Saying AI is Intelligent is Parasocial. Reading Sartre on Bergson

The image, in Sartre’s criticism of classical philosophy, is a thing. It is given or has certain properties, and from there, we build our systems of thought. In reaction to, in acceptance of, in denial or overruling of, those images and their properties. He spends the first chapter of The Imagination on Descartes, Leibniz and Hume. He ends the chapter reinforcing the problem of the thingness of the image.

In chapter two he spends many pages unpacking Henri Bergson’s approach to the thing of the image. In Sartre’s framing Bergson uses a temporal metaphor for the image. Rather than saying the most relevant fact of the image is its spatiality, distance, extension, foreground or aft of consciousness, perception, thought, Bergson, in Sartre’s telling, addresses its place in time. It arrives now, is stored for later and becomes present again in the future: pulling the past into the present. Creating the present consciousness, the impression and activation of self, in the now. Building the capability for future being, in the moment. Projected.

The mechanics of imputing intelligence and selfhood to AI feel similar. There is input, to which we ascribe thingness. From that thingness of the input, returned to us in new forms, new configurations, we impute being. We read what we’ve fed it, in different shapes, in broader range than what any single human can remember or hold, and impute a selfhood to it. We forecast an evolving consciousness, a growing being, and fear what comes next. We are anthropomorphizing AI.

The narrower the application, the tighter the context, the less remarkable AI can feel. It is computationally fast. It is search and retrieval, archivially, fast. But it is more likely to be wrong or non-useful. It often goes down rabbit holes or seeks to widen when it shouldn’t. The probability functions guiding the shape of its output slip gears, unnecessarily seek, weigh and output other responses. Clearly show itself a tool. As a user I project forward improvements to the tool, future utility. Not future beinghood.

There may be something to the Bergsonian phases of consciousness (as described by Sartre) and schema. Some metaphoric application to AI’s function and our use of it. There are characteristics of the input. The circuitry flattens them. In the space between thing input and thing output, in the circuitry and nervous system, can generalities arise, universals held by the machine itself to be the case, in all applicable cases?

AI has been trained on our language and built to output in our language. There is a machine, circuitry and programming in the middle. It is connected to a McLuhanian nervous system of all digital data. But the output and the patterns are not intelligence. They are not selfhood or consciousness. We impute meaning to it, often based on our use cases, the instrumentalist application of the tool. Like social media or a music video we can project closeness, connection, she’s singing to me. We are somehow connected. People who like these posts are my friends, are trustworthy, competent.

Ascribing intelligence to AI is parasocial anthropomorphizing.

EXCAVATION
The Imagination (end of chapter 1, chapter 2) by Jean-Paul Sartre, 2012 Routledge translation by Kenneth Williford and David Rudrauf