Theme: Psychology

I love bandwagons. I’m present at most crazes, sometimes too late, but I don’t want to miss anything. In 2000, at the peak of the internet boom, I went to work at Lost Boys, one of the most high-profile internet firms of the era. They were so good at branding that the company seemed omnipresent. Bikes bearing large advertisements for the firm could be seen everywhere. Incidentally, that generated some confusion – for a long time, my sister thought I worked at a bike rental company.

Lost Boys made what they called ‘digital interfaces’. At first, they produced beautiful CD-ROMs for museums, but progressed to building websites which after some time, got much broader functionalities. These started as a sort of digital brochure but grew into the digital interfaces we recognize in airline booking systems or online banking systems.

There is an unexpected link between the Lost Boys’ interfaces and the German philosopher Martin Heidegger. His 1927 book, Being and Time (Sein und Zeit), may well be one of the strangest works in the history of philosophy. It contains sentences like: Das Gefragte der auszuarbeitenden Frage ist das Sein, das, was Seiendes als Seiendes bestimmt, das, woraufhin Seiendes, mag es wie immer erörtert werden, je schon verstanden ist. Das Sein des Seienden ‘ist’ niet selbst ein Seiendes. (Which was translated by Macquarrie and Robinson in 1962 to: ‘In the question which we are to work out, what is asked about is Being – that which determines entities as entities, that on the basis of which [woraufhin] entities are already understood, however, we may discuss them in detail. The Being of entities “is” not itself an entity.’)

I never believed I’d be doing humanity a favor by tormenting myself over what ‘being’ actually is. So I wrote off Heidegger in the early years of my studies. But later, when I went on to immerse myself in how computers work, I began to appreciate certain of his ideas. I came across one of them in practice, at Lost Boys, in the ‘usability’ department. There, they worked on the user experience of digital interfaces. What works, what doesn’t?

Heidegger has an interesting set of ideas for tools: Zuhandenheit and Vorhandenheit. If we use a tool and it works well, we no longer notice it. The tool becomes transparent, an extension of our body and our intentions, allowing us to concentrate on the task rather than on the instrument we’re using. That’s Zuhandenheit (Ready-to-hand).

Conversely, if that tool malfunctions or becomes unavailable, it becomes Vorhanden (Present-at-hand) and visible. Frustration draws our attention to the object itself, and we become aware of it. Let’s say you’re driving in a car. But then you hear a scraping sound when you change gear and you can’t get it into fourth. That makes you aware of the transmission, and the car. The object, as it were, suddenly appears into view. That’s Vorhandenheit.

Now, an interface has to function well in both its Vorhandenheit and its Zuhandenheit. But the two are not the same. If you ask people to describe how they work with something, they rarely describe the functionality (i.e. the Zuhandenheit), instead, they describe the object (its Vorhandenheit). So such research is not easy. It’s easier to study how an object works if it’s malfunctioning and Vorhanden, than when it’s Zuhanden. It’s impossible to put into words since if it functions well, you don’t note or notice much.

Irrespective of practical implications about how we deal with interfaces or tools, and what does and doesn’t work, there is a broader implication in Heidegger’s ideas – and it’s relevant to artificial intelligence. I’ll condense things, to avoid making this too complex. If you look at Large Language Models like ChatGPT, you’ll know that they fabricate answers in response to prompts, based on the knowledge they already have. It’s reasonably straightforward to incorporate data of Vorhandenheit, malfunctioning tools, and their objective workings, into such a system, because you can easily describe them with language.

That doesn’t apply to the Zuhandenheit, which concerns the subjective use of tools in an undefined context. With this implicit understanding of how to use a tool, a Large Language Model system would not know where to begin or where to end. The Zusammenbruch (breakdown), as Heidegger calls it, we experience once an object goes from from Zuhanden to Vorhanden indicates that we fall back from a completely different world, back to a world where linguistic representations can work.

For those interested and to iron out any misunderstandings, I’d like to say just one more thing on this topic. This does not mean that AI cannot deal with such problems. But AI that processes and generates language would need to be integrated with a form of ‘artificial life’ that can mimic human consciousness and nuances, to achieve a meaningful interaction with our environment. POTUS’s first buddy Musk seems to understand this, given his focus on the fast roll out of robotics. That is where the future of AI will be.