Thread by @doctorow

Here's a fun AI story: a security researcher noticed that large companies' AI-authored source-code repeatedly referenced a nonexistent library (an AI "hallucination"), so he created a (defanged) malicious library with that name and uploaded it.

1/

If you'd like an essay-formatted version of this thread to read or share, here's a link to it on , my surveillance-free, ad-free, tracker-free blog:



2/pluralistic.net
pluralistic.net/2024/04/01/hum…

Thousands of developers automatically downloaded and incorporated it as they compiled the code:



3/theregister.com/2024/03/28/ai_…

These "hallucinations" are a stubbornly persistent feature of large language models, because these models only give the illusion of understanding.

4/

In reality, they are just sophisticated forms of autocomplete, drawing on huge databases to make shrewd (but reliably fallible) guesses about which word comes next:



5/dl.acm.org/doi/10.1145/34…

Guessing the next word without understanding the meaning of the resulting sentence makes unsupervised LLMs unsuitable for high-stakes tasks. The whole AI bubble is based on convincing investors that one or more of the following is true:

6/

I. There are low-stakes, high-value tasks that will recoup the massive costs of AI training and operation;

II. There are high-stakes, high-value tasks that can be made cheaper by adding an AI to a human operator;

7/

III. Adding more training data to an AI will make it stop hallucinating, so that it can take over high-stakes, high-value tasks without a "human in the loop."

8/

These are dubious propositions. There's a universe of low-stakes, low-value tasks - political disinformation, spam, fraud, academic cheating, nonconsensual porn, dialog for video-game NPCs.

9/

But none of them seem likely to generate enough revenue for AI companies to justify the billions spent on models, nor the trillions in valuation attributed to AI companies:



10/locusmag.com/2023/12/commen…

The proposition that increasing training data will decrease hallucinations is hotly contested among AI practitioners. I confess that I don't know enough about AI to evaluate opposing sides' claims.

11/

But even if you stipulate that adding lots of human-generated training data will make the software a better guesser, there's a serious problem. All those low-value, low-stakes applications are flooding the internet with botshit.

12/

After all, the one thing AI is unarguably *very* good at is producing bullshit at scale. As the web becomes an anaerobic lagoon for botshit, the quantum of human-generated "content" in any internet core sample is dwindling to homeopathic levels:



13/pluralistic.net/2024/03/14/inh…

This means the cost of adding another order of magnitude more training data isn't just massive computation - data will be orders of magnitude more expensive to acquire, even without additional liability arising from new legal theories about scraping:



14/pluralistic.net/2023/09/17/how…

That leaves us with "humans in the loop" - the idea that an AI's business model is selling software to businesses that will pair it with human operators who will closely scrutinize the code's guesses.

15/

There's a version of this that sounds plausible - the one in which the human operator is in charge, and the AI acts as an eternally vigilant "sanity check" on the human's activities.

16/

For example, my car has a system that notices if I activate my blinker with a car in my blind-spot. I'm pretty consistent about blind-spot checking, but I'm also fallible and there've been a couple times where the alert saved me from making a potentially dangerous maneuver.

17/

As disciplined as I am, I'm also sometimes forgetful about turning off lights, or waking in time for work, or remembering someone's phone number (or birthday). I like having an automated system that does the robotically perfect trick of never forgetting something important.

18/

There's a name for this in automation circles: a "centaur." I'm the human head, and I've fused with a powerful robot body that supports me, doing things that humans are innately bad at.

19/

That's the good kind of automation, and we all benefit from it. But it only takes a small twist to turn this good automation into a *nightmare*.

20/

I'm speaking here of the *reverse-centaur*: automation in which the computer is in charge, bossing a human around so it can get its job done.

21/

Think of Amazon warehouse workers, who wear haptic bracelets and are continuously observed by AI cameras as autonomous shelves shuttle in front of them and demand that they pick and pack items at a pace that destroys their bodies and drives them mad:



22/pluralistic.net/2022/04/17/rev…

Automation centaurs are great: they relieve humans of drudgework and let them focus on the creative and satisfying parts of their jobs.

23/

That's how AI-assisted coding is pitched: rather than looking up tricky syntax and other tedious programming tasks, an AI "co-pilot" is billed as freeing up its human "pilot" to focus on the creative puzzle-solving that makes coding so satisfying.

24/

But hallucinating AI is a *terrible* co-pilot. It's often good enough to get the jobm but it also sneakily inserts booby-traps that are statistically *guaranteed* to look like good code (that's what next-word-guessing programs do: guess the statistically most likely word).

