Read on Substack (canonical source).

A few recent conversations have me thinking about an old idea that sounds banal but is rarely applied. A discipline you teach quietly stops being yours the moment you stop operating it. The shift is invisible from the inside, because the vocabulary keeps working, the concepts keep getting invoked, and the inner sense of still being a professor of X does not change. Only the content behind the word changes. That is what makes self-observation so difficult: there is no internal alarm to warn you. Prof. Humberto Maturana built his epistemology around this point. The ontology of the observer, in its serious form, says that the observer is not outside the system being observed. Every distinction one makes also constitutes the observer. There is no balcony from which to view the world, because one is in the street making distinctions from a particular position. The ethical consequence is demanding: if you want to operate inside this framework, the first thing your framework asks of you is to observe yourself observing. Self-observation is not optional, it is the constitutive act of the method. What strikes me is that this aligns almost point by point with what cognitive psychology has discovered along entirely different routes. Festinger’s theory of cognitive dissonance shows that humans tolerate internal contradictions poorly, but we usually resolve them by adjusting our beliefs to match our behavior rather than the other way around. Robert Trivers devoted a whole book, The Folly of Fools , to the idea that self-deception is adaptive: we lie to ourselves because lying to others is cheaper when we believe the lie. The capacity to observe oneself in the act of contradicting oneself is a costly skill, not a spontaneous one. Hardly anyone trains it, and the few who do stand out. Philosophy arrived at the same place much earlier. Socrates put it in one line: the unexamined life is not worth living. Marcus Aurelius wrote his Meditations as a daily exercise in self-examination. Husserl proposed the epoché, suspending one’s own assumptions in order to see what one is seeing. Dewey insisted that honest inquiry requires welcoming the discomfort of discovering one was wrong. Wittgenstein, in one of his most quoted lines, said that the limits of my language are the limits of my world. I would add that the limits of my self-observation are the limits of my understanding. Popular culture occasionally arrives at the same intuition by other routes too. Michael Jackson’s “Man in the Mirror,” back in circulation because of the recent biopic, puts it in one line: I’m starting with the man in the mirror, I’m asking him to change his ways . It is not quite the same point as Maturana’s, but it is the same instinct. The 1988 pop song and the Greek philosopher are pointing at almost the same place. Let me bring this into concrete territory. Computer science departments around the world are expanding their programs in cybersecurity and artificial intelligence, areas where the demand for trained people is real and growing. At the same time, the institutions teaching these subjects often face their own operational problems in exactly those areas: security incidents, opaque algorithmic decisions, fragile technological infrastructure. A simple reading, common enough on social media, says this is contradictory. How can you teach what you do not operate well? That reading is precisely the trap the ontology of the observer is meant to dismantle. It assumes that teaching cybersecurity presupposes having solved cybersecurity, which is only possible from a balcony that does not exist. The reality of cybersecurity and AI is the opposite. These are fields where problems are structurally open, where the attacker or the system itself keeps moving, where operation teaches as much as textbooks, and where an honest stance is only possible if one recognizes oneself inside the system one studies. Teaching cybersecurity from an institution that has cybersecurity problems is not hypocrisy if the institution takes its own problems as part of the shared object of study. What would be dishonest is teaching it as if we were above it. Pedagogy begins precisely when one includes one’s own university as a live case, observes oneself operating, and acknowledges publicly the distance between what one says and what one achieves. The same holds for artificial intelligence. Those of us teaching AI in universities have not solved AI. Nobody has. Companies that publish responsible AI principles while shipping reckless products are the extreme case of what happens when you teach without observing yourself. The reasonable alternative is to assume that we are inside the system we study, that our technical decisions are also ethical and political decisions, and that the only way to teach AI with integrity is to admit what we do not know while observing what we are doing. If I had to take one thought from all this, it would be that academic coherence does not consist in having solved the problems of the discipline one teaches. That would be impossible in any living field. Coherence consists in not using the discipline as an ornament while operating against it. In noticing yourself when you invoke one framework and operate inside another. In accepting the friction of peer review, of critical colleagues, of students who ask awkward questions, because that friction is the only thing that keeps us operational in the discipline we claim to teach. Without it, the discourse empties slowly and nobody notices, because the voice still sounds the same. The more problematic a field, the more reason to teach it. But also, the more reason to teach it from inside, without pretending one is above the problem one is describing. Maturana would have put it better, but that is the idea: the ontology of the observer begins with observing oneself.

Read the full post on Substack.