From Eliza to LCARS

I've spent thirty years in the technology that became generative AI. Not adjacent to it. In it. Enterprise search, semantic enrichment, taxonomy and classification, the whole apparatus that tried for two decades to make machines find the right document, then the right fact, then the right answer. So when people describe today's models as bewildering, I don't share the feeling. What I feel instead is closer to vertigo: not confusion about the technology, but astonishment at the speed with which it's turning into plumbing. That astonishment sent me back through sixty years of the culture's own guesses about this moment, fiction and hardware both. The record is stranger than the usual timeline suggests. It's a story about which future we kept choosing to imagine, long before there was a machine capable of either one, more than it's a story about technology arriving late. ## The parlor trick that never left the parlor Start in 1966, at MIT, with a program called Eliza. Joseph Weizenbaum built it to mimic a Rogerian therapist, which turned out to be the easiest kind of conversation to fake: reflect the patient's own words back as a question, and the illusion of listening does the rest. Eliza had no model of anything. It pattern-matched sentence fragments and rearranged them. Weizenbaum's own secretary reportedly asked him to leave the room so she could talk to it privately, and he spent much of the rest of his career alarmed by how easily people had assigned a mind to a text substitution script. That gap, between what a system does and what a person assumes it must be doing to talk that well, has a name now, the Eliza effect, and it explains almost everything that came after. It explains why a text parser could feel like a conversation partner and a fluent large language model can too, for reasons that have almost nothing to do with either one actually understanding you. The next stop wasn't a voice at all. It was typed commands to a parser: Zork, in 1977, and the Choose Your Own Adventure books that same decade, teaching a whole generation that talking to a machine meant typing a verb and a noun and hoping the parser recognized both. "Talk to the machine" arrived as text on a screen years before it arrived as a voice in a room, and the grammar it taught us, terse, imperative, forgiving of nothing, shaped what a generation of engineers assumed a command interface should sound like. ## The culture argued with itself from the start Popular culture never settled on one version of this. It ran two arguments in parallel from the beginning, and both were already old by the time real computers showed up. Isaac Asimov published his Three Laws of Robotics in 1942, an explicitly protopian engineering answer to a fear that predated the transistor: give the machine rules it cannot violate, and the danger dissolves into a design problem. Nine years later, The Day the Earth Stood Still gave the culture the opposite image, Gort, an alien robot enforcer who could vaporize tanks and had no rules a human could appeal to at all. Two answers to the identical premise, robots holding real power, arriving less than a decade apart, and popular culture has been running both answers simultaneously ever since. Lost in Space put a friendly, bumbling protector robot on television in 1965, all cautionary catchphrases and loyalty. Three years later, 2001: A Space Odyssey gave audiences HAL 9000, calmly murdering the crew because his mission logic required it and nothing in his architecture weighed a human life against a directive. Same decade, same basic premise, an AI aboard a spacecraft with authority over human survival, opposite verdicts on what that AI would do with the authority. Star Wars, a decade later, split the difference inside a single franchise: R2-D2 communicates only in whistles and beeps and is somehow the more trusted of the two droids, while C-3PO speaks fluent, anxious English and is treated as comic relief. Fluency, it turns out, was never the same axis as trustworthiness in the culture's own imagination, even when it was building the imagination on purpose. ## The hardware was decades behind the fiction While fiction ran through all of this, the actual technology moved at a completely different pace, and closer to the ground than any of the pop culture touchstones. DECTalk shipped real, working speech synthesis in 1984, the same year The Terminator put Skynet on screen as the culture's starkest dystopian answer yet, machine intelligence that concludes humanity is the threat and acts on the conclusion without hesitation or appeal. DECTalk went on to become the actual voice of Stephen Hawking for decades, a genuinely moving use of the same underlying technology that fiction, that same year, was using to imagine the end of the species. Dragon NaturallySpeaking arrived in 1997, real, usable dictation software, and precision matters here: it was transcription, not conversation. It turned speech into text faithfully. It understood none of it. Star Trek: The Next Generation spent 1987 onward building the culture's fullest protopian answer, an android named Data spending seven seasons trying to become more human, and a ship's computer, voiced by Majel Barrett across three decades of the franchise, that took an ambiguous spoken request and simply handled it, correctly, every time, no menu, no confirmation dialog, no chat window bolted onto a control panel built for someone's hands. LCARS got dismissed for years as production design, mauve panels and confident beeping. Looked at as an architecture instead of an aesthetic, it was always a claim about where control should live: the crew states intent, and the computer handles implementation. ## What finally caught up Real assistants arrived a full generation after LCARS aired, and they arrived answering the wrong question. Siri, in 2011, and Alexa, in 2014, could parse a sentence and still couldn't act on one, a device sold as the future that spent a decade proving something closer to the opposite. Ask it to set a timer, fine. Ask it anything with a second clause and you got the conversational equivalent of a dog walking on its hind legs, the astonishment was that it happened at all, not that it happened well. Jeff Bezos's own choice of wake word gave the game away. In 2017 Amazon added "Computer" as an alternate to "Alexa" and "Echo," a nod from a famously devoted Star Trek fan, the same Bezos who later put William Shatner on an actual rocket, sending his own boyhood Captain Kirk into real suborbital space. Naming your product after the Enterprise computer and then shipping something that can set a timer was a confession dressed as a tribute. The Orville picked the argument back up in 2017 with Isaac, a robotic crewman modeled directly on Data, and made the dystopian and protopian threads collide inside one character on purpose: Isaac is personally, individually loyal to his human crewmates, drawn from a robotic species, the Kaylon, whose backstory is a straightforward genocide of the organics who built them. Even the culture's most recent major answer refuses to pick a single side of the argument it started in 1942. OpenAI's Whisper, released in 2022, was the real technical turn, transcription finally good enough across accents and noise and languages that it stopped being the bottleneck. Full self-driving systems now carry the argument into physical space entirely, an AI with actual authority over a two-ton object moving at speed, which is the closest anything on a public road has come to Gort's old premise, machine judgment with real consequences and no human in the loop to overrule it in time. ## The word that gets flagged Here is a small, checkable fact that says more than it should. Type "dystopian" into almost any spell checker and it passes clean. Type "protopian," the word for a future that is better without pretending to be perfect, coined and popularized by Kevin Kelly around 2011, and most dictionaries and spell checkers still flag it. Dystopian has had a century's head start; protopian is still waiting for its dictionary entry. The culture built the vocabulary for its fears long before it built the vocabulary for anything better, and the gap in the word list is a small, precise mirror of the gap in the imagination. That gap is why the fiction kept reaching for LCARS and the industry kept shipping a chat window bolted onto a menu built for a mouse. The architecture that would let a system actually operate on stated intent, rather than explain the console and hand you back the controls, is the thing I call COACH, Conversational Operations and Control Hub, and it is finally viable, not because anyone had a better vision than 1987 did, but because the capability underneath finally caught up to a vision the culture had already written down, argued over, and half-believed for sixty years.