As people grapple with ChatGPT 3 and other instances of generative artificial intelligence, we sometimes turn to imagination in order to describe and understand the technology.
I’ve seen folks raise the Terminator movies, HAL-9000, and generic scary robots to express their fears, for example.
Others use history, bring up comparisons from prior technological revolutions, to think through possibilities: AI like the European printing press. One of my young students told me AI feels to him like the appearance of the world wide web must have felt to me in the 1990s. (And sometimes it does.)
In this post I’d like to offer two historical analogies or imaginative figures for how we might experience generative AI.
I: The Calculator
When I started playing with generative AI and thinking about its educational implications I thought back to the rise of the commodity digital calculator, back in the 1970s. I remembered debates over its potential impact which occurred around me. Students might not learn basic math, went one claim, and would instead outsource those crucial skills to handhelds. Numerical literacy would dwindle, just when it was needed as society became every more closely tied to rapidly advancing science and technology.
Over time we integrated calculators into teaching in various ways. To oversimplify, some classes went on require graphing calculators. Others, instructing younger students, teach the operations first, then let students outsource the work later on. In other words, a pedagogical consensus emerged which included the technology. The consensus persisted even as the physical calculator migrated into software forms.
There are already practices emerging now which follow the calculator’s story in apprehending generative AI. For example, some instructors want to have students use ChatGPT to create essays, then critique them as bland, badly cited writing, which helps them advance their own composition abilities. Alternatively, students can use large language model tools to create first drafts of content, then edit, amend, and improve them on their own. And so on.
Remember, too, that while we are accustomed to calculators in our lives, as they are embedded deeply in them to the point of being background noise, we don’t actually use them simply. The world rarely gives us simple math problems we can enter, then results we can use straight away. Instead, life mostly presents us with those dreaded story problems, which we need to translate into an operation calculators will perform. Then we need to do stuff with the results. So it is with generative AI. On the front end we need to formulate useful parameters, which takes some doing (and how many classes teach this?). On the back end we often have to work with the results: picking the best image of a set, redoing the prompt, perhaps editing the best image in another app. Or we take a chatbot’s text as a draft to revise.
Now, this calculator comparison assumes generative AI is competent as a calculator. It isn’t the case in reality now, as a range of AI make mistakes or produce terrible results. Yet we might see the technology’s quality improve to a point where many people find it sufficient for their purposes. As one observer put it, “ChatGPT and its fellow essay bots are simply the scientific calculators of writing in a world that is still obsessed with four-function calculators.” Heck, you can use ChatGPT to code a calculator.
II: Igor, the mad scientist’s unstable assistant
Alternatively, Bing’s chatbot et al might not become that reliable. Instead, these tools might act erratically. Like a mad scientist’s hunchbacked assistant. As Igor.
Igor wants to help, but sometimes gets… creative, and provides results far from what we asked for. Igor usually obeys us (the mad scientist), but sometimes wants to follow his own plan or the voices of others (think of the famous “guardrails”). Remember the strangest art which you’ve coaxed from Stable Diffusion or Craiyon, those transmissions from the uncanny valley, or read about a New York Times writer’s weird Bing chat. ChatGPT and Bing’s chatbot do quickly leap to churn out the text you require, yet at times will just balk, as per its internal (and sometimes mysterious) guidelines. And ChatGPT is capable of cheerfully producing horrors on demand.
There’s a great, relevant scene in the fantastic Bride of Frankenstein (1935). The two mad scientists (it’s such a fine film that it won’t settle for just one), Frankenstein and Praetorius, complain about the quality of hearts in the cadavers they have.
One assistant hears this and helpfully stalks off to grab and murder a casual passerby, then provide the resulting fresh corpse, which pleases the mad scientists. The assistant did help out, albeit in an, er, unorthodox fashion.
Playing the Igor off of the calculator, we see two very different understandings of generative AI.
Either we see it as stable or unstable. We view these tools as easily understood (most calculator functions are clear to many users) or susceptible to following commands other than our own. Its quality is durable or sometimes risible.
Both technologies and their analogies can be frightening at times. The calculator is less so, yet like many labor-saving devices, threatens to weaken our individual capacity to perform, or even understand, that labor. The hunchbacked assistant can easily wreck havoc, even when trying to follow our instructions. While a calculator is cool, unremarkable, and office-friendly, an Igor is ungainly, warped, unpredictable, and strange.
There are other figures and metaphors for us to use, of course. Today I’d just like to offer and contrast these two, at least for entertainment, and perhaps for a touch of imagination.
I’d like to close with an image from Bride of Frankenstein, which combines a freaking assistant with cool, reliable tech: