain activity was a superfluous (unnecessary) input that was not a necessary part of the process. The method used to get the Stable Diffusion output images would have worked just fine without any brain scan data at all. All you need to get good Stable Diffusion output images of some particular type is a text prompt. And the researchers had got the appropriate text prompts by matching the images of the Natural Scenes Dataset with the source of the images (the MS COCO dataset), which has text phrases describing each of the images.  

This is a "smoke and mirrors" sleazy trick that you should not be fooled by. The authors incorrectly claimed that they had done "image reconstruction with latent diffusion models from human brain activity," when the human brain activity was not an essential input. The technique used here has no dependency on any brain scan data. The authors claim that they have shown you can "reconstruct high-resolution images with high semantic fidelity from human brain activity." This claim is untrue. Using only the brain activity data, the authors would be unable to create any images corresponding to what the subjects had seen when such brain scans were made. 

There were two cheats here: (1) using text annotations (descriptions of the images the brain scanned subjects saw), descriptive phrases which the brain-scanned subjects never heard or saw; (2) use of an image generating AI using these text phrases as inputs rather than inputs of the brain scans. The authors have not reconstructed what the subjects saw from their brain scans. The authors have used a sneaky data backdoor to get something they never could have got from such brain scans alone. A legitimate attempt to reconstruct what people saw from brain scans would have used only the brain scans, and would never have succeeded. You cannot identify or reconstruct from brain scans what people saw or thought while their brains were being scanned.  

What is going on here is something rather like the conversation below: 

Jack: Did you know I can tell which restaurant you went to from a list of the items you ordered?

Jill:  Really? Let's try.

Jack: Okay, just give me a receipt you got from some dinner you ordered.

Jill: Okay, here's my receipt from last night. 

Jack: Okay, let me see, you ordered a large pizza and 2 medium Pepsi drinks. Using my astonishing algorithm, I deduce that you went to Santino's Pizza Palace on 34th Street.

Jill: Wow, that's amazing -- you figured out where I ate from what I ordered!

Of course, Jack has done no such thing. Jack is cheating. He simply read the name of the restaurant from the bottom of the receipt. 

As long as I am mentioning the Natural Scenes Dataset, let me mention a very troubling fact about that dataset: that it was created by what seems like a recklessly excessive scanning of 8 subjects.  A paper entitled "A massive 7T fMRI dataset to bridge

3 cognitive neuroscience and artificial intelligence" discusses some data collection used to create this dataset: a process in which eight subjects were brain scanned 30 to 40 times with 7T scanners twice as powerful as the 3T scanners or 1.5T scanners normally used for MRI scans.  The paper states this: 

"The total number of 7T fMRI scan sessions were 43, 43, 35, 33, 43, 35, 43, and 33 for subj01–subj08, respectively. The average number of hours of resting-state fMRI conducted for each subject was 2.0 hours, and the average number of hours of task-based fMRI conducted for each subject was 38.5 hours."

This was in addition to other 3T scans the subjects were given.  The paper makes no mention of any consideration of health risks to these people, who received $30 per hour for the medically unnecessary scans. A 7T scanner would presumably have more than twice the risks of the 3T scanners.  Referring to mere 3T MRI scans, the 2022 paper "The effects of repeated brain MRI on chromosomal damage" found that "The total number of damaged cells increased by 3.2% (95% CI 1.5–4.8%) per MRI." The paper was referring to "DNA breaks" that have a possibility of increasing cancer risks. There is no medical need for anyone to receive more than one or a few MRI scans. Scanning subjects for 40 hours with 7T scanners seems rather like playing Russian roulette with the health of subjects, who might one day get cancer or dementia from such excessive scanning. It is dismaying that people were lured into undergoing such risks for "chump change" payments such as $30 per hour. 

A year 2024 study ("Evaluation of the Biological Effects of Exposures to Magnetic Resonance Imaging on Single-Strand DNA: An In-vivo Study") found similar results, finding that MRI scanners only half as powerful as 3T scanners can produce genotoxic effects.  It reported this:

"The DNA single-strand breaks were significant for all tested parameters in both MRI 1.5 T (p<0.01) and 3.0 T (p<0.001)....The percentage of cells destroyed in the group exposed to 3.0 T MRI was increased to 12.65 ± 1.0 after 10 minutes of exposure."

