Erlebnis and Artificial Intelligence
Every so often I reread some of Gadamer’s Truth and Method to try to understand a little bit more of it. At one point he digs into the genealogy of the word and concept Erlebnis (an experience), particularly in Dilthey’s attempt to come up with a justification for the human sciences (Geisteswissenschaften, what we now mainly call the humanities). Gadamer traces the idea of Erlebnis as a particular concept from Goethe through the Romantics to Dilthey, where what is important is a sense of human experience. Gadamer’s and Dilthey’s main focus is on hermeneutics (interpretation) as an alternative methodology for the humanities (alternative to the method of the natural sciences which had become dominant during the 17th century).
For the Romantics, Erlebnis - human experience - became vital as a protection against the dehumanization of industrial capitalism then developing across Europe. Where science sought to dominate and control nature, and machinery sought (as Marx wrote) to devalue and replace human labour, all areas where the human experience of work were under threat: artistic and intellectual work just as much as artisanal and agricultural work. By focusing on the scientific or productive outcomes, capitalism had no place for any concept of human experience. Indeed, human experience - the experience of childhood, for example - was wiped out as factories and their logic took over the developed economies.
It seems to me that the the concept of Erlebnis - experience - is precisely what is lacking from any discussion of “artificial intelligence”. I have written before about how new technologies like Large Language Models can only be considered a threat to academic work if such work is focused solely on outputs rather than the process learning and writing. The concept of Erlebnis is what was missing from my argument. If a student uses ChatGPT to write an essay, the student has missed out on the Erlebnis - the experience - of writing it. We can avoid the pitfalls of ChatGPT by restoring Erlebnis to the pedagogical realm.
But more than that, we will only be fooled into thinking that an LLM is in any sense intelligence or smart (let alone “smarter”) if we completely disregard the concept of Erlebnis. We will not be fooled if we adhere to a position something like the following: intelligence is not a question of outcomes or behaviours, but of experience. Intelligence is something experienced, first of all, by the intelligent being. A ChatBot cannot be intelligent because it cannot experience intelligence.
In a recent video on AI, Sabine Hossenfelder argued that she thinks AIs can understand because they use language. But here again, the concept of Erlebnis helps avoid a pitfall: a ChatBot does not - cannot - use language because it has no experience of language use. It cannot, in any sense, experience language. By the same token it cannot understand because it cannot experience understanding.
If we want to bring behaviourism back in - and here I think we connect with criticisms of the Turing Test - we can say, with Searle in his Chinese Room, that intelligence and understanding cannot be behaviourally limited to the exchange of texts (of singular genres of outputs). What we look for in our human interlocutors is evidence of experience: eye contact, body language, tone of voice, stumbles, errors, slips of the tongue. Surely various technologies may be able to mimic those things - especially once they are embodied as robots - but as long as we bear in mind that machines cannot experience; that Erlebnis is, as yet, completely separate from all that they are, then we won’t be fooled by claims to current artificial intelligence capabilities, still less AGI or the singularity.
Of course, the reason Erlebnis is never mentioned in AI discussions is because the current technologies are - like all the technologies developed since the 18th century, capitalist technologies. At best they are designed to ignore questions of experience; at worst they are designed to crush it. Experience is as foreign to military-industrial-technological discourse as it is to natural science which is its helpmeet. Only the humanities - as Dilthey and Gadamer insisted - care at all about human experience. That’s what they study, that’s what they want the world to be full of. We could almost say - and I think this idea would not be foreign to the young Marx of the Paris manuscripts - that human experience is the mortal enemy of capitalism.
Do we need to worry about some future where machines can experience? I suppose so, but only speculatively. There is no indication, given any of the current technologies on offer, that Erlebnis is even a remote possibility. And without Erlebnis, as the Romantics maintained, there can be no human aptitudes, skills, capacities, desires, or functioning. Erlebnis, experience, can and should become the watchword of a critical distancing from the hype surrounding “artificial intelligence” as well as a broader anti-capitalist “structure of feeling”.