The Hidden Moment You Kneel Before AI

We judge others according to their actions, while we tend to judge ourselves according to our intentions. This is hardcore attribution theory—the brainchild of the Austrian psychologist Fritz Heider, as far as I can tell. In a nutshell: if someone cuts you brusquely in traffic your initial thought is “this dude’s an asshat” but, if you do it, it’s likely that your first thought will be “I was distracted” or maybe “it wasn’t my intention” or the classic “I was in a hurry”. While everybody is (justifiably) quarreling about lines in the sand regarding the switch of AI from useful tool to existential threat, few—if anyone at all—look at the precise moment when humans will surrender to AI, effectively hollowing themselves out, i.e. delivering the spoils of their nature to the machine god.  

Heider’s most captivating premise is this: every human that has ever existed is an amateur psychologist. Fellas go around life deciphering “why” someone else did something, trying to make sense of the social world around them—either by attributing behavior to internal factors (motives or character) or external factors (like situational pressure). While watching someone shout at the poor dude working the counter at McDonald’s, the two options inside a human head are “how rude” or “having a bad day”, but the logic is inconsistent, particularly in an us-versus-them line. 

This is what is called dispositional attribution, I think, i.e. blaming people’s nature. It’s the path of least resistance towards making sense of the world, which is the hunger for predictability inside all persons, mind you. Since a bloke knows his “inner story”, made of intentions and context, he will attribute every screw-up to stress, urgency, and what not—of course: You are inside your head, not other people’s head! And the rabbit hole goes deep. In philosophy, Kantian ethics zeroed in on intention as the heart of moral truth, i.e. actions are good if they’re propelled by good will. But a Sartrean avalanche of esoteric verbiage—cascading down the cliffs of coherence—might differ, accusing humans as hiding behind intentions to avoid facing the painful truth of their actions. And Nietzsche? Well, he would just be nasty and say that self-leniency is the morality of a slave. In theology, Matthew 7:1-5 delivers a hammer blow to these inner musings with an admonition to first look at one’s plank before the speck in others’ eye. St. Augustine grappled with the issue of sin being not only what you do but what you meant to do. And lawyers? Man, lawyers love serpentine discussions about intention—as I wrote elsewhere. You don’t get the same jail sentence for offing a chap intentionally than accidentally. And so on and so forth. 

Now, if I may, let me bring in a blender and try to make salsa with your thoughts. Do we treat the outputs of AI (specifically LLMs) like human actions or do we tweak attribution because it’s a machine? This may surprise you but the behavior of habitual users of the main parrots—such as ChatGPT, Claude, and Grok—are currently on the fence. Let’s waltz into this mosh pit of concealed jackassery. There are (at least) two interesting angles to this: actor-observer bias and some sort of uncanny valley vis-à-vis agency. 

Since AI does not self-reflect (yet), humans are always the observer in the actor-observer bias framework. Therefore, in theory, LLM users are expected to attribute outputs by their ChatGPT in a dispositional rather than situational manner. The problem is that AI does not have intent and as a consequence humans might not attribute motive the same way than they do to other humans. Instead, they would leap and attribute behavior to the programmers tweaking the application or to the environment, i.e. the internet from where the data was harvested. But there’s the issue of LLMs getting better and better at mimicking human language, causing users to treat them like people. Confess it, you’ve done it, for example saying “please” and “thank you” to a bloody machine. I wonder if you would do the same to your blender: “Hi blender, can you please make this smoothie?” Bonkers, right? But people do it. And they only snap back into reality when the AI glitches or hits its limits, and the human will go back to the “it’s the code” mode. A duality emerges, then. The user will perceive the LLM human-like when it’s impressive and revert to mechanical when it fails. Remember that in human-human situations, intent and emotion anchor judgments in a more consistent manner. 

No need to twirl my own mustache here but these are the very first stages of what I bet will become the true line in the sand in the "risks of AI” world. There are a few peripheral, early approaches to the vicinity of the subject. For example, the “computers are social actors” paradigm by Nass and Moon states that humans instinctively apply their unique social rules to technology even though it is not sentient. Another study on Ai-generated content bias (last year on arXiv, maybe?) showed that LLMs inherit social biases from training data but there’s some sort of 50/50 blame between the tech and its maker, which is uncommon with people. 

There’s nothing out there specifically about attribution theory pairing humans and AI. So, once I realized that, I did what any rigorous, responsible, and proper chimp of letters would do: go to social media and check out the high-minded and sophisticated contributions by shitposters. It’s a mixed bag. On one hand, there are countless comments humanizing LLMs, like “chat got sass” or “it responded too polite so it made me wonder”. That’s dispositional attribution and it suggests that, in the background, users are anthropomorphizing the application. However, when the LLM hallucinates or is biased, users consistently blame the developers, moaning about OpenAI, Meta, xAI, or Anthropic infusing their applications with slop, wokeism, conservatism, racism, and whatever –ism the user wishes to whine about any given Sunday. It’s fair to say, though, that there is also a sizable crowd of netizens resisting human-like attribution, insisting that LLMs lack agency and it’s just a parrot, a probabilistic device connecting words to words. 

That is, my courageous and unscientific dive into the dark waters of social media reinforces what was suggested before, i.e. users perceive their chat app as nearly human when everything goes well but retrieve into the “just a machine” zone when the so-called shoggoth blunders. There is an undeniable separation between those who are awestruck and the doomsayers. It’s a mess. People don’t know what to blame. 

So, what’s all the fuzz about? Here it is, wrapped up with a bow: intent. Humans sway in its absence. I do not have a prediction but strongly suggest that people currently putting all their attention into purely mathematical metrics start keeping an eye on how AI users judge its outputs. Will humans go out one day into the cybernetic forest submerged in an inner battle to decipher the actions of machines in a dispositional way, i.e. blaming their nature? That, more than processing power or guessing how many r’s are in strawberry (look it up), is the line in the sand. Thirsting for predictability, humans may break today’s ambivalence and end up attributing behavior to machines in the same way they do to another human. That’s the moment we’ll know that AGI has been achieved inside the human soul. 

I am not worried about Terminators roaming a devastated earth covered in debris while hunting the last remnant of underground-dwelling humans. I am concerned about humans willingly erasing themselves.