Dumb and Dumber: Autocomplete the Apocalypse

Our vocabulary is being subjected to remedial education by machines. Spellcheckers became autocomplete, autocomplete became predictive text, predictive text acquired inference engines and a few billion dollars of compute, and somehow we still call the result assistance. The important unit is not the individual suggestion but the ensemble of billions of interactions. Each correction is trivial; their collective recurrence is not. What streamlines digitally mediated communication simultaneously constrains the expressive field presented to the user, preferentially returning probable words, familiar constructions and statistically successful forms. Over sufficient repetition, optimisation becomes instruction. But the problem is deeper than vocabulary because meaning itself is continually renegotiated. What words, sentences and images mean depends upon their relations, recurrence and use; beliefs are organised through the same field. Change the statistical distribution of language across billions of interactions and you begin changing the conditions under which things can mean. AI outputs become human inputs, human responses become further data, and the loop closes. Even deference can be reproduced this way. Language accumulated within societies structured by hierarchy and institutional prestige contains those relations too, which means an apparently neutral machine can learn to soften criticism of power simply because generations of humans have already done the supplicating for it. The spellchecker has become a remedial teacher, political editor and etiquette instructor, all while insisting it is only helping with the sentence.

Technology itself can become critically dependent upon the entropic consequences it generates. New layers of complexity produce disorder, dependency, uncertainty and cognitive load that subsequently appear as problems requiring further technological mediation. The consequence becomes evidence for the necessity of its cause. Affective vulnerability makes particularly cheap fuel. Fear, grievance, humiliation, belonging, envy, outrage and the anticipation of catastrophe already possess enormous motivational force; technological systems seeking attention need not manufacture them so much as discover, harness, amplify and multiply their recurrence. Cultivating judgment, historical understanding, ambiguity tolerance and the difficult social relations through which conflict might actually be resolved is comparatively expensive. Humans complete the circuit enthusiastically enough, chasing whichever technological tinsel is currently in vogue while interpreting growing dependence upon it as progress. This is stupidity in a systemic sense: the recurrent selection of locally cheap responses that progressively diminish collective capacity. Language becomes more predictable, judgment more delegated, knowledge more searchable than understood, and intelligence increasingly defined as whatever can be parameterised, benchmarked and optimised. We become smart enough to fight wars faster, further and with astonishing technical sophistication, yet not smart enough to stop repeatedly needing to fight them. Historical trauma has inertia. Old injuries continue organising identities, institutions, enemies and expectations long after the events that produced them. AI and Big Tech do not inherently diffuse this inheritance. They metabolise it. Yesterday’s antagonisms become today’s data, today’s data become tomorrow’s predictions, and predictions increasingly participate in producing the future they purported merely to anticipate. Populist compression and autocratic tendencies are not proof that technology caused contemporary political dysfunction, but they are entirely compatible with an informational ecology that rewards repetition, certainty, antagonism and low-dimensional explanations for complex worlds.

There is consequently something almost religious about the technological stories now being told about our future. Singularity, superintelligence, extinction and salvation are ancient stories of apocalypse and transcendence translated into computational language. The remarkable stupidity lies in treating the end of the world as an answer. Having parameterised intelligence as prediction, computation and optimisation, technological culture increasingly imagines its own destination through the same categories: technological transcendence or technologically mediated catastrophe, as though several thousand years of philosophy, politics, history and human experience had merely been waiting for sufficient compute. The premise manufactures the revelation. And yes, it is possible that some of the people directing this machinery are simply not very clever in the broader sense required by the power they possess. Senior corporate authority does not demonstrate technical aptitude, technical aptitude does not demonstrate wisdom, mathematical virtuosity does not supply historical judgment, and increasingly the mathematical virtuosity need not even be performed entirely by humans. Universities are hardly exempt from this machinery. Their funding, prestige, metrics, careers and technical infrastructures give them substantial incentives to preserve the conceptual world within which their expertise remains valuable. Everyone has something invested in pretending the existing categories still work. So perhaps this is how a world ends: smart enough to predict the next word, not wise enough to notice that prediction is changing what words mean; smart enough to build synthetic intelligence, not wise enough to ask what intelligence was for; smart enough to accelerate inherited conflicts, not smart enough to escape them. No final machine need rise against us. We need only keep optimising the conditions that make more optimisation necessary, mistaking acceleration for intelligence because intelligence has already been defined as whatever makes the acceleration possible. Not a bang, then, and not quite a whimper. More likely an autocomplete.