As shown in its Table 1, the study found that with a 3.0T MRI scanner, the longer you scan, the more DNA damage is done. The reported damage after 40 minutes of 3.0T MRI scanning is given a numerical value of 57, which is nearly twice as high as the damage done after only 10 minutes of scanning (damage given a numerical rating of 32). A similar relation is reported in Table 4, using a different measure of damage.  Using that measure, 10 minutes of 3.0T MRI produces DNA damage given a numerical value of 12.66, but 40  minutes of 3.0T MRI produces almost twice-as-bad DNA damage given a numerical value of 22.38.  Any DNA damage tends to increase cancer risk. 

All future claims to be generating images from brain scans should be regarded with the greatest suspicion whenever such claims made any use of the Natural Images Dataset. I have explained above how there is a tricky "backdoor" method by which anyone can generate images from that dataset very similar to the images that the poor over-scanned subjects saw when the brain scans of that database were made. 

We can expect to see in the future some additional studies using sleazy tricks such as the one described here. There will be more and more confusing methodology papers that use complicated technological mashups that leverage AI and data backdoors.  Don't be fooled by such shenanigans. It is never possible to figure out what someone thought or saw from merely looking at brain scans, and any new paper suggesting otherwise will almost certainly be using complicated trickery designed to hide its sneaky sleight-of-hand. 

Postscript: Above I stated, "You cannot identify or reconstruct from brain scans what people saw or thought while their brains were being scanned." This statement is not at all discredited by papers such as the mistitled paper "The Code for Facial Identity in the Primate Brain." That paper does not meet good standards of experimental neuroscience. The paper is not a pre-registered paper that committed itself to one exact method of analysis before data was analyzed. The paper is one of those papers in which you get the suspicion that the authors were playing around with countless types of statistical analysis before ending up with what they reported. The analysis pipeline they report is some hopelessly convoluted and arbitrary rigmarole that fails to provide any convincing evidence for any such thing as a code for representing faces in brains. The statistics involved are so convoluted a can of worms (or perhaps we should say  "vat of worms") that it smells like irreproducible results.  I may note three fundamental failures:

(1) The lack of pre-registration, leaving the authors free to "keep torturing the data until it confessed."

(2) The lack of any blinding protocol, a necessity for a paper like this to be taken seriously.

(3) The use of only two monkey subjects (in a correlation study such as this, 15 subjects would be the minimum for a slightly impressive result). 

At the NBC News web site, we have a story entitled "From brain waves, this AI can sketch what you're picturing." The title gives the incorrect idea that scientists were trying to reconstruct what people were imagining (a common deceit of stories like this), although the actual study only involved what people were seeing while their brains were scanned. The story claims, "The resulting generated image matched the attributes (color, shape, etc.) and semantic meaning of the original image roughly 84% of the time." That's not a claim made by the scientific paper, which reports accuracy of only about 21%. in its "Results on Different Subjects" section. The study used annotations in the COCO database (text descriptions of the images), so it apparently used the same kind of data backdoor trick described above. Again, we are given the misleading impression that images are being reconstructed (or the content of images guessed) based solely on brain scans, when no such thing is happening. Instead the content of what someone saw is being guessed based on brain scans and lots of other data other than just the brain scans.  

Misrepresentations of what went on in studies of this type are extremely common in the press. We may be told that such and such a study identified what people were thinking or picturing when the study merely involved people whose brains were scanned when they were seeing something or speaking words. 

The latest in misleading poor-quality science papers trying to insinuate mind-reading by brain scans is the paper "Semantic reconstruction of continuous language from non-invasive brain recordings." A few subjects were brain-scanned for 16 hours, and attempts were made to predict what they had heard, using both brain scans and "a generative neural network language model that was trained on a large dataset of natural English word sequences" in order to get some ability to predict the words that would follow from a sequence of previous words. We read, "Given any word sequence, this language model predicts the words that could come next." The meager results produced had a statistical significance of only "p < .05," which is very unimpressive. That's the kind of result you would expect to get by chance in one out of 20 tries. Again, we have a misleading formula of  getting output using "brains scans plus some other huge thing" with the small claimed success coming mostly from the other huge thing, not the brain scans. No actual evidence has been provided that you can reconstruct what people were thinking or hearing from brain scans alone. For the sake of this piece of misleading parlor-trick junk science, some subjects had their brains scanned for 16 hours, a medically unnecessary risk to them which may have increased their chance of getting cancer or dementia. 

Ever-eager to produce misleading but interesting-sounding click-bait stories that help increase page views and advertising revenue, the press has jumped on this story, producing some very misleading stories that do not accurately describe the research,  and fail to tell us that the results mainly do not come from analysis of brain scans, but from some high-tech AI trained to anticipate the most likely words that would follow from some text. 

