Following up on Ted Underwood’s recent post on the recent article “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” So yes my post title is a joke, but in my defense, they started it. Very briefly put, this is a conversation about natural language processing. NLP is about the nonhuman-computational processing of human speech/text (as when you say something to Siri) but also about the nonhuman-computational generation of text (such as GPT-3).
It’s with the generation of text that the parrot comes in.
Text generated by an LM is not grounded in communicative intent, any model of the world, or any model of the reader’s state of mind. It can’t have been, because the training data never included sharing thoughts with a listener, nor does the machine have the ability to do that. This can seem counter-intuitive given the increasingly fluent qualities of automatically generated text, but we have to account for the fact that our perception of natural language text, regardless of how it was generated, is mediated by our own linguistic competence and our predisposition to interpret communicative acts as conveying coherent meaning and intent, whether or not they do [89, 140]. The problem is, if one side of the communication does not have meaning, then the comprehension of the implicit meaning is an illusion arising from our singular human understanding of language (independent of the model). Contrary to how it may seem when we observe its output, an LM is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot.
“On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” 616-7
In other words, computers are like parrots. They might sound like they are making sense but they don’t know what they are saying. That is, more precisely, that whatever desires propel parrots to make those sounds, they are not clearly revealed by humans parsing the meanings of the words they hear in the parrot’s voice. However, it would be incorrect, I’d argue, to suggest that there is no desire behind parrot vocalizations.
What about nonhuman “stochastic parrots”? Are there desires shaping their synthesized speech? I imagine we’d all agree there are desires. At the very least there are the desires of the engineers and corporations that produce these products and also the desires of human interlocutors who ask Siri (etc.) something. That said, we do not have to imagine humans as the alpha and omega of desire. Some might argue for the spectral force of ideology or capitalism as the generator of these apparently human desires. Alternately, desiring-production might be viewed as a kind of vitalist force.
The parrot’s desire for expression arises through its participation in biological and geological strata as (part of) an assemblage… or so a Deleuzian, new materialist argument might begin.
The nonhuman stochastic parrot is (part of) an assemblage that intersects not only the geological and biological strata but the techno-semiological stratum as well. Whatever desires fuel its activity and regardless of who really means what, the text/speech of these nonhumans is coded in a way that a biological parrot’s speech is not.
In a way Bender et al raise a similar concern to that raised by Plato in the Phaedrus when he observes that writing lacks the living voice of the author. Of course these days, New Critical intentional fallacies aside, we usually have little problem ascribing our interpretations of texts to their authors. And really, as rhetoric scholars, we’ve been up and down this road in the post-structural age. However, unlike the lost intentions of writers, Bender et al want to argue there is no originary intention, no desire, behind the voice of the stochastic parrot. It’s dead, meaningless words from the start: static parading as meaning. But I’m suggesting that the stochastic parrot isn’t dead; it’s just pining for virtual fjords. Nonhuman does not mean without agency, and desires or intentions may be understood as emerging from an assemblage.
There are several other interesting and important points raised in this article calling for a reconsideration of the ethics around NLP. And who could argue against the importance of ethical considerations? I won’t go into them here. In that context this topic may seem esoteric and of little relevance, but I would argue that isn’t so (well, ok, maybe it’s a little esoteric). However if we are to consider the ethical implications of computational work in terms of its effect upon our climate, in its representations of cultural and human differences, in terms of the politics and values promulgated by its work, then we will benefit from an investigation of coded computational expression that does not begin with the assumption that it is meaningless and/or absent of intent or desire.
cyborg parrot image: https://www.deviantart.com/blonde-phoenix/art/cyborg-parrot-559481306
Alex Reid 