By: INDRAYUDH SINHA
“If we want things to stay as they are, things will have to change,” says the antihero Tancredi, the nephew of Don Fabrizio, the eponymous “Leopard” of Lampedusa’s novel. With Garibaldi at shore, the Sicilian prince’s fate hangs in midair, uncertain in front of the abrupt force of revolution. We might be at a similar precipice in our time, and one not any less dramatic. The force of automation has always loomed over our heads. In our society presently, there are already significant degrees of automation: the electric switch and bulb, the lamps in the street, ceiling fans, word processors and so much more. Automation has made the modern world.
The historian Carl Benedikt Frey wrote in his book “The Technology Trap” about New York’s class of lamplighters, men who were tasked with lighting lamps, every single one of them, every evening. This mere fact might surprise some readers of this Op-ed, that there was a time when lamps needed to be lit individually, and thousands of men earned a living as lamplighters. Likewise, in Colonial India, a class of “punkah puller” natives were employed to manually fan their European employers all day long. In 1907, a large section of the lamplighters in New York went on a strike, resisting their incoming displacement by the tiny article of the bulb. Yet the lamplighters and the punkah pullers, like very many other professions throughout history, couldn’t resist the wave of technological automation through strikes or legal remedies. This time wouldn’t be different.
The economic historian Joel Mokyr, who was an Upton Scholar here at Beloit, splits technological innovation into two categories: macro and micro innovations. The micro developments in linguistics and cognitive science gave rise to the macro innovation of large language models. LLMs themselves have spurred an even bigger development in terms of how textual data is processed. There are no absolutes in terms of what counts as macro or micro. Technological innovations are mostly deemed consequential in terms of their impact on the economy. The microscope is certainly more consequential to the economy, but without the lens maker it simply wouldn’t exist.
Most new technologies seek to align themselves with some precedent (one can find traces of the military funded DARPAnet in the contemporary internet) while some technologies establish entirely false precedents within their functions. Anybody who has used social platform apps such as Instagram has noticed that abominable traffic of short videos which the app neatly terms “reels”; such a formulation is meant to evoke familiarity for users, and psychologically make the experience seem coherent like a film reel. But reels on Instagram or Facebook are what might be called false precedents, when a new technology claims ancient ancestry.
But it is these false precedents in new technologies which are interesting. Many of these false elements in present technologies are solely meant to humanize the application. ChatGPT, Claude, Perplexity, the whole bunch of LLMS, don masks of communicability, which have been aptly falsified by critics decrying their “functionalism,” as in just because these models can externally emulate human communication, doesn’t mean that they work like humans internally. These brackets of human interaction within contemporary technologies sets it up for what might be termed “petty usage.”
Technological development throughout history has been fueled by large imperatives, but their petty usages have often derailed them from their intended application. Gunpowder, which was used to make weapons like fire arrows in China, also found usage in making fireworks for court entertainment (but also later arms and ammunition). Similarly, contemporary LLMS came out of the needs of harvesting “big data,” but now finds itself entangled in the lives of regular people due to its increasing petty usage.
But there might be a reason why founders and owners of technology built for large, consequential usage are interested in sustaining their petty usages. What can be observed is that petty usage creates a market for the technology. It’s not as many companies or professionals paying for LLMS, but mostly petty users. Students like us are petty users, but academic inquiries also help train models. ChatGPT announced free access to paid models to users with college IDs just at the nick of spring semester.
The economist Mariana Mazzucato highlights in her book “The Entrepreneurial State” how state funded military innovations like the DARPAnet, GPS, Calo and so on manifested into later consumer products of petty usage, like the internet, map applications and voice recognition.
South Park has a farcical episode on exactly how petty petty usage can be. The episode is titled “Deep Learning”; Bebe, a fourth grader, is awestruck by the long winded, romantic messages her boyfriend Clyde (also a fourth grader) sends her. Her female friends are envious, Clyde’s a real charm, and transgresses his fourth grade limitations with language to write his girlfriend exquisite texts. When Eric Cartman, one of the show’s protagonists, approaches Clyde about how he writes such great responses to his girlfriends texts, Clyde simply says, “ChatGPT dude!” Somewhere out there a fourth grader is doing this in actuality.
Maybe the frat boy sitting next to you in class isn’t exactly Clyde (even though that just might be his name), he is still among that group of petty users enjoying the spillovers of large technology.
But AI isn’t the sole technological upgrade impacting students. The student now is not the student then; there have been significant upgrades in terms of information accessibility, technological assistance and so on. Student input in LLMs is among the petty usages of large language models, furthering the training of data. Even though it provides an incentive in exchange, that incentive only sustains itself in the face of a largely static gradation system. It would be interesting to somehow quantify the effects of curricular change in a largely static gradation system.
We need to think about grades like an economist thinks about money: constantly inflating or deflating will only debase grades themselves, not in any manner be better at assessment. To assign work in the domain of repeatability is to ultimately consign it to tools like LLMS which at this phase generate results targeted around a median of academic performance.
If we wish to grow an immunity to academic automatons, then our only choice is to embrace new forms of assessments in the classrooms, and perhaps more importantly, new forms of assessing.
Leave a Reply