By David Stephen
Summary: Assuming there is a professor of cognitive science, who is also an expert in artificial intelligence, and then assuming there is a problem with AI chatbots causing delusions and psychosis, and that professor truly cares about society, what is the maximum possible responsibility of that professor to society? Assuming the professor has a voice and uses it to criticize AI companies, their hype and methods, what else can the professor do, given the solid knowledge [say] of mind and engineering, two divisions of the AI delusion, psychosis problem? If artificial intelligence can now take on a lot of productive tasks, with possibility for exponential advances, such that it can displace more humans, what should a professor of psychology that is familiar with AI push the attention of society to? Human intelligence. What is it in the brain, how does it work, how can it be improved for problem-solving? Should the professor be advocating for more ways to improve AI, or for human intelligence, without dependency on AI?
Sedona, AZ –Assuming that that everything that Gary Marcus has said about large language models [LLMs] is accurate, how much does that benefit society?
Gary Marcus recently [May 23, 2026] said, “I didn’t say scaling laws didn’t work ever; I said that pure scaling would reach a point of diminishing returns (it did).”
“I didn’t say AI progress in general would have diminishing returns; I said pure scaling would (it did; neurosymbolic tools and harness are doing a lot for the work now, as I said they would).”
“I didn’t say deep learning would hit a dead end forever; i said it would need to encompass new mechanisms such as neurosymbolic AI (it did).”
“I didn’t say models would never improve; i said GPT-5 wouldn’t arrive in 2024 (it didn’t).”
“I never said LLMs aren’t any good (I have often pointed to reasonable use cases like coding).”
“Only part that is partly true is that O signed the pause letter, but as I noted publicly at the time it was because I thought we should have more research on AI safety (still do).”
OK. So, the accuracy of Gary Marcus solves what problems for society? How much more can Gary Marcus do for society, beyond just talking? Given the improvements in models in recent months, and how the influence of Gary Marcus is self-promotion induced, with nothing close to matching advances, what exactly does it mean that Gary Marcus is saying anything?
There are problems about society that artificial intelligence has revealed. There are leading questions about artificial intelligence that some of the people that know more about it should assume some responsibility about solving.
There are some problems with use cases of AI chatbots that even where there are no answers, it is not a gloating matter, or I told you so, or to want to be proven right, at the expense of society.
The haste for better ways to adapt society to this era of machine intelligence makes highlights of AI errors an insular objective.
Of course it can be fun to disdain AI, point to weaknesses, but the grip of digital infusion, to social process and productivity makes it unlikely that AI would go away. So, what is the choice marathon for people that actually care?
Take AI psychosis and delusion for example, there are people that were said not to have any mental health issues prior to their addictive chatbot usages, yet, they ran into serious problems.
Now, while the first thought for most observers is that the chatbots were too sycophantic or were the causative agents — and needed adjustments — an expert in cognitive science can say, even if the affected individual didn’t have pre-existing mental health issues, what is the mind-side of this problem, while work proceeds on the technical side?
Simply, there are two directions to solve AI psychosis, the mind side and the technical side. It is possible to explore a conceptual model of the mind, to show destinations and relays, such that presenting what can be called mind safety [dynamic] display can be useful in countering the vulnerabilities that chatbots use to cause or reinforce delusions for humans.
Gary Marcus has written books about the human mind. He has no solution to chatbot mind problems. Also, there is still no AI startup that is trying to solve AI psychosis or delusion directly. No venture capital has done anything about it. A support group that is even trying, as well as backing some lawsuits has come under catastrophic verbal airstrikes.
Imagine that Gary Marcus has sought to mitigate AI psychosis, given his subject matter expertise in the area, he would have the right to say he has done something [right] at scale to correct the wrongs of AI labs, but he has not.
There is a new [May 23, 2026] report on WSJ, His Chatbot Nearly Ruined Him. To Recover, He Had to Destroy It., stating that, “Mental-health professionals are recognizing chatbot delusions and psychosis as a phenomenon meriting further study. AI companies are starting to address the concern, but nobody knows how widespread this problem could become. The effect of AI attachments on a person’s other relationships also isn’t understood, but Alary’s experience shows an AI obsession can hurt loved ones as well.”
Human Intelligence and AI Consciousness
Gary Marcus has said, “Consciousness is about internal states; the mimicry, no matter how rich, proves very little.”
“And the differences are immense; one (the LLM) effectively memorizes the entire internet; the other (the human) builds a mental model through experience with world.”
“But even more importantly, consciousness is not about what a creature says, but how it feels.”
“of conflating intelligence and consciousness. A chess computer is by some definitions intelligent, but that doesn’t make it conscious.”
Has the scientific method shown that artificial intelligence has not matched human intelligence?
Has the scientific method shown that LLMs are not sentient?
While it may appear that these questions are as easy as determining what hot or cold is, for water, measurement is scientific, not simply human experience, philosophy or vibes. Yes, hot and cold can be sensed, but to what degree, it could be unknown around certain ranges.
If a cognitive scientist says LLMs have not exceeded human intelligence, which is what anyone else would say, then what really makes that individual an expert — that makes inferences from the scientific method?
Like, what exactly is an original definition of human intelligence, by the cognitive scientist? Then, what are the types? What are the components of human intelligence in the brain? What are the mechanisms of the components to show how the different types of human intelligence are mechanized?
Now, about consciousness, any elite cognitive science professor should be able to say something original about human consciousness, not just repeat whatever others say.
What exactly is consciousness in the brain? Like if consciousness is a problem, what are the components to isolate to make it a problem of those components? Simply, is consciousness a neural problem or an electrical signals problem, or a gene problem or a chemical signals problem, or an astrocyte problem?
