The Destructo Test
When will life without AI seem unthinkable?
Back in my TechCrunch days I wrote a lot about cryptocurrencies1, until five years ago I gave up. The main reason was a test I call destructo ad absurdum2: if a technology were to vanish overnight, how many people outside that field would be profoundly affected? Fifteen years into the so-called ‘crypto revolution,’ that answer remains ‘very few.’ Some, sure: Russian expats who escaped the Ukraine war to Argentina and now get paid in Bitcoin, gamblers with their net worth in memecoins. But few, and fewer yet if you kept Bitcoin and erased everything since. One should expect more of a field accompanied by fervent claims of imminent global transformation.
The Destructo Test is a good lens through which to look at emerging technologies in general. AR/VR? Fail: some gamers, a few industrial uses, Meta Ray-Ban adopters, but no profound changes. Drones? They pass: we may not be getting Amazon packages by quadcopter yet … but warfare as we now know it has been utterly transformed.
AI? Speaking of fervent claims of imminent global transformation? Well. That's interesting. This week AI pioneers were awarded Nobel Prizes in both Physics and Chemistry, which gives a sense of its all-important aura. And it's certainly true a few external-to-AI-proper fields, e.g. protein folding, have been profoundly reshaped. But “popular AI,” LLMs and diffusion models as we know them? ChatGPT, Claude, Grok, Midjourney, Runway, etc.? If they vanished tomorrow, how many outside the field would be profoundly affected?
That question is kind of a cheat. A huge number of people would be somewhat affected. Hundreds of millions use ChatGPT weekly for a multitude of purposes. Millions of developers rely on Copilot, Cursor, etc., to automatically format their loops and generate glue code; Amazon has outsourced thousands of person-years and $250 million of grunt coding work to LLMs. Midjourney’s Discord has 20 million members. All that from a standing start of near-zero only two years ago! Popular AI is, clearly, a stunning, spectacular, skyrocketing hockey-stick success.
But The Destructo Test is still instructive. If all the above were to suddenly disappear, their users would simply go back, sad and grumbling, to working without them as they did before. AI is not yet a cornerstone, or even a building block, of much economic or artistic or scientific activity. (I said much.) It is not yet woven into our quotidian lives and workplaces such that its removal would be wrenching or even difficult. That is the expectation; and there are countless experiments, pilots, proposals, and startups aiming to it make it moreso; but few are much more mature than that. For most people AI is still far from passing The Destructo Test.
So when will we get there, and how?
The transformation of a single industry would be enough, and the most likely such industry is software. In fact the kind of profound transformation we’re talking about might not even be necessary to justify all the money poured into AI. A Fermi estimate: there are ~30 million software developers worldwide; figure an average global salary of ~$35K/year; that’s $1 trillion/year. If AI makes that 20% more efficient, even if only by shoring up the reduce-technical-debt, add-test-coverage side of software rather than building sexy new things, that’s $200 billion/year of value.
To reiterate the obvious, $200 billion of value in a single industry is a lot! And puts into perspective the investment of single-digit billions into OpenAI and Anthropic, especially when AI is expected to reshape all industries, and by doing much more than just making them more efficient.
As such one answer to “how and when will AI pass The Destructo Test?” is “in the GDP numbers.” The sunny Pollyanna perspective is that AI will bring a productivity boom the likes of which we have never seen before; its doom-and-gloom obverse is that it will render human jobs redundant at a scale we have never seen before. But that answer seems handwavey and indeterminate. We should expect to pass The Destructo Test well before AI transforms the entire economy. Again: how and when exactly?
The Candidates
Software
The more specific answer “by obsoleting software engineers—especially mediocre ones—en masse over the next, say, three years” is, awkwardly, one that both fits our criteria and seems entirely plausible, as software engineers are beginning to realize. When we live in a world that continues to incorporate ever more software … coupled with decreasing demand for software engineers … we will pass The Destructo Test.
Robots
It may be underappreciated how much modern AI has transformed robotics. From Google to startups, the same transformer architecture that powers LLMs also drives modern robotics, and is incorporated into the driverless Waymos that prowl San Francisco, Phoenix, and Los Angeles today, and many more locations tomorrow. Today, they are a (very impressive) option for a small minority, but it’s easy to imagine a near future where the absence of Waymos—and then, home robots—is unthinkable.
Animation
Hollywood in general is both fascinated and freaked out by AI, with good reason in both cases. Animation, already computer-generated, seems its lowest-hanging / most-vulnerable fruit. It’s easy to envision a near future in which an entire screenplay is converted into a set of characters, settings, and scenes, then rendered into shots that get stitched together into a full animated movie … in near-real-time … after which each shot can be edited by simply telling the AI what to do. Will the character voices need to be redubbed by humans? Probably. Would this still change the creative process forever? Yes. Would the result be art? This is a deep epistemological debate, but I note it’s fairly rare for individual shots to be where a film’s ‘art’ lies; rather, it’s in how they are structured and combined, still an entirely human-driven process here.
Business Processes
I know, I know, my eyes just glazed over too. But enterprises have processes, and people are using AI to optimize them, in both private and public sectors. Customer service, since it was already a chat interface, was only the first step. Accenture is now booking a billion dollars a quarter, and climbing rapidly, in Gen AI revenue. 1 in 6 public employees already use Gen AI regularly. At some point this will stop being a new tool and start being the tool.
Journalism
I’ve written about this very recently, and to quote myself: “Try to imagine the contrary case, in which an extraordinary new technology that can make sense of language, audio, video, and imagery—one that can turn both structured and unstructured data into meaningful narratives, and vice versa!—does not have a profound impact on a vital industry, and pillar of society, that is all about narratives, and is currently hemorrhaging money, people, trust, and relevance. I think you’ll agree that’s a very hard case to make.”
Amanuenses
This is my favorite option, and all of the above can be recast as this. The world’s greatest mathematician already thinks we’re one step away from AI being as good as “a competent graduate student” for his purposes. Imagine a world in which everyone has a graduate student on call, to perform research, draft papers, run experiments, do paperwork, handle administrivia—to handle all the gruntwork and let you focus on what’s actually important.
We actually took the first step towards this in the oughts, when people started calling Google their “exocortex,” because it stored knowledge for them. Google couldn’t actually do anything for them, though. Now that’s changed. Here, AI becomes everyone’s amanuensis—OK, fine, ‘assistant’ if you prefer, or evern ‘Jarvis’, although ‘amanuensis’ is a far better word —which we soon struggle to imagine working without, just as kids today struggle to imagine a world with no Internet. It doesn’t replace us; it complements us. I don’t deny that, again, there will be gloomy obverses to this optimistic outcome. But every change comes with both costs and benefits, and there’s little denying that the latter seem increasingly immense.
A very weird side effect of this is that of late, strangers have bombarded me with questions about Satoshi Nakamoto’s identity, because an HBO documentary has brought the question back to mind and I publicly corresponded with one of the predicted suspects who is alas no longer with us. (I’ve written about my theories on the subject here before.)
Yes, this should really be excisio ad absurdum; but, with apologies to my late mother who once taught Latin, destructo really rolls better off the English-speaking tongue.