Conscious realities

Apple’s Machine Learning Research group recently released a paper that outlines the current “reasoning” capabilities of Artificial Intelligence (AI). The details of the study are fairly technical from a layman’s perspective, but the results are clear: AI isn’t nearly as close to actual “intelligence” as many had hoped or feared.

The Apple study examined several state-of-the-art Large Reasoning Models (LRMs), a form of AI that purports to be an advance over the Large Language Models (LLMs) that power chatbots such as OpenAI’s widely used ChatGPT. In short, the study found that LRMs demonstrate an advantage over LLMs in certain medium-complexity tasks, but both LRMs and LLMs experience a “complete collapse” when faced with more complex tasks. As AI content creator and entrepreneur Ruben Hassid summarized the findings, even when the researchers gave the LRMs the exact solution algorithm to certain problems, the models’ accuracy eventually reduced to 0% on high-complexity tasks.

AI presents a series of societal challenges that few have the expertise or the will to address. Entire industries fear job displacement, and many worry about the restructuring of the global economic order. Governments have been reluctant to provide any regulatory guardrails. The detailed AI 2027 report outlines a possible scenario that has provoked fear and stoked controversy. Higher education is facing a crisis of rampant AI cheating. The newly elected Pope Leo XIV has described his choice of a papal name in terms of the “industrial revolution” caused by AI (Leo XIII defended the rights of workers during the original industrial revolution). Some are even openly warning of species extinction if we don’t shut the whole thing down.

But perhaps the most pressing challenge posed by AI is the one the Apple study highlights: the uniqueness and dignity of human beings as reasoning creatures. Both enthusiasts and detractors of AI are concerned with the question of consciousness: Can AI ever achieve the ability to think, to self-reflect, to reason, to intend? Some AI developers are quite certain that AI consciousness is possible (and perhaps already realized), but they can only arrive at such a conclusion by series of disastrously mistaken metaphysical assumptions. They assume, first of all, that consciousness is simply an emergent property that supervenes upon vast amounts of data and computational powers. They further assume that this is actually how consciousness works in human beings as well. We are simply highly advanced brains that somehow, as an accident of evolution, achieved the ability to think about thinking, and to have a first-person awareness of the world. Having reduced humans to machines, it is a small step to assume that machines can become humans, or at least humanlike in their reasoning capacities.

But there is an infinite qualitative chasm between the machinery of computation and language prediction, on the one hand, and actual consciousness, on the other. Mind is not reducible to brain, even if we grant that the two are fully integrated (as all of the classical philosophical positions on the mind-body problem readily acknowledge). But mind transcends neurobiology precisely by its first-person experience of the self and the world. Even if you could exhaustively map the neurological activity in my brain as I sip my morning coffee, it still wouldn’t give you access to my first-person experience of the flavor notes in the bean.

Something is always left out in purely material descriptions of qualia—our conscious awareness of the phenomena of experience. Of course, Christians have always recognized this. We believe in an immaterial soul of some sort (though theologians have cashed out the precise relationship between soul and body in different ways). But even many non-theistic philosophers recognize that mental states cannot be easily reduced to physical states. Consciousness remains a mystery, but it is a mystery that all of us experience in the simplicity of our own being in the world. As theologian David Bentley Hart has articulated it, “Electrochemical events are not thoughts, even when they may be inseparably associated with thoughts, and no empirical inventory of such events will ever disclose for us either the content or the experiential quality of an idea, a desire, a volition, or any other mental event.”

In the end, the Apple study reveals what we already knew: machines can’t think. They do not—indeed, they cannot—possess reason, self-awareness, and intentionality. Unique among the material substances in the world, only the human being can think, intend, desire, and reason. Theologically, this is tied to our personhood—our being a “who” and not just a “what,” those made for union and communion with the divine persons. The most AI can do is produce a highly advanced simulacrum of these things—one that, to date, has failed to live up to the billing. None of this means that AI is harmless. It poses real promises and perils for humanity. But considering the nature of consciousness in philosophical and theological perspective should temper our fears that AI could ever displace the dignity and beauty of human existence.