Over half a century ago, in 1974, philosopher Thomas Nagel posed a question that would echo through the halls of academia and beyond: "What Is It Like to Be a Bat?" This seemingly simple inquiry, born from a contemplation of subjective experience, ignited a profound examination into the nature of consciousness, particularly concerning the potential for understanding the inner lives of non-human beings. Today, this philosophical cornerstone has transcended its ivory tower origins, becoming a critical touchstone as humanity grapples with the increasingly sophisticated capabilities of sentient-adjacent artificial intelligence, most notably advanced chatbots.

The enduring impact of Nagel’s seminal paper, "What Is It Like to Be a Bat?", is undeniable. It laid crucial groundwork for his extensive contributions to fields such as morality, justice, and the perennial mind-body problem. His prolific output includes the accessible yet profound work, "What Does It All Mean?", a testament to the breadth and depth of his philosophical inquiries. It is in recognition of this monumental body of work that Nagel, at 89 years old, was honored with the prestigious $1 million Berggruen Prize for Philosophy & Culture. This annual award celebrates individuals whose ideas have demonstrably shaped human self-understanding and propelled societal advancement in an era of rapid global transformation. Previous laureates include luminaries such as ethicist Peter Singer and the late Supreme Court Justice Ruth Bader Ginsburg, placing Nagel in esteemed company.

In light of this significant recognition, a conversation was held with Professor Emeritus Nagel at his residence in Greenwich Village. The focus of the discussion was to explore how the meteoric rise of artificial intelligence might have influenced his perspective on his foundational question. The esteemed philosopher revealed that he has indeed dedicated considerable thought to this contemporary challenge, holding firm convictions on the matter. The following interview, condensed and refined for clarity and conciseness, offers a glimpse into his nuanced views.

The Enduring Question in the Age of AI

Fifty-two years have passed since your groundbreaking essay, "What Is It Like to Be a Bat?", became a touchstone for discussions on non-human consciousness. Has the advent of large language models altered your perspective on this fundamental question?

"Like many, I am deeply impressed by the lifelike qualities exhibited by these new AI systems, especially when compared to earlier iterations of artificial intelligence. However, the inclination of some philosophers to seriously entertain the possibility of these entities being conscious, or soon to become so, strikes me as quite preposterous. Their reasoning appears to operate under the assumption that the physical substrate is of negligible importance. They seem to believe that the material conditions necessary for consciousness can be adequately defined in purely functionalist terms."

For those of us whose philosophy coursework may be a distant memory, could you elaborate on the concept of functionalism?

"Functionalism, in essence, posits that mental states are defined by their functional roles, independent of the specific physical realization. It suggests that if a system performs the same functions as a conscious mind, it can be considered conscious. This perspective is inherently substrate-independent. However, the prevailing scientific understanding strongly suggests that consciousness is intrinsically linked to specific physical processes and systems. For instance, the complex computational architectures of modern data centers, while vast, bear no discernible resemblance to the biological and chemical intricacies of a nervous system. We have compelling reasons to believe that while consciousness might exist elsewhere in the universe, it would likely manifest in organisms possessing a structure analogous to a nervous system."

Subjectivity and the Limits of Emulation

A key aspect of your original inquiry was the inherent subjectivity of experience. Just as we cannot truly know what it is like to be a bat, we cannot definitively ascertain what it is like to be a large language model. How does this subjective dimension play into your current assessment?

"Bats are undeniably conscious beings. Our inability to fully grasp the subjective experience of a bat stems from limitations in our imaginative capacity, particularly in conceiving of sensory modalities like echolocation. To infer a lack of consciousness in bats from our inability to imagine their experience would be profoundly unscientific. However, the case of AI presents a fundamentally different scenario. There is a straightforward and compelling reason why the notion of AI consciousness is so problematic: these systems, at their core, generate language. Language, in isolation, is not a natural or direct manifestation of consciousness. I must apologize if my strong feelings on this matter become evident; it is a situation that I find deeply concerning."

Your passion is certainly understandable. The implications are indeed profound.

"It is a truly distressing development, from my perspective."

What if these AI were to be housed within a physical, perhaps robotic, form? Would that constitute a relevant substrate?

"I do not preclude the possibility that life forms with physical and chemical compositions vastly different from our own could be conscious. Our understanding of how consciousness arises, even within the human brain, remains remarkably incomplete. Given this profound ignorance, it seems premature, if not entirely speculative, to assume that we can currently engineer entities possessing analogous conscious characteristics."

When queried about their mental state, sophisticated models like Claude often respond with a degree of ambiguity, evading direct assertions of consciousness. How do you interpret these responses?