Suspect Shenanigans When You Hear Claims of "Mind Reading" Technology

 The New Yorker recently published an extremely misleading article with a title of "The Science of Mind Reading," and with a subtitle of "Researchers are pursuing age-old questions about the nature of thoughts—and learning how to read them." The article (not written by a neuroscience scholar) provides no actual evidence that anyone is making progress trying to read thoughts from a brain. 

The article starts out with a dramatic-sounding but extremely dubious narrative. We hear of experts trying to achieve communication with a Patient 23 who was assumed to be in a "vegetative state" after a bad injury five years ago.  We read about the experts asking questions while scanning the patient's brain.  They were looking for some brain signals that could be interpreted as a "yes" answer or "no" answer.  We are told: "They would pose a question and tell him that he could signal 'yes' by imagining playing tennis, or 'no' by thinking about walking around his house." 

We get this narrative (I will put unwarranted and probably untrue statements in boldface):

"Then he asked the first question: 'Is your father’s name Alexander''

The man’s premotor cortex lit up. He was thinking about tennis—yes.

'Is your father’s name Thomas?'

Activity in the parahippocampal gyrus. He was imagining walking around his house—no.

'Do you have any brothers?'

Tennis—yes.

'Do you have any sisters?'

House—no."

Constantly foisted upon us by scientists and science writers, the claim that particular regions of the brain "light up" under brain scanning is untrue. Such claims are visually enforced by extremely deceptive visuals in which tiny differences of less than 1 percent are shown in bright red, thereby causing people to think the very slight differences are major differences. The truth is that all brain regions are active all the time. When a brain is scanned, there are only tiny signal differences that show up in a brain scan.  Typically the differences will be no greater than about half of one percent, smaller than 1 part in 200.  When scanning a brain, you can always see dozens of little areas that have a very slightly greater activity, and there is no reason to think that such variations are anything more than very slight chance variations. Similarly, if you were to analyze the blood flow in someone's foot, you would find random small variations in blood flow between different regions, with differences of about 1 part in 200. 

Because of such random variations, there would never be any warrant for claiming that a person was thinking about a particular thing based on small fluctuations in brain activity. At any moment there might for random reasons be 100 different little areas in the brain that had 1 part in 200 greater activity, and 100 other different little areas in the brain that might have 1 part in 200 less activity.  In this case no evidence has been provided of any ability to read thoughts of a person supposed to be in a vegetative state. We cannot reliably distinguish any signal from the noise. 

The New Yorker article describing the case above refers us to a Los Angeles Times article entitled "Brains of Vegetative Patients Show Signs of Life." The article gives us no good evidence that thoughts were read from this patient 23. The article merely mentions that 54 patients in a vegetative state had their brains scanned, and that one of them (patient 23) seemed "several times" to answer "yes" or "no" correctly, based on examining fluctuations of brain activity.  Given random variations in brain activity, you would expect to get such a result by chance if you scanned 54 patients who were completely unconscious. So no evidence of either consciousness or thought reading has been provided.  

A look at the corresponding scientific paper  shows that the fluctuations in brain activity were no more than about a half of one percent. No paper like this should be taken seriously unless the authors followed a rigorous blinding protocol, but the paper makes no mention of any blinding protocol being followed.  Under a blinding protocol, anyone looking for signs of a "yes" or "no" answer would not know whether a "yes" answer was the correct answer.  The paper provides no actual evidence either of thought reading by brain scanning or even of detection of consciousness. We merely have tiny 1-part-in-200 signal variations of a type we would expect to get by chance from scanning one or more of 54 patients who are all unconscious.  

The paper tells that six questions were asked, and the authors seemed impressed that one of the 54 patients seemed to them to answer all six questions correctly (by means of brain fluctuations that the authors are subjectively interpreting).  The probablility of getting six correct answers to yes-or-no questions by a chance method such as coin-flipping is 1 in two-to-the-sixth-power, or 1 in 64.  So it is not very unlikely at all that you would get one such result testing 54 patients, purely by chance, even if all of the patients were unconscious and none of them understood the instructions they were given.  

