ALDaily, that inestimably worthy arts aggregator website, pointed me overnight to this fascinating article in Intelligencer, "You Are Not a Parrot" by Elizabeth Weil.
The article poured me down a rabbit hole into a whirlwind of speculation and intrigue about linguistics and computational learning. It enters and addresses areas of computer science well outside my comfort zone, as it explores with more rigor than I knew existed the growing conflict between Art and AI. I have been sporadically pondering some of this stuff in the poetic shallows here from time to time. It is slightly giddying to find out that this subject might have some real heft, depth and contemporary relevance. Even Glenn "Instapundit" Reynolds is now expressing doubts about the benefits of technology
Elizabeth Weil's article is largely about an impressive academic at the University of Washington, Emily M. Bender, and her actions in linguistics studies to attempt to challenge the headlong pursuit of the singularity in technology. The singularity is that apparently imminent putative time when it will no longer be possible to distinguish between human and machine. Some even say we are almost there already if only we could know it.
The article introduced me to the SALAMI: Systematic Approaches to Learning Algorithms and Machine Inferences. This is suggested as a less threatening and more apt way of describing Artificial Intelligence ("AI"), as it does not inadvertently instill in the concept some superior existential presence.
We also meet the stochastic parrot, an entity “for haphazardly stitching together sequences of linguistic forms … according to probabilistic information about how they combine, but without any reference to meaning. Stochastic means (1) random and (2) determined by random, probabilistic distribution."
And then the parrot story unfolds when on 4 December 2022, four days after ChatGPT was released, Sam Altman, the CEO of Open AI (it's creators), tweeted, “i am a stochastic parrot, and so r u.” A million people signed up to use ChatGPT in the first five days. It was the end of writing by humans!
In my limited observation this whirlwind is still in full spin, with young people naturally gravitating to the ease with which they can now appropriate ChatBOT functions to reduce the demands on their creative resources and the need for research to make content. And like all old people throughout history I instinctively respond with "but what is this going to mean for our culture, learning, civilization, truth and authenticity?" This ChatBOTTing stuff does have the whiff of bullshit about it after all. A bullshitter doesn't care if something is true or false. They only care about the rhetorical power of what is said, whether the audience is persuaded by it. And this AI generated content is persuasive. Maybe there is something there...
Bender and others wrote an influential article entitled “On the Dangers of Stochastic Parrots, Can Language Models be Too Big?” in March 2021 . It’s a review of Large Language Model (LLM) critiques looking at the biases encoded in such models and the near impossibility of studying what’s in the underlying data.
“People want to believe so badly that these language models are actually intelligent that they’re willing to take themselves as a point of reference and devalue that to match what the language model can do.”
Apparently until recently linguists agreed with Bender that: referents, actual things and ideas in the world, like coconuts and heartbreak, are needed to produce meaning. But a new school has now evolved called distributional semantics: “The meaning of a word is simply a description of the contexts in which it appears.”
Bender invokes Wittgenstein to counter this by defining language as inherently relational:
“a pair of interlocutors at least who were working together with joint attention to come to some agreement or near agreement on what was communicated.”
She believes there is too much effort trying to create autonomous machines rather than trying to create machines that are useful tools for humans. The idea that the human brain is a computer, and a computer is a human brain is seen as a dangerous, if powerful, metaphor. Such a notion affords “the human mind less complexity than is owed, and the computer more wisdom than is due.”
Bender seems to suggest that we need to live with more humility, not to create new artificial entities but to accept that we are creatures among other creatures, matter among other matter. Trees, rivers, whales, atoms, minerals, stars ...
"Artificial people will always have less at stake than real ones, and that makes them amoral actors, ...Not for metaphysical reasons but for simple, physical reasons: They are sort of immortal.”
It gets a bit weird around this point. It is suggested that if corporations make counterfeit people they will need to be held responsible. It's unclear how or by whom. Creating counterfeit people is characterized as a weapon of destruction against human society. These AI creating corporations need to take such creations as seriously as molecular biologists take the prospect of biological warfare or that physicists have taken the prospect of nuclear war. But they don't seem to understand what cosmic forces they are unleashing ...
The notion that AI is human or sentient is a seen as a misconception. Emily Bender says:
“We haven’t learned to stop imagining the mind behind it.”
Weil says: "Bender knows she’s no match for a trillion-dollar game changer slouching to life... LLMs are tools made by specific people — people who stand to accumulate huge amounts of money and power, people enamored with the idea of the singularity. The project threatens to blow up what is human in a species sense."
The argument against the singularity is built around a perceived inherent danger in blurring the line between human and machine. The fear is that if we share our society with counterfeit people, who we can’t differentiate from real ones, there will soon be no society at all.
But what this is not saying is that a singularity can't be achieved.
Only that it shouldn't.
And the gate for that horse to bolt through looks to have already been fully opened.
And here was me thinking that it was just that AI was poor at writing poetry. Maybe there's a little bit more to this.
Comments