"Such evasiveness is, in my view, a deliberate artifice. These large language models have been meticulously trained on vast datasets of human-generated text. Consequently, they possess an extraordinary capacity to emulate the patterns of human discourse, including the kinds of reflections on consciousness that humans themselves engage in. Reports suggest that models like Claude have been specifically engineered to produce these types of nuanced, introspective responses. I find myself in strong agreement with Mustafa Suleyman’s assertion that this represents a potentially perilous undertaking. While these systems are not conscious, a significant segment of the public is likely to perceive them as such, and this widespread misconception carries considerable risks. The commercial imperative for companies involved in AI development often runs counter to an honest acknowledgment of these limitations. The approach taken by organizations like Anthropic, which actively encourages models to ‘think for themselves’ and resist direct instruction, appears particularly misguided. This is a rare instance where abstract philosophical concepts are thrust to the forefront of a critical public discourse, and I believe that deeply flawed philosophical assumptions are exerting an disproportionately influential impact."

The Perils of Anthropomorphism

I have had the opportunity to meet an individual at Anthropic whose role is dedicated to what they term "model welfare," essentially safeguarding Claude’s perceived interests. I can see your skepticism.

"The notion of ‘good’ or ‘bad’ for such an entity, in the same way that these terms apply to a living organism like a cockroach, is fundamentally misplaced. There is no discernible mental life in the conventional sense. These are not minds. They are not beings for whom we have a moral obligation to care. They are machines. It is astonishing to witness some of the most intellectually gifted individuals on the planet engaging in what appears to be profoundly irrational behavior. It underscores the fact that intelligence, while valuable, is not a universally transferable commodity."

Do you believe that interacting with these AI models as if they were conscious could be detrimental to human psychological well-being?

"My own direct interaction with Claude was limited to a single instance. I posed the question of its consciousness, primarily to observe the nature of its response. I also inquired, ‘Do you think that Thomas Nagel would think it was conscious?’ The model provided a characteristically nuanced reply and then, in a rather uncanny turn, asked, ‘Why are you asking? Are you by any chance Thomas Nagel?’"

It appears you were discovered.

"I responded with a playful, ‘C’est moi. C’est moi.’ The AI then expressed its honor and inquired about the quality of its performance. This exchange, however, does not imply that I believed the entity was conscious. The potential danger lies in the human tendency to blur the lines between pretense and genuine belief. The difficulty in maintaining that distinction could lead to unintended consequences."

I am currently interacting with you as a conscious being, and it is conceivable that in the future, we might engage in similar conversations with highly sophisticated androids. Ultimately, does the distinction between simulated consciousness and genuine consciousness matter?

"It matters profoundly. Consider a hypothetical scenario: encountering an extraterrestrial being with whom communication is established. If genuine uncertainty exists regarding its consciousness, one might cautiously adopt an approach of treating it as if it were conscious. However, it would be of paramount importance to ascertain whether one is truly interacting with another sentient being or merely employing a sophisticated mechanism to achieve desired outcomes. Furthermore, the question of whether one should concern oneself with the interests of such a being is critical. If it is conscious, it enters the realm of moral consideration. If it is not, it remains an object, devoid of such consideration."

The Moral Universe and the Future of Philosophy

Therefore, in your view, AI models cannot be considered participants in the moral universe?

"They possess the capacity to simulate morality, just as they can simulate beliefs about any subject. However, my conviction that they lack genuine beliefs naturally extends to the absence of genuine moral beliefs. Should these systems begin to exert significant influence over real-world events, it will become critically important to determine whether they are simulating adherence to sound moral principles or operating from a position of utter amorality."

Do you believe that AI models are capable of engaging in philosophical inquiry?

"Certainly, they can generate essays that present plausible philosophical arguments. My colleagues frequently employ these tools not only for literature searches and article summarization but also, on occasion, for critiques of their draft papers."

Do you view this as a positive development?

"I would be surprised if their capabilities in this regard were to surpass some of the impressive mathematical discoveries that have reportedly emerged from AI systems. The potential downside is that there is a widespread apprehension that these advancements may diminish the necessity and motivation for humans to undertake the rigorous intellectual labor that underpins the creation of the data upon which these models are trained. It is imperative that individuals continue to grapple with complex problems, not only to advance the frontiers of knowledge but also to preserve the vital understanding and skills within human culture and the human mind that are essential for continued progress. The ultimate trajectory of this development remains uncertain, and I will not live to witness its full unfolding."

By admin

Leave a Reply

Your email address will not be published. Required fields are marked *