My old professor was a committed nominalist. He abhorred reification, making what is abstract something concrete, and mining this rich vein of thought he would discourse on the notion that we do not have a mind. Ergo, there is no such thing as mental illness. As you might imagine, in the age of the psychologist, he came under a lot of fire as a reactionary ‘denier’.
Of course, such responses to his reasoning merely unmasked the bias of his interlocutor — the unwillingness to contemplate anything that runs counter to their perceived understanding. This professor spent a career as a working psychologist, including in institutions with societies most troubled, he was all too aware of the joy and torment that goes on in our heads. Thus, his objections were not to deny, rather to challenge how people use language to talk about the acts of thinking, feeling, knowing, understanding. Language which he thought, if misused, only compounded the suffering of his clients.
Hearing his and others reasoning on the subject and augmenting it with what is now enough personal experience to bore people at dinner parties, I cannot help but return to Wittgenstein’s notion that ‘the limits of my language are the limits of my world’. When we appropriate words, when we conflate distinct concepts, when we seek to change the world by changing the meaning of words, we may win the moment but we mortgage the future. This is because in redefining words to limit or prevent any possibility of counter-debate, we cut ourselves off from imagining things that are undreamt of in our philosophy.
In a recent article on ChatGPT for The New Yorker, Ted Chiang explored the notion of not viewing the service as AI — a view I share — but instead to:
Think of ChatGPT as a blurry jpeg of all the text on the Web. It retains much of the information on the Web, in the same way that a jpeg retains much of the information of a higher-resolution image, but, if you’re looking for an exact sequence of bits, you won’t find it; all you will ever get is an approximation.
Ted Chiang
To counter this gap in language, and because the chatbot cannot think, ChatGPT leverages it’s modelling to fill in the gaps in an attempt to sharpen the image. This leads to an output which is plausible enough to seem true to the untrained mind, but that in reality requires checking and correcting against third party data.
What I find particularly interesting about this is the speed with which people are willing to devalue the output of ChatGPT because of the same process that happens in us. Our memory is, to use Chiang’s analogy, a blurry image that our present self then interpolates to form a coherent vision of topic.
Returning briefly to my old professor, he would expound on this conceptualisation to the seminar, only to face howls of derision from the cohort. Living in Sydney, he would invite everyone to engage in a small experiment — close their eyes and count the verticals of the Harbour Bridge. If our memory is truly lossless, we could simply retrieve the image of the bridge and then count off the verticals. But instead, our memory of the bridge is one part retrieval and one part ‘hallucination’. The greater our familiarity with topic, the more detailed or ‘lossless’ our memory. At least in theory.
As with the experiment of the bridge, Chiang observed another factor in play with respect to ChatGPT. The notion of repeatedly ‘saving’ information:
Repeatedly resaving a jpeg creates more compression artifacts, because more information is lost every time. It’s the digital equivalent of repeatedly making photocopies of photocopies in the old days. The image quality only gets worse.
Ted Chiang
Though he used this analogy in the context of a newer iteration of a Chatbot learning from an earlier bot’s output, it is apt in conceptualising the trap into which we can easily fall when thinking about a topic. If we meditate on subject matter with ill understood or reimagined definitions of words, we are the equivalent of resaving a jpeg — the quality of our thinking only deteriorating the more time we spend reflecting on the topic.
This behaviour is writ large on social media, where people pickup on a redefinition of language they already know, feel heartened because it confirms their bias — I am not alone or wrong to think this — and then the algorithm provides dopamine hits in the form of an endless stream of ‘likeminded’ people. All of which permits us to form a view of the world that reinforces our ideology but may be devoid of truth — unless everyone else adopts our outlook in a Jonestown approach to reality.
With nothing but arbitrary re-definitions of words to fall back on, and Thought Police to brainwash and torture dissidents, it is striking that the most bile invariably comes from those who favour concