What Would Kant Think of AI?

Volume 25

Dec 12

“Experience without theory is blind, but theory without experience is mere intellectual play.”

Immanuel Kant is seen as the intellectual godfather of cognitive science, and next year we will celebrate the German philosopher’s 300th birthday. However, with the genesis of artificial intelligence and the idea of the so-called deep neural network structures, it’s become apparent that there are now overlaps between classical philosophies and modern technology. Kantian doctrine underscores that our capacity for reason and sensory experiences are integral to shaping our comprehension of the world, reminding us that even as we draw parallels with AI gaining sentience, the depth of human understanding is rooted in this synergy. But because humans are responsible for the programming of such algorithms using their own biases, these algorithms will continue to be as limited as their creators.

Back in 2015, when I was involved in AI, the technology was still in its infancy. Much like nurturing a hatchling, each piece of information helped it grow stronger and more capable. Eventually it transformed from a tiny creature into a hulking albatross, and it was teaching me what I required without the need for a prompt. It was providing a chorus of life-and-death information at lightning speed, pinpointing potential attacks around the globe — both in the physical realm and the digital expanse of cyberspace. Hence it was my first direct contact with human-in-the-loop feedback, the symbiotic partnership between person and machine. It was far from perfect given the need for constant correction.

Like these imperfect devices, Kant was also not without his faults. His steadfast objectivism, coupled with the progression of secular thought, made his theistic account of a morally reasoning immortal soul appear somewhat outdated to many. Nevertheless, many have been turning to his seminal work “Critique of Pure Reason” as a way to understand so-called “self-generated brain activity.” This includes the Ruhr-University Bochum Philosophy Professor Tobias Schlicht, who talks about how AI's feasibility is often functionalism, which views mental states in terms of their functional roles and causal relations within a system, suggesting that mental processes are multiply realizable and not only restricted to organic brains.

As written in “Kant and Artificial Intelligence,” edited by doctoral researcher Hyeongjoo Kim and philosopher Dieter Schönecker, some suspect that Kant's neutrality on the mind's substance could potentially be replicated in various scenarios, not only biological ones. This begs the question of whether our minds just consist of ones and zeros in order to make choices.

In both the “Jäsche Logik” and the “Critique of Pure Reason,” Kant believes that cognition is not just about having representations — it also requires real consciousness. The Enlightenment thinker articulates a spectrum where simple perceptions or representations evolve into full-blown awareness when they are accompanied by consciousness, distinguishing between mere sensations and objective cognition.

In the context of current AI, Kant's framework would suggest that for a machine to truly cognize in the narrow sense, it would need to be aware of its own existence. However, this requirement in cognition is not always shared among modern scientists or those working in AI. They often view AI systems as capable of reason in a much broader sense, where the apparatus can still process information and make decisions without the need for what we deem as human consciousness.

With the rise of deep learning architectures in AI research, we see new questions presented surrounding perception, learning, and understanding. These systems, however, which have contributed to significant advancements in machine intelligence, operate without the ability to be self-aware, challenging traditional notions of reasoning. They align more with an empiricist approach, learning from vast amounts of data, as opposed to a rationalist approach that might suggest the necessity of an underlying conscious understanding.

Speaking to Dr. Naomi S. Baron about our Franken-creation, the linguistic expert questioned whether we actually know what it means to emote. The American University professor emerita of linguistics told me that we see the symptoms of it, we can probably describe what it looks like, and chances are “we will be able to program in the parameters of what we describe linguistically as sentience.” Dr. Baron wrote the 2023 book “Who Are We? How AI and the Lure of Efficiency Threaten Human Writing,” which describes the evolution of language as a result of machine learning softwares. She noted that in the 1970s and 1980s, sign language was being taught to non-human primates such as the bonobo monkey — is communication with a non-human but sentient being still considered a language? That’s the camp that AI falls under, and just as Kant asks people to rationalize their thinking, perhaps Kantian concepts would want us to consider AI equally as capable.

