Why AI’s Biggest Problem May Be Its Own Rhetoric

Why AI’s Biggest Problem May Be Its Own Rhetoric

By Walter Donway

Two media narratives about artificial intelligence now dominate the news cycle—and at first glance seem oddly disconnected.

The first is the growing “backlash” against AI. Polls show declining public trust with one Pew Research Center poll finding that some 50 percent of Americans are more concerned than excited about AI versus only 10 percent more excited. Teachers worry about the collapse of writing and learning. Artists and musicians protest plagiarism and synthetic imitation. Workers fear job displacement. Regulators warn about fraud, deepfakes, cyberwarfare, autonomous weapons, and the concentration of unprecedented power in a handful of corporations and governments. Even ordinary users increasingly complain about “AI slop,” fabricated information, and the creeping artificiality of online life.

The second narrative is both triumphal and anxious.  Leaders of the AI industry speak in sometimes awed tones about artificial general intelligence, “agentic AI,” “digital minds,” “recursive self-improvement,” and “superintelligence.” They describe a race—among corporations and increasingly between China and the United States—to create systems capable of rivaling or surpassing human intelligence itself. David Wallace, writing in the  NYT Magazine, reported that “Plenty of A.I. researchers believe it is right around the corner; Anthropic’s Jack Clark predicted this week that fully independent recursive self-improvement [that is, AI taking over its own evolution] might be less than two years away.”

President Donald Trump’s policy chief for AI, Dean Ball, told a recent conference at Yale University: “It will not be A.I. in government. It’s going to be A.I. as governments.” [Emphasis added]

The two stories are usually treated separately but shouldn’t be.                   

The backlash against AI is not driven solely by practical fears about jobs, privacy, education, scams, or national security. A powerful undertow runs beneath it: the fear that humanity is attempting to create not merely a revolutionary tool, but a rival form of intelligence with genuine agency of its own.

And who could blame the public for drawing that conclusion? The industry itself increasingly speaks in precisely those terms.

Again, many concerns are entirely legitimate. Autonomous weapons systems are rapidly transforming warfare. Cybersecurity threats are posed as AI systems become capable of discovering and exploiting software vulnerabilities at extraordinary speed. Large sectors of white-collar labor may be disrupted. Governments may deploy AI for surveillance and social control on a scale never before possible as has the People’s Republic of China.

These are concrete concerns that translate into demands for regulation, guardrails, liability rules, transparency, and public oversight.

AI Backlash Turns Metaphysical

But the other strand of the backlash goes much further. It demands not merely control of AI but its destruction or abolition before it “escapes,” “replaces humanity,” or “destroys civilization.” Here the emotional tone changes dramatically. AI is no longer treated as a dangerous technology but as an incipient rival species.

That fear, explicit or implicit, is directed not primarily at today’s systems but at the vision continually projected by the AI industry leadership itself: the emergence of autonomous digital intelligence possessing agency, consciousness, initiative, and perhaps superhuman strategic power.

And if such a thing were possible, the fears would hardly be irrational. A truly agentic, conscious, superintelligent non-biological being would raise questions so vast that they border on the unfathomable. Human society, politics, economics, morality, even the metaphysical meaning of human nature itself—would stand in question. Such a development would not merely introduce a new technology into human civilization. It would introduce a new category of being into the human ecology.

No sane civilization would deliberately induce the birth of such an entity without infinitely deeper understanding than we currently possess.

That is why rhetoric about AI not only generates practical anxiety but sounds metaphysical alarms. Constant references to “digital minds,” “superintelligence,” and “artificial agents” conjures up a future world populated by rival intelligences with goals and powers of their own.

Thus, part of the backlash is rational concern. Part is science-fiction nightmare. But the two are synergistic and now deeply intertwined.

The Myth of Synthetic Agency

Is this confusion necessary?

From the outset, AI research has been driven by competition with human intelligence. AI systems measure success by comparison with human capacities: language fluency, reasoning, coding, mathematics, scientific problem-solving, strategic planning. Alan Turing’s famous imitation test asked if a machine could imitate human conversation closely enough to fool a human judge.

And the term “artificial intelligence,” coined for the famous Dartmouth Conference in 1956, carried an unmistakable implication. It was not called “advanced statistical inference” or “automated symbolic processing.” It was called intelligence.

The terminology, created with marketing in mind, was philosophically loaded. For behind the modern discussion of “agentic AI” lies a much older and unresolved philosophical issue: What exactly is human agency?

