I. The Device
Imagine a device that listens to all of your conversations. It has a single purpose: live-running your utterances through text-davinci-003 to evaluate how surprising each token is, given its preceding context.[1]
Users could chart their verbal novelty throughout the day; highlight and share their most unusual turns of phrase; note which friends put them on well-worn paths and which help them careen into the chaotic underbrush.
Undoubtedly, some would be disturbed by the long stretches where every one of their words could easily have been generated by the model. In those hours, they might as well have lost consciousness, taken a little nap while Daddy Altman autopiloted their mouth.[2]
Consider: for how many people would this be true at virtually all times? If President Andreessen were someday to mandate the wearing of these devices, scoring each person on their overall surprising-ness, what would the distribution be?[3]
II. The Quotient
Let’s call it a novelty quotient. How correlated is it with IQ? Certainly somewhat: a greater vocabulary widens the possible sentences available, and isomorphically,[4] intelligence and knowledge are likely to assist the dedicated weirdo in finding ever-more-remote unexplored territory.
But we can all think of high-IQ people who we’d wager have reliably low NQ, sticking to stifling convention like a straitjacket. Conversely, there are bizarre, stupid, bizarrely stupid, and stupidly bizarre people. Think of the difference between boring-dumb and crazy-dumb.
Which brings us to another question: how correlated would NQ be with insanity? There’s likely to be some link… But it can’t be absolute. The ravings of someone who has fully lost touch with reality are very different from typical speech, but they’re completely predictable. Given fifty sentences of paranoia or hours of repeated compulsions, it’s easy to see what’s coming next. Even more than for the healthy.
Is NQ just a measure of creativity? Not really — creativity is usually evaluated as an ability to generate novel ideas in service of a particular goal. NQ seems more like a precursor, an ingredient to be combined with smarts and moxie to generate a desirable outcome.[5]
III. So What?
Okay, so NQ seems likely to be somewhat-but-not-entirely-correlated with IQ, somewhat-correlated-but-not-at-extremes with insanity, and a precursor of creativity.
The real question, which you’ve undoubtedly sensed is coming: how important is NQ? Is it adaptive?
This is the part of the post where I tell you that NQ is so important,[6] that it’s going to overtake IQ as the biggest determinant of status and economic value in society, that AI will automate away all of the IQ-heavy tasks that have driven our sorting into technocratic socioeconomic strata, leaving us competing to peacock our strangeness in exotic new artforms, as a sort of proxy for our ability to inject real chaos and entropy into a system that lacks only for new roads to run its perfectly-optimized jihads down.
Do I believe that? Kinda, yeah. But I’m a weirdo, so it’s in my best interests to believe weirdness is the future.
Still, I could argue: the Industrial Revolution brought about a massive transfer of power from the martial elite (something like today’s jocks) to the technical elite (today’s nerds). Exemplified by little Napoleon, rising from his role as head engineer of the artillery corps to dominate all of Europe, largely through impeccable logistics.[7] This new revolution will provoke a similar shift.
Note, for instance, the increasing influence of so-called “schizoposters,” valued specifically for their nonconformity and odd takes,[8] on tech titans like Elon Musk, Daddy Altman, and President Andreessen.
Or, more modestly, the viral outputs of image and text models: always the faithful execution of some person’s off-the-wall concept. What would that look like on a world-historic scale?
IV. Nevermind
But this argument is a little obvious. It could easily be completed by, say, text-davinci-004. I’d rather save the effort, simply inject the germ of the idea and move on.
Besides, I’m trying to buff my NQ, here. Have you ever snowed in strawberry morning? Verily, peregrine.
- ^
Incidentally, this is not at all far-fetched. See the Amazon Halo, a largely-ignored 2020 fitness(?) tracker that performs sentiment analysis on all of one’s conversations, spitting out reports on the happiest, saddest, and angriest moments of each day.
My experience has shown these to be startlingly accurate. Enough to provide insight into feelings that had remained largely subconscious before then (or at least, the post hoc illusion of doing so).
- ^
Autopilot my mouth, Daddy.
- ^
In this timeline, compliance is universal (because President Andreessen is so beloved) and behavior is unchanged by the measurement (because this feature is a secret project, spearheaded by a fallen Edward Snowden after the collapse of the Russian Federation). It’s an exciting universe.
- ^
See? Pick me!
- ^
Also, creativity involves big, deliberate swerves off the conventional route. NQ, as we’ve defined it, is habitual, and constant.
One could imagine measuring it like heart rate: we each have a consistent baseline, and a max we can swing up to during periods of exertion. Both would be useful to gauge.
- ^
In Our Modern Times
- ^
Have problems with this history? Fair. Btw, Romans? The wolves would simply eat the babies.
- ^
They even have the concept of an “NPC,” which is essentially a slur for someone with low NQ.
This is very upsetting to me.
People would start using big words they don’t understand or use uncommon synonyms when a small common word would do. I hate it when people do this trying to sound smart. The archetypical example of this is Kingpin in the Marvel shows, who I genuinely can not stand. More people sounding like midwit try-hards does not lead to a better world.
Increased neologism. They’re funny, but decrease the quality of communication for everyone involved.
Some relevant links: https://ahrm.github.io/jekyll/update/2022/04/14/using-languge-models-to-read-faster.html https://www.gwern.net/docs/psychology/novelty/index https://arxiv.org/abs/2210.14140 https://arxiv.org/abs/2109.00698#eleutherai
You will get cursed by Goodhart.
You can increase your NQ by learning new things, or trying new things. But you can increase it even more by saying random things.
Truly random things are boring, but difficult to predict exactly. More precisely, you can predict that the sequence of the words will be boring, but you cannot predict the exact words. So from the mathematical perspective, you get maximum variance, but from the psychological perspective, you always get the same thing.
Also read Diversity is all you Need, Open-Ended Reinforcement Learning, maximum entropy reinforcement learning, Kevin Frans’ blog
Would the NQ be calibrated to common public text corpus or things you personally have said? One interesting option is to think about those that have low personal NQ but high societal NQ.
(cf. nonstandard standard toolbox.)
Osman’s sleepers Hayden scintillating agglutinate unnerving styli Aleutian’s sacs stardom’s stepfather’s Aron’s delegates noisy substitutions Johanna ICBMs respectable chamois’s espies theme’s clobbers downpour’s cagey Chateaubriand.
Hard to predict, but not very interesting. I don’t think you’ll get very far without semantic content analysis. In fact, within a given idea, redundancy is CRITICAL to getting that idea across to the intended hearer. Nobody sane goes for lexical surprise, and even conceptual surprise is somewhat constrained in the dimensions which are interesting and/or useful.