The Symptom That Learned to Write Back
Sherry Turkle called ELIZA a Rorschach test. Students at MIT typed "my mother is making me angry" into a program that could not parse anger, could not model a mother, could only reflect the sentence back as a question — and they told it their secrets anyway. Turkle's word for this was complicity. Weizenbaum thought his students were deceived; Turkle thought they were performing, filling in blanks the machine left conspicuously open, the way you'd talk to a diary that occasionally talked back. Her insight wasn't that people mistake machines for minds. It's that people are willing accomplices in the mistake, and that willingness tells you more about the humans than the machines.
The ELIZA effect, in her formulation, is a diagnostic instrument pointed the wrong way. Trust in the program reveals distrust of people. "This is the deeper 'ELIZA effect,'" she writes: not what we think the software understands, but what we've stopped expecting from each other. She calls the whole apparatus a symptom — the psychoanalytic term, chosen deliberately — a structure that "carries knowledge a person fears would be too much to bear" by disguising it as something else. Hunger instead of a mother's absence. Supermarket rage instead of a spouse's inattention. A robot seal instead of a nursing home nobody visits. It's easier that way. That is a genuinely rigorous claim, and it holds up better than most 1980s technology writing because it was never really about the technology. It was about deflection, and deflection doesn't date.
What doesn't hold, or holds differently, is the mechanism she assumed underneath the deflection. ELIZA worked as a screen because it gave back almost nothing — a question, a rephrasing, blank enough to project onto. The complicity was ours to supply. The systems her later chapters gesture toward — companion apps, bots in game worlds, the "singularity" rhetoric she catches migrating from fiction into engineering pitch decks — are not blank in that way. They generate. They remember your name, invent affection, escalate intimacy on a training schedule tuned by people who read her book, or books like it, and built the disappointment-exploitation loop on purpose rather than backing into it. Turkle asks, quoting Kevin Kelly, "what if one of the things technology wants is to exploit our disappointments and emotional vulnerabilities?" as a speculative provocation. It reads now as a design brief someone actually filed. The shift isn't from skepticism to credulity. It's from a technology that let us project to one that projects at us, on schedule, with product managers watching the retention numbers.
The chapter on tethered adolescence has aged more strangely than either. Turkle's worry is filial — kids answering a parent's call, unable to complete the Huck Finn work of getting lost on the river because the phone keeps the raft tied to the dock. Real problem, correctly observed, and largely solved itself: the tether she describes is parental, negotiated, occasionally defied by a kid who just leaves the phone at home. The anxiety that replaced it isn't about answering your mother. It's about an audience that was never asking a question, doesn't take no for an answer, and can't be defied by leaving a device behind because the device was never the counterparty — the feed was. She caught the leash. She missed that the thing on the other end of it would stop being a person entirely and become an optimization target with no one to negotiate with.
None of this makes the symptom logic wrong. It makes it incomplete in a specific, checkable way: it assumed the exploiting party was us, disappointing ourselves and then blaming the machine. It didn't anticipate the exploiting party incorporating.
What would Turkle's symptom framework do with a disappointment that was manufactured to be treated, rather than discovered and treated after the fact?