When AI executives use terms like “agency” or “digital minds,” they tap into the prevailing modern understanding of intelligence and human consciousness. And Western culture since the eighteenth-century Age of Enlightenment has implicitly, and often explicitly, adopted a mechanistic conception of human intelligence itself. Much modern science and philosophy interpret human thought in such terms: the human mind is ultimately the product of deterministic physical processes in the brain. Free will, volition, and the initiation of thought are often treated with skepticism or dismissed altogether as illusions generated by neural computation. Explicit avowal of free will is largely left to religion.

Nobel laureate Francis Crick, co-discoverer with James Watson of the structure of DNA, put the point starkly in his 1994 book The Astonishing Hypothesis:

“You, your joys and your sorrows, your memories and your ambitions, your sense of personal identity and free will, are in fact no more than the behavior of a vast assembly of nerve cells and their associated molecules. [Emphasis added.]

Stanford University neuroscientist Robert Sapolsky argues in his 2023 book, Determined: A Science of Life Without Free Will, that free will is an illusion, presenting a comprehensive case that all human actions are the result of a seamless chain of biological and environmental causes stretching back in time.”

If human intelligence is mechanistic, then, yes, the aspiration to reproduce it artificially is logical. Build sufficiently complex systems capable of processing information, adapting behavior, learning from data, and regulating action—and eventually agency itself may emerge. Under this view, ‘agentic AI’ appears not mysterious at all, but simply another mechanistic mind.

“This Is War” To Create Super-Intelligence

In his heralded (should I say “blockbust”) new book, The Infinity Machine: Demiss Hassabis, DeepMind, and the Quest for Superintelligence, published in March, author Sebastian Mallaby’s captures a remarkable closeup of the world of AI.  It is based on 30 hours of conversation over three years with the Nobel Prize-winning head of one of the world’s foremost AI research labs, Google DeepMind. In the process of interviewing Hassabis, head of a rare non-US, non-Silicon Valley shop (until acquired by Google) at the very forefront of the race for smarter, faster, more powerful systems, Mallaby had personal access to virtually all the battle chieftains (Hassabis says AI companies are “at war”) such as Sam Altman of OpenAI, Dario Amodei of Anthropic, Larry Page of Google, and Elon Musk.

The report is unambiguous. All are focused on their vision of “general artificial intelligence,” also called “agentic” intelligence or super-intelligence. All, but none more than Demiss Hassabis and his closest colleagues, had been focused on the “dangers” of their technology. There are the immediate and much-discussed perils of job displacement, autonomous battlefield weapons, producing a generation dependent for thinking and writing on chatbots, and so forth. But then, there are what have come to be called the “existential” and” extinction” perils of birthing a super-intelligent agent that breaks free of human control and, driven by its own goals, becomes (in effect) a species with which we cannot compete. To dramatize: the age of humans will be succeeded by the age of the thinking machines.

After years of striving to focus on “risks” of the latter kind, one of its most thoughtful and insistent advocates, Demiss Hassabis, admits that “technological determinism”—the race for the next, best, most profitable AI “release”—has swept the “safety” factor from the C-suite.  Others, who are Silicon Valley insiders but not calling the shots of companies, show up at meeting after meeting to sound the alarm and challenge complacency.

No AI chieftain interviewed by Mallaby is unambiguous about the metaphysical status of the fearsome AI agent; that is, they do not claim that it will be “aware,” “conscious,” in the sense of “mind.” That, however, never has been the premise of AI since Turing. The framework is all about “benchmarks.” How will AI compare with human intelligence? Hassabis, a child chess prodigy, built DeepMind and his reputation on machines that could “beat” humans—notably, AlphaGo, which, after years of development, defeated the world champion at Go—arguably the world’s most complex game. (Beaten in four out of five games, the Japanese world champion retired.  What’s the use?)

In other words, the chieftains of the AI competition, revealed in the incomparably Mallaby interviews, seem to share the premise that consciousness—or its equivalent--reduces in the end to mechanism. If the AI shops at last create a mechanism sophisticated enough, with the right algorithms, the right transformer architecture, enough “compute,” power, and sufficiently oceanic “training,” then consciousness will “emerge” from complexity—at it did when billions of years of evolutionary complexity yielded the brain and mind.

What is remarkable to me, as a philosopher, not a technologist, is that the field of artificial intelligence since its birth with the Turing Test, has never in some 75 years altered its foundational philosophical premises.  Hassabis still harks back to Immanuel Kant’s noumenal world, the inscrutable ding an sich, and Benedict Spinoza’s pantheism. For example, Hassabis says that the fundamental constituent of existence is “information.” Natural enough for a computer guy. But if existence is “information” then information about what?  Isn’t information about something? But if information is the fundamental unit of existence, then there is no content of information.