25/

This turns AI-"assisted" coders into *reverse* centaurs. The AI churns out code at superhuman speed, and you, human in the loop, must maintain perfect vigilance to review its code, spotting cleverly disguised malware hooks the AI can't stop inserting into its code.

26/

As "Lena" writes, "code review [is] difficult relative to writing new code":



Why is that? "Passively reading someone else's code just doesn't engage my brain in the same way. It's harder to do properly":



27/

https://twitter.com/qntm/status/1773779967521780169

https://twitter.com/qntm/status/1773780355708764665

There's a name for this phenomenon: "automation blindness." Humans are just not equipped for eternal vigilance. We get good at spotting patterns that occur frequently - so good that we miss the anomalies.

28/

That's why TSA agents are so good at spotting harmless shampoo bottles on X-rays, even as they miss nearly every gun and bomb that a red team smuggles through their checkpoints:



29/pluralistic.net/2023/08/23/aut…

"Lena"'s thread points out that this is as true for AI-assisted driving as it is for AI-assisted coding: "self-driving cars replace the experience of driving with the experience of being a driving instructor":



30/

https://twitter.com/qntm/status/1773841546753831283

In other words, they turn you into a reverse-centaur. My blind-spot double-checking robot allows me to make maneuvers at human speed and points out the things I've missed.

31/

But a "supervised" self-driving car makes maneuvers at a computer's frantic pace, and demands that its human supervisor tirelessly and perfectly assesses each of those maneuvers.

32/

No wonder Cruise's murderous "self-driving" taxis replaced each low-waged driver with 1.5 high-waged technical robot supervisors:



33/pluralistic.net/2024/01/11/rob…

AI radiology programs are said to be able to spot cancerous masses that human radiologists miss. A centaur-based AI-assisted radiology program would keep the same number of radiologists in the field, but they would get *less* done.

34/

Every time a human assessed an X-ray, the AI gives a second opinion. If the human and AI disagree, the human goes back re-assess the X-ray. We'd get better radiology, at a higher price (the price of the AI software, plus the additional hours the radiologist would work).

35/

But back to making the AI bubble pay off: for AI to pay off, the human in the loop has to *reduce* the costs of the business buying an AI. No one who invests in an AI company believes that their returns will come from business customers to agree to *increase* their costs.

36/

The AI can't do your job, but the AI salesman can convince your boss to fire you and replace you with an AI anyway - that pitch is the most successful form of AI disinformation in the world.

37/

An AI that "hallucinates" bad advice to fliers can't replace human customer service reps, but airlines are firing reps and replacing them with chatbots:



38/bbc.com/travel/article…

An AI that "hallucinates" bad legal advice to New Yorkers can't replace city services, but Mayor Adams still tells New Yorkers to get their legal advice from his chatbots:



39/arstechnica.com/ai/2024/03/nyc…

The only reason bosses want to buy robots is to fire humans and lower their costs. That's why "AI art" is such a pisser.

40/

There are plenty of harmless ways to automate art production with software - everything from a "healing brush" in Photoshop to deepfake tools that let a video-editor alter the eye-lines of all the extras in a scene to shift the focus.

41/

A graphic novelist who models a room in The Sims and then moves the camera around to get traceable geometry for different angles is a centaur - they are genuinely offloading some finicky drudgework onto a robot that is perfectly attentive and vigilant.

42/

But the pitch for "AI art" is "fire your graphic artists and replace them with botshit." They're pitching a world where the robots do all the creative worl (badly) and humans work with robotic pace and vigilance, to catch the mistakes the robots make at superhuman speed.

43/

Reverse centaurism is *brutal*. That's not news: Charlie Chaplin documented the problems of reverse centaurs nearly 100 years ago:



44/en.wikipedia.org/wiki/Modern_Ti…

As ever, the problem with a gadget isn't what it does: it's who it does it *for* and who it does it *to*. There are plenty of benefits from being a centaur - lots of ways that automation can help workers.

45/

But the only path to AI profitability lies in *reverse* centaurs, automation that turns the human in the loop into the crumple-zone for a robot:



46/estsjournal.org/index.php/ests…

I'm touring my new, nationally bestselling novel *The Bezzle*! Catch me in Boston with Randall "@XKCD" Munrow (Apr 11), then Providence (Apr 12) and beyond!



eof/ pluralistic.net/2024/02/16/nar…