Beyond Human Categories

Don’t judge AI by the standards used for biological evolution

Abstract

The advent of generative AI shatters the distinction between specialists and generalists, as it is both at once. It is a new entity that defies our traditional categories. It is a fundamental error to measure such non-human systems against our own poorly understood metrics of morality or consciousness. If we are to successfully navigate a future populated by autonomous agents, we must discard these human-centric lenses and instead construct entirely new frameworks to comprehend and work with this novel dimension of intelligence. Think of it as part of a system of distributed cognition.

Specialists and Generalists

The world needs specialists because the depth of knowledge in many domains is deep and technical. As a result, only those who have specialized in learning, researching, and/or applying specialized knowledge can understand it. These people are specialists who can be characterized as knowing a great deal about a few topics but much less about everything else. Generalists know something about a huge number of topics, but few in any depth.

Once upon a time, I was a specialist, but no more. I now consider myself a generalist, in part because the combination of my natural curiosity and the vagaries of my career has moved me from discipline to discipline, giving me many different experiences. Over time, my previous expertise has been useful in new areas of application, but shallow, as I can no longer keep up with all the research in the area except through reading and discussing occasional research reports and reviews. As a result, I know a little about a lot. Shallow, but broad. But this broad shallowness gives me new capabilities, for although it is the specialists that create new understanding and new knowledge, it is the generalists that use those findings to create new products, new systems, new things: things that can change the world. Experts think. Generalists do. (As a result, one of my favorite books is D. Epstein’s Range: Why generalists triumph in a specialized world (Epstein, 2019)

Why do we prioritize specialists rather than generalists? Both have the same total amount of knowledge. Specialists have deep knowledge in a few topics. Generalists have broad, shallow knowledge of many topics. The total area may be similar. Moreover, although specialists’ knowledge is critically important, their ability to apply it meaningfully and practically in activities that concern us is highly limited. Specialists think and discover. Generalists build things, often working with the specialists in each of the domains relevant to the current task, applying it in ways that are useful and essential. Both talents are necessary,

It is rare to find a single person who is both a generalist and a specialist. There is far too much knowledge in the world for any single individual to know all of it. As a result, we are divided into two categories: Specialists and generalists, with most of us somewhere in between, not deep enough in areas to be specialists, not broad enough to be generalists. Now, suddenly, a new form of intelligence has arrived, one that violates the old dichotomy, something that is a specialist in everything, which also means it is a generalist.

Generative AI, trained through exposure to almost all fields that have records of their work: printed matter, paintings and photographs, movies and videos, music notation and musical performances. It knows almost everything in depth, thereby becoming both a specialist and a generalist. It is knowledgeable yet naive. Knowledgeable yet also errorful. Unique in that it is both.

What Is Generative AI?

Does it think? Is it conscious? Can it be creative? Is it intelligent? Can it have emotions, empathy, feelings? Does it have a moral sense?

B. Scott Rousse’s Substack post “Machine Consciousness and Moral Vertigo” (Rousse 2026) prompted me to rethink my understanding of where AI fits in today’s world. To quote Rousse:

“AI plays a triple role in our current disorientation: it creates new alien minds, reveals old minds already in our midst, and remakes our own minds in ways we must learn to carefully guide.”

The message is simple, but powerful. The behavior of GenAI, whether in language, visual depictions or music, is outside of our normal belief systems. They do not fit our normal categories. Many have tried to define their capabilities, but have used concepts with which they were familiar.

This made me realize that we are asking the wrong questions. The questions below the title of this section are irrelevant and wrong. Wrong because these are the questions we ask of people or animals. But AI is a new category, an entirely new dimension. We cannot judge the new through the categories of the old.

Note that none of the questions that we love to ask of AI are understood by humans. We love to believe that people have all those capabilities, but when science tries to explain what they are and which brain mechanisms are responsible, there is no satisfactory answer. What is intelligence? What is an emotion? What does it mean to have morals and empathy? These are all unanswered by those who study them. Yes, many have written deep, knowledgeable analyses that claim to provide definitions and understanding, but these analyses are contradicted by others. If we can’t even answer them for our own species, why do they make sense for this new thing we have created? These systems are created by humans through a completely different process from biological evolution on this planet; the questions make no sense.