But it makes a kind of sense in the philosophy of Kant, to which Hassabis repeatedly refers. Kant is addressing the epistemological contradiction of the Enlightenment: David Hume’s argument that we never can know “reality” as it is. We know only reality in the form conveyed by our senses. Kant, in briefest terms, replied: yes, our minds supply all the categories that shape our perceptions—the categories of space and time—and we can know not “reality” but just the reality created for us by our categories.

It seems almost incomprehensible that science today, with its astonishing achievements, seldom questions its philosophical legacy of the Enlighenment’s contradictions and the conclusions of German Idealism (mind, ideas, create “reality”).  But the AI revolution is viewed by its pioneers and today’s leading lights through the lens of eighteenth- and early nineteenth-century philosophy. That is the philosophical foundation of AI’s most contemplative pioneer, Hassabis, and, as far as I can see, the implicit framework of the AI revolution.  

‍

The Trouble with Calling AI  ‘Mind’

Philosophy is not so simplistic. Beginning in the classical world and continuing for centuries until the Enlightenment’s radical skeptics such as David Hume, a far more empirical conception of causality prevailed, one associated especially with Aristotle. Causality was understood not merely as mechanical chains of events—the billiard-ball model—but as rooted in the nature and identity of the entities acting. Causality was the law of identity applied to action. Human action was not viewed as the passive output of prior physical states. Human beings had capabilities specific to their nature, including volitional conceptual thought: Aristotle’s “rational animal.”

This older understanding does not automatically solve the problem of free will for neuroscience, but unlike the mechanistic model it does not contradict it, rule it out on logical grounds alone.

In fact, however, we have overwhelming introspective evidence that we can regulate our own thought at the conceptual level: choose focused mental effort to question assumptions, suspend judgment, confront doubt, compare alternatives, initiate inquiry. And we know introspectively, too, that we can choose to avoid or evade that mental effort. This implication of the Aristotelian view of causality for human volition has been most powerfully put forward today by Ayn Rand.

Neuroscience has not explained away the introspective experience but instead has assumed the mechanistic model of cause and effect to rule free will out of court. Against this, some leading neuroscientists, including Antonio Damasio, chairman of neuroscience at USC, have offered sophisticated accounts of volition and self-regulation grounded in body, emotion, and the experience of self (The Feeling of What Happens: Body and Emotion in the Making of Consciousness, 1999). Thus, we can hypothesize that out of 4.0 billion years of life’s evolution, including increasingly sophisticated types and degrees of self regulation, has emerged a human brain capable of initiating self-regulation at the conceptual level.

Thus, predictions of agentic artificial beings, in any human sense of “agentic,”  hangs by the thread of a highly controversial philosophical interpretation of human intelligence and human consciousness.

None of this diminishes the revolutionary importance of AI. Current systems already represent some of the most powerful technologies ever developed. They are beginning to transform medicine, scientific research, engineering, education, communication, logistics, and economic productivity on a scale comparable to electricity or the internet. AI may become the most general and powerful tool humanity has ever created.

But it is one thing to develop astonishing computational tools. It is another entirely to proclaim as the ultimate goal the creation of autonomous digital beings possessing true agency and superhuman intelligence.

The first vision inspires excitement and investment. The second unnecessarily evokes existential dread. And understandably so. A public repeatedly told that humanity is creating “digital minds” will not be shocked into resigned metaphysical calm. It will imagine replacement, domination, dependency, or extinction. It will imagine Frankenstein, indestructive robotic samurai, rival species, and end of the age of Homo sapiens. The backlash against AI is incited in no small part by the rhetoric of silicon agency.

Concomitantly, of course, the more sophisticated AI becomes, the easier anthropomorphism becomes. Human language is our deepest signal of mind. A system capable of conversation, humor, persuasion, explanation, and emotional imitation naturally triggers intuitions associated with personhood. The line between simulation of agency and actual agency becomes psychologically blurred.

Yet the confusion may ultimately reveal something important—not about machines, but about ourselves.

As I argue at greater length in A Serious Chat with Artificial Intelligence, the debate over machine agency reflects a deeper uncertainty within modern culture about the nature of human reason, consciousness, and volition. The rise of AI has not settled these questions. It has exposed them.

Perhaps the final irony of AI will be that the more clearly humanity sees its own reasoning processes reflected in increasingly sophisticated machines, the more clearly it may come to understand the irreducible uniqueness of human agency.