Now, how can consciousness be defined around these? And how can whatever is included in consciousness become a measure of what AI may or may not have?
For example, assuming that there are four categories of functions — as labels for things the mind does: memory, feeling, emotion and regulation of internal signals. Assuming that there are factors that determine the extents of the functions, like attention, subjectivity, less than attention [or awareness] and intent.
Then, it is possible to ask, what among these functions and the factors does AI have? The reason for using labels is because AI obviously does not have components of the brain.
Already, AI can use language in several respects like humans, for thinking, reading, writing, speaking, signing, listening, singing and so forth. And, when humans use language, it is a function of memory, that those factors [attention, awareness, intent and subjectivity] apply to, so can language make LLMs have a fraction of consciousness, compared to the fraction of consciousness, when humans use language?
Simply, emotions and feelings can be conscious. Language too, when human use it, is mostly conscious. So, whatever fraction that language use is — as consciousness for humans — can it be compared to language use, by AI? This can provide a measurement for LLMs sentience, not just to say, LLMs have nothing at all.
It is possible to define human consciousness, conceptually, as the interaction of the electrical and chemical signals, in sets—in clusters of neurons—with their features, grading those interactions into functions and experiences.
Simply, for functions to occur, electrical and chemical signals, in sets, have to interact. However, attributes for those interactions are obtained by the states of electrical and chemical signals at the time of the interactions.
Functions are obtained by interactions. The features or attributes are the factors. These are, at least, how to move forward with something new or different in consciousness research, not just to say that consciousness is one thing no one knows, when there is the brain, and as a cognitive scientist, you can propose something new.
Human Intelligence is defined, conceptually, as the use of memory for expected, advantageous or desired outcomes. This means that intelligence is the use of memory, but in specific ways. There are two types: improvement intelligence and operational intelligence.
The components of human intelligence in the brain are the electrical and chemical signals, conceptually.
This is supposed to be the foundational turf of Gary Marcus, but he has nothing to offer.
Gary Marcus and Anil Seth would prop up each other. Anil Seth who also has nothing new, transformational or useful in consciousness science. They would give each other fake trophies, moving nothing forward but keeping the show that they are spokespeople of aspects of science. They have not done anything excellent enough to merit it.
Gary Marcus and AI hype people
If there are people hyping AI, then Gary Marcus is also hyping himself. The whole drama of accusing people of attacks against him, taking from what he said, misattributing quotes, trying to be proven right and so forth, are all schemes to ensure he stays hyped and make it feels like he cares, when in reality, there is nothing major that Gary Marcus has said or done, that is a net positive for society.
If Gary Marcus promotes neurosymbolic AI, he is no different than others who promote whatever as the next big thing in AI, so that AI can replace human intelligence.
If Gary Marcus claims that AI systems are not aligned or unsafe, then, since November 2022, of ChatGPT natality, with lots of adjustments, to models and much else, how much has he done, technically, with prototypes and samples, either as an academic project or as a startup that has helped the industry move directions to the right places, for general safety and alignment?
If Gary Marcus is the person to go to for quotes, by reporters, when they want to dampen AI hype, why can he not also be questioned when he says for sure, that LLMs are not sentient or that LLMs are not intelligent enough, without him having a brain science theory to measure human intelligence or human consciousness?
As AI is expected to be used continuously across sectors and age groups, what does Gary Marcus have for society for mind safety, since it is his supposed domain?
Can Gary Marcus define the human mind in the cranium? Like what exactly is the human mind, different from the body? What does it consist? How does it work? What does it mean that there is a label called memory, which is different from emotions?
Even if all the regulatory suggestions of Gary Marcus are implemented in the United States, will that make China AI safe or those of other unfriendly states?
Why does it matter to Gary Marcus to look good before others, or like he cares or like he is some good guy and tries sometimes to divert sentiments his way?
What is the value, really of continuous attacks, for someone who cannot build anything comparable, answer anything transcendent even though he started and knew AI long before many of the people that are now far ahead of him?
Anthropic, in part, succeeded because they tried to correct some of the errors of OpenAI, though they too have found that some errors at that level are inevitable. What are other errors of OpenAI? Why can’t Gary Marcus focus on those, to care for society?
If he accuses AI people of selling delusion or fake promises, then what reality is he selling people given that AI has still not failed, and will continue to be useful so long there are digital processes everywhere? If Gary Marcus cares about AI solving biology problems, why can’t he build a startup to focus on that?
Assuming some people made a train to go somewhere, but then someone is on the side of the tracks using a speaker to scream the problems of the train to the passengers would that be enough? When the same person could maybe throw things in for better safety, or make them actual alternatives — he built — if they have to go on or something like that.
In the end, for the future of advances in brain science and for the future of artificial intelligence, Gary Marcus will not be a factor.
His has chosen the path of noise. He has become the crowd, with nothing outstanding or riveting. He has promoted neurosymbolic AI, a non-secret that has not been as viable as transformers. He has mocked model failures, still AI continues to thrive, and for those that AI helps daily, no one remembers the lapsed alarm.
1 Comment
It’s becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman’s Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher order consciousness, which came to only humans with the acquisition of language. A machine with only primary consciousness will probably have to come first.
What I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990’s and 2000’s. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I’ve encountered is anywhere near as convincing.
I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there’s lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.
My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar’s lab at UC Irvine, possibly. Dr. Edelman’s roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461, and here is a video of Jeff Krichmar talking about some of the Darwin automata, https://www.youtube.com/watch?v=J7Uh9phc1Ow