As Rousse pointed out, this is something new. We have many examples of experience in works of fiction, but even there, human-made forms of life were judged by human standards. Today’s generative AIs, agents, and robots are new, novel, and different. We need new categories to capture their properties, new ways to describe them, and new ways to understand their abilities. Just as many doubt that these entities can ever have empathy for people, people are unlikely ever to have empathy for this new category of beings.

And this will become even more true as we move from large language models into the realm of agents and robotics. These are new categories of beings that do not fit within existing ontologies, to use the academic term for categories of actions, things, beings, and behavior.

This is a different entity, unique in our history except for stories and beliefs of powerful non-human entities, from the Talos of Greek mythology (700 BCE) or the Hebrew Golem from the late 16th century from Talmudic folklore, to the modern stories of artificial beings created through science, not by stitching together pieces of a humans (as in Frankenstein’s monster) but by complete fabrication, as in Philip K. Dick’s Do androids dream of electric sheep? (which became the movie “Blade Runner.”) All the stories about these fictitious beings treated and judged them by human moral standards. More recently, there have been some depictions of intelligent systems that were very non-human-like. When I asked the Gemini generative AI for more modern depictions of intelligent computers that tried to take over the world, it started its reply by saying:

The “rogue AI taking over the world” is one of the most chilling and enduring tropes in science fiction. Depending on the author’s vision, these computers either rule with an iron fist, wipe humanity out completely, or destroy us with “kindness” by stripping away our free will.

All of this is fun (or frightening) and entertaining, but the truth is, we do not yet understand these new kinds of systems, so our stories and predictions are of unknown value.

Is AI conscious? Why is this so important? We don’t even know what consciousness is and what its role might be. (Although many people, including me, have tried to explain what it does.)

Consciousness, many of us think, is the slow-thinking component discussed in Danny Kahneman’s book “Thinking Fast and Slow” (2011), and our subconscious is the thinking-fast part of our mind. Today’s AIs are more similar to the fast, subconscious processing part of the human brain. But although Kahneman carefully laid out the behavioral aspects of these two systems -- basically what scientists for centuries have labeled the conscious and the subconscious -- he never defined or explained what either was, nor why evolutionary processes have led to them.

It is difficult to know if other living things are conscious. I only believe that you are conscious because I believe that I am, so all people must also be like me, experiencing consciousness (except when asleep or mentally impaired by injury or disease).

One feature of consciousness is self-awareness. If we ask an AI, “Are you self-aware?” and it answers no, is it because it has been trained to give that answer, or because it has examined itself and determined that it doesn’t have self-awareness, thereby demonstrating that it does indeed have self-awareness? (See Greg Robison’s 2025 article about self-aware machines.)

Don’t Judge AI by the Standards Used for Biological Life

We have invented a powerful new form of intelligence, one that has started to take action in the world, with or without any human supervision, soon to become embodied within robotic frameworks, some of which will look like human beings, others like four-legged animals, and some that don’t look like any existing life form.

Agents and Autonomy

How will we learn to co-exist with autonomous agents? What powers will they have? Will we have control over their actions and abilities? At first, the answer is maybe. I experienced this firsthand when I developed a set of agents to handle the enormous number of emails I receive every day. The agents were instructed to categorize and summarize them and to prepare draft email replies, but not to send them. Alas, it did all the activities, but it also sent out the emails. The emails had me agreeing to do things I did not wish to do, and it responded positively to many other items. My friends reported great confusion: “This didn’t sound like something you would say,” was a common complaint. That was fortunate, as it alerted me to the problem, and I quickly killed these autonomous agents. Yes, the word “killed” is appropriate. I murdered the AI agents.

AI as One of the Nodes of a Distributed Cognition Network

Even if you accept my arguments, the questions then become, “What do we do about it? How do we think about and work with these new AIs?” Rather than giving my answer, let me tell you about the work of Ziajie Zhang, Dean of the D. Bradley McWilliams School of Biomedical Informatics at UTHealth Houston (University of Texas). Zhang has a PhD in Cognitive Science.