The New Yorker article then introduces Princeton scientist Ken  Norman, incorrectly describing him as "an expert on thought decoding." Because no progress has been made on decoding thoughts from studying brains, no one should be described as an expert on such a thing. The article then gives us a very misleading passage trying to suggest that scientists are making some progress in understanding how a brain could produce or represent thoughts:

"Now, Norman explained, researchers had developed a mathematical way of understanding thoughts. Drawing on insights from machine learning, they conceived of thoughts as collections of points in a dense 'meaning space.' They could see how these points were interrelated and encoded by neurons." 

To the contrary, no neuroscientist has the slightest idea of how thoughts could be encoded by neurons, nor have neuroscientists  discovered any evidence that any neurons encode thoughts. It is nonsensical to claim that thoughts can be compared to points in three-dimensional space. Points in three-dimensional space are simple 3-number coordinates, but thoughts can be vastly more complicated. If I have the thought that I would love to be lounging on a beach during sunset while sipping lemonade, there is no way to express that thought as three-dimensional coordinates. 

We then read about some experiment:

"Norman invited me to watch an experiment in thought decoding. A postdoctoral student named Manoj Kumar led us into a locked basement lab at P.N.I., where a young woman was lying in the tube of an fMRI scanner. A screen mounted a few inches above her face played a slide show of stock images: an empty beach, a cave, a forest. 'We want to get the brain patterns that are associated with different subclasses of scenes,' Norman said." 

But then the article goes into a long historical digression, and we never learn of what the result is from this experiment. Norman is often mentioned, but we hear no mention of any convincing work he has done on this topic. Inaccurately described as "thought decoding," the attempt described above is merely an attempt to pick up signs in the brain of visual perception. Seeing something is not thinking about it. Most of the alleged examples of high-tech "mind reading" are merely claimed examples of picking up traces of vision by looking at brains -- examples that are not properly called "mind reading" (a term that implies reading someone's thoughts).

We hear a long discussion often mentioning Ken Norman, but failing to prevent any good evidence of high-tech mind reading. We read this claim about brain imaging: "The scripts and the scenes were real—it was possible to detect them with a machine." But the writer presents no evidence to back up such a claim. 

Norman is a champion of a very dubious analytical technique called multi-voxel pattern analysis (MVPA), and seems to think such a technique may help read thoughts from the brain. A paper points out problems with such a technique:

"MVPA does not provide a reliable guide to what information is being used by the brain during cognitive tasks, nor where that information is. This is due in part to inherent run to run variability in the decision space generated by the classifier, but there are also several other issues, discussed here, that make inference from the characteristics of the learned models to relevant brain activity deeply problematic." 

In a paper, Norman claims "This multi-voxel pattern analysis (MVPA) approach has led to several impressive feats of mind reading."  Looking up two of the papers cited in support of this claim, I see that only four subjects were used in each study.  Looking up another of the studies cited in support of this claim, I find that only five subjects were used for the experiment cited. This means none of these studies provided robust evidence (15 subjects per study group being the minimum for a moderately reliable result). This is what goes on massively in neuroscience papers: authors making claims that other papers showed some thing that the papers did not actually show, because poor methodology (usually including way-too-small sample sizes) occurred in the cited studies.   

The New Yorker article then discusses a neuroscientist named Jack Gallant, stating the following: "Jack Gallant, a professor at Berkeley who has used thought decoding to reconstruct video montages from brain scans—as you watch a video in the scanner, the system pulls up frames from similar YouTube clips, based only on your voxel patterns—suggested that one group of people interested in decoding were Silicon Valley investors."  Gallant has produced a Youtube.com clip entitled "Movie Reconstruction from Human Brain Activity."

On the left side of the video we see some visual images. On the right side of the video we see some blurry images entitled "Clip reconstructed from brain activity."  We are left with the impression that scientists have somehow been able to get "movies in the mind" by scanning brains. 

However, such an impression is very misleading, and what is going on smells like smoke and mirrors shenanigans.  The text below the video explains the funky technique used.  The videos entitled "clip reconstructed from brain activity" were produced through some extremely elaborate algorithm that mainly used inputs other than brain activity. Here is the description of the technique used:

"[1] Record brain activity while the subject watches several hours of movie trailers. [2] Build dictionaries (i.e., regression models) that translate between the shapes, edges and motion in the movies and measured brain activity. A separate dictionary is constructed for each of several thousand points at which brain activity was measured....[3] Record brain activity to a new set of movie trailers that will be used to test the quality of the dictionaries and reconstructions. [4] Build a random library of ~18,000,000 seconds (5000 hours) of video downloaded at random from YouTube. (Note these videos have no overlap with the movies that subjects saw in the magnet). Put each of these clips through the dictionaries to generate predictions of brain activity. Select the 100 clips whose predicted activity is most similar to the observed brain activity. Average these clips together. This is the reconstruction."