However, a side effect of AI is that it stops many people from thinking for themselves, which defeats the purpose of moral appraisal and moral responsibility. When testing out the AI companion reading software, YouAI, on Plato’s “Republic,” it provided generic responses without the need to read the book myself. Additionally, as it processes the text, it also interweaves elements of your own personality into the narrative, creating a customized interpretation. Evidently, it may help you seem profoundly wise, except for the fact that you haven’t actually read the book. As a result, Dr. Baron says by allowing AI to do all the work, peoples’ ability to think critically for themselves is “being eaten away.”

One saving grace is that our anxiety about machines making decisions for themselves rather than following the instructions of their creators is misplaced. Despite the fact that Kant’s philosophy could apply to AI in some instances, there is a misconception that, just like a machine, a human being’s intelligence amounts to following a series of instructions, one so complex that it creates the illusion of autonomy. But long before the advent of computing machines, Kant realized that human cognition could not be reduced to following instructions, no matter how complex. He recognized that a creature whose entire mental capacity boiled down to executing a series of directives would inherently be unable to differentiate between those directives and the actuality of the external world. To this entity, its entire understanding would be synonymous with its environment, one indistinguishable from the other. This notion of obeying commands starkly opposes our actual lived experiences, as Kant observed.

Kant views memory as a crucial cognitive faculty that encompasses two core processes: the storage and retrieval of information. These processes are significantly influenced by our experiences, and reflect the vital role of memory in shaping human thought. Essentially, our ability to retain and recall information is often tied to how compelling the subject matter is to us personally. The more captivated we are by a topic, the more effortlessly we can commit it to memory and subsequently access it. The memory often transforms into a bitesize version in order for us to store it — one that isn’t always accurate.

Consider a cherished moment in your childhood. While the desire to experience it again is strong, the nature of memory doesn't allow for an exact re-experience. If it were possible to relive it precisely as it occurred, it would strip the moment of its status as a memory. Memory functions as a repository of information, but it's the contrast with lived experience that gives it significance. Without this contrast, there would be no way to distinguish whether you're encountering the moment for the first time or the hundredth. Memory, then, is not a mere replay of events; it's a conscious recollection that informs and enhances our lived reality.

This enduring feature of our cognition, the ability to experience the world sensually, sets us apart from machines and any other entities that might process information without sensory perception. This intrinsic trait of human existence — that we are beings who feel and utilize information through our bodies — underpins our unique capacity for decision-making. Unlike machines that operate purely on programmed directives, our choices are shaped by a rich hodgepodge of sensory experiences and embodied interactions with the world.

Kant implies that the path to true autonomous decision-making for machines does not lie in the escalation of complex algorithms or the extensive analysis of data to predict subsequent language patterns. Essentially, if machines ever attain a state where they can independently make choices, they will be fundamentally different from the generative AI systems we are familiar with today.

But we are Victor Frankenstein in Mary Shelley’s classic novel. Kant argued that the only world we can know is the world created by the innate structure of our minds, and thus reality “as it is in itself” is unknowable. We are responsible for the architecture of this reality within AI’s construct, and thus an algorithm selecting a most-likely next word or a program calculating the best move in a game of chess isn’t necessarily ‘choosing.’ Since its 'knowledge' is indistinguishable from its 'reality', it hasn’t chosen because it's not exploring options, rather it’s executing pre-determined pathways.

While scientists hail Kantian philosophy to explain away its ability to become self-aware, he also leaves us with this: “All false art, all vain wisdom, lasts its time but finally destroys itself, and its highest culture is also the epoch of its decay.” Despite its impressive capabilities, AI still operates within the confines of its human programming, and lacks the rich sensory and conscious experience that define ‘intelligence’ — at least for this iteration.

What Would Kant Think of AI?

Stravinsky at the End

Broken World