1. Do emergent phenomena exist?
Yes.
Here’s a picture of several, caught in the act of emerging.
This is a figure from a famous paper from 2022, Emergent Abilities of Large Language Models, by then-Google employee Jason Wei and his associates. The paper describes how, at a certain size, language models suddenly jump in capability:
“Figure 2: Eight examples of emergence in the few-shot prompting setting. Each point is a separate model. The ability to perform a task via few-shot prompting is emergent when a language model achieves random performance until a certain scale, after which performance significantly increases to well-above random.” (p.4.)
If you have ever wondered why we call them “Large Language Models,” (LLMs) this is the answer: Small Language Models don’t reason. While there is some debate about the context of the numbers, and the specific mechanics behind the threshold, there is no serious question whether the threshold exists. After all, the term “generative” is widely used, by proponents and skeptics alike, referring to outputs that the models were not programmed to create, and yet do.
The outputs emerge.
We have all enjoyed an eight-week course on complex adaptive systems, focused entirely on evolution. Surely this single focus was appropriate in 2020. Today, though, I wonder if we should not consider LLM development alongside evolutionary development. Where a decade ago there was one complex adaptive system; now, in many ways, there are two.
In other words, if we wish to juxtapose—presumably for our didactic pleasure—systems alternative to the evolutionary programming that has driven us out of the seas and into designer sunglasses, ChatGPT and its cousins are about the best thing we’ve got.
2. Who Is “We”?
Here is some hero text from the Climate and Clean Air Coalition’s webpage (with original emphasis):
WHY WE NEED TO ACT NOW
That same “we” has been invoked throughout our course discussions and during our lectures; “we” is thoroughly canonized in the literature, in fact. You know the “we” we are talking about.
Except, there is no we.
While one might be forgiven for replying “hah, not lately there isn’t,” I suspect the truth is that there has never been one, not since humanity scooted down out of the trees for the last time.
If we agree—I do—that evolution is a cosmic program that acts on every complex system it can get its hooks dug into, then we must accept a corollary: all human beings are subject to the laws of natural selection. Our social systems are just as susceptible to regulatory frameworks and path dependence as our cells are, and the nature of the body politic is just as “red in tooth and claw” as is the open plain.
The result is a society that operates not always how “we” want, but universally how it must.
Attempts to instruct humanity on the errors of its collective ways tend to ignore the reality that humans have very little agency because of selective pressures. Coalitions of the “better-minded” are themselves ineligible to apply for an exemption.
This is not to say that it would be impossible to change things for the better for humanity. It is to say that inventing a window screen or a dishwasher is your best bet. One might observe, with concern, that uptake on window screens and dishwashers is not universal, and these technologies seem unambiguously awesome.
Still, talking somebody into defeating their own self interests is not only less useful—it violates everything we know about how our universe works. Including everything we have learned the past eight weeks.