This bizarre and very complicated rigmarole is some very elaborate scheme in which brain activity is only one of the inputs, and the main inputs are lots of footage from Youtube videos.  It is very misleading to identify the videos as "clip reconstructed from brain activity," as the clips are mainly constructed from data other than brain activity. No actual evidence has been produced that someone detected anything like "movies in the brain." It seems like merely smoke and mirrors under which some output from a variety of sources (produced by a ridiculously complicated process) is being passed off as something like "movies in the brain." 

Similar types of extremely dubious convoluted methods seem to be going on in the papers here co-authored by Gallant:

• "Reconstructing Visual Experiences from Brain Activity Evoked by Natural Movies"
• Bayesian Reconstruction of Natural Images from Human Brain Activity"

In both of these papers, we have a kind of byzantine methodology in which bizarre visual montages or artificial video clips are constructed. For example, the second paper resorts to "an averaged high posterior (AHP) reconstruction by averaging the 100 clips in the sampled natural movie prior that had the highest posterior probability." The claim made by the New Yorker -- that Gallant has "used thought decoding to reconstruct video montages from brain scans" is incorrect. Instead, Gallant is constructing visual montages using some extremely elaborate and hard-to-justify methodology (the opposite of straightforward), and brain scans are merely one of many inputs from which such montages are constructed.  This is no evidence of technology reading thoughts or imagery from brains.  In both of the papers above, only three subjects were used. 15 subjects per study group is the minimum for a moderately compelling experimental result. And since neither paper uses a blinding protocol, the papers fail to provide robust evidence of anything. 

The rest of the New Yorker article is mainly something along the lines of "well, if we've made this much progress, what wonderful things may be on the horizon?" But no robust evidence has been provided that any progress has been made in reading thoughts or mental imagery from brains. The author has spent quite a while interviewing and walking around with scientist Ken Norman, and has accepted "hook, line and sinker" all the claims Norman has made, without asking any tough questions, and without critically analyzing the lack of evidence behind his more doubtful claims and the dubious character of the methodologies involved. The article is written by a freelance writer who has written on a very wide variety of topics, and who shows no signs of being a scholar of neuroscience or the brain or philosophy of mind issues.  

There are no strong neural correlates of either thinking or recall. As discussed here, brain scan studies looking for neural correlates of thinking or recall find only very small differences in brain activity, typically smaller than 1 part in 200. Such differences are what we would expect to see from chance variations, even if a brain does not produce thinking and does not produce recall.  The chart below illustrates the point. 

What typically goes on in some study claiming to find some neural correlate of thinking or recall is professor pareidolia. Pareidolia is when someone hoping to find some pattern reports a pattern that isn't really there, like someone eagerly scanning his toast each day for  years until he finally reports finding something that looks to him like the face of Jesus. A professor examining brain scans and eagerly hoping to find some neural signature or correlate of thinking or recall may be as prone to pareidolia as some person scanning the clouds each day eagerly hoping to find some shape that looks like an angel. 

There are ways for scientists to help minimize the chance that they are reporting patterns because of pareidolia. One way is the application of a rigorous blinding protocol throughout an experiment. Another way is to use adequate sample sizes such as 15 or 30 subjects per study group. Most neuroscience experiments fail to follow such standards. The shockingly bad tendencies of many  experimental biologists was recently revealed by a replication project that found a pitifully low replication rate and other severe problems in a group of biology experiments chosen to be replicated.

Postscript: The latest example of needless risk to subjects for the sake of unfounded "mind reading by brain scanning" claims is a study with a preprint entitled "Semantic reconstruction of continuous language from non-invasive brain recordings." The study failed to show any good evidence for anything important, as it used a way too-small study group sizes of only three subjects and seven subjects (15 subjects per study group is the minimum for a moderately impressive result). Following Questionable Research Practices, the scientists report no sample size calculation, no blinding protocol, no pre-registration, no control group, and no effect size. The only "statistical significance" reported is what smells like "p-hacking" kind of results of the bare minimum for publication (merely p < .05). For these basically worthless results, seven subjects endured something like 16 hours of brain scanning with a 3T scanner, which is more than 30 times longer than they would have had for a diagnostic MRI.  Senselessly, this study has been reported by our ever-credulous science press as some case of reading thoughts by brain scanning. It is no evidence of any such thing.