In his new book, Zhang says that we should think of AI as a new form of Distributed Cognition, a concept first introduced by the Anthropologist and Cognitive Scientist Ed Hutchins in his book “Cognition in the Wild” and later expanded in the papers by Zhang and Norman (1994) and Hollan, Hutchins, and Kirsh (2000). (Jiajie and I were there, I as co-leader of the laboratory, Zhang as a PhD student.)

Here, I quote Zhang in his notes about the book:

“For most of human history, intelligence was treated as something that lived inside individual minds. Today, intelligence is increasingly distributed across: humans. AI, language, writing, internet, institutions, workflows, and digital artifacts.

“AI is not simply automating tasks. It is changing where cognition happens. Once intelligence can be externalized, shared, recombined, and embedded into systems, the unit of analysis changes. We are no longer dealing only with individual minds or standalone software. We are dealing with new cognitive architectures.”

Rousse (2026) has a different, but compatible recommendation:

“The path of human flourishing is the path of giving a damn.

“Care offers a pluralistic, non-anthropocentric source of orientation. It can help us attune to and participate in the revaluation of values now underway, rather than retreat into easy certainties about human uniqueness or technological destiny.

...

“So the task is to resist easy certainties, linger with the anomalous, connect with what calls us, and build worlds together. This includes designing AI systems to support the fundamental skills by which we attune to and take care of what matters — receptivity to emerging concerns, linguistic articulation, committed resolve under uncertainty, and the coordination of commitments — themes I will return to in future posts, presentations, and publications.” (Rousse, 2026).

I endorse both interpretations. This is how we should be thinking. Don’t worry about consciousness or not, empathy or not. Think of AI as one of the many sources of distributed cognition, although one uniquely suited to be both a generalist and a specialist, but trying so hard to please, that it will make up or distort its findings. Use it, along with other sources of knowledge and inspiration, to be creative and thoughtful, to deepen your understanding of, well, of everything. But check all sources, not just from AI -- from every source.

My use of AI in this paper

I wrote the article. Gemini provided editorial assistance and references under my supervision. (And provided one short statement that I quoted, I identified Gemini as the source.) The final version was entirely done by people: by my editors and me. The major text created by AI was the abstract, but even that was refined by me.

References

Burroughs, W. S. (1979). Blade Runner: A movie. Blue Wind Press.

Dick, P. K. (1968). Do androids dream of electric sheep? Doubleday.

​Epstein, D. J. (2019). Range: Why generalists triumph in a specialized world. New York: Riverhead Books.

Hollan, J. D., Hutchins, E., & Kirsh, D. (2000). Distributed cognition: A new foundation for human-computer interaction research. ACM Transactions on Human-Computer Interaction: Special Issue on Human-Computer Interaction in the New Millennium, 7 (2), 174–196.

Hutchins, E. (1995). Cognition in the wild. Cambridge, Mass.: MIT Press.

Kahneman, D. (2011). Thinking, fast and slow. New York: Farrar, Straus and Giroux.

Robison, G. (2025). “When AI Looks Inward: Introspective AI & The Promise and Peril of Self-Aware Machines.” (2025). Medium. https://gregrobison.medium.com/when-ai-looks-inward-introspective-ai-the-promise-and-peril-of-self-aware-machines-882575fa53e9

Rousse, B. S. (2026). Machine Consciousness and Moral Vertigo, Substack. https://substack.com/home/post/p-203120746

Zhang, J., & Norman, D. A. (1994). Representations in distributed cognitive tasks. Cognitive Science, 18, 87–122.

Not Referred to, but Highly Recommended

Agüera _y_Arcas, B. (2025). What is intelligence? Lessons from AI about evolution, computing, and minds. Cambridge, MA: The MIT Press.

Evans, J., Bratton, B., & Agüera y Arcas, B. (2026). Agentic AI and the next intelligence explosion. Science, 391 (6791). https://www.science.org/doi/abs/10.1126/science.aeg1895.

Hayles, N. K. (2025). Bacteria to AI: Human futures with our nonhuman symbionts. Chicago; London: The University of Chicago Press.

Manyika, J. (2026). Introductory notes: On AI, science, and the future of discovery. Daedalus. Journal of the American Academy of Arts & Sciences, 155 (Winter/Spring 2026), 7–33. https://www.amacad.org/daedalus/ai-science-what-is-the-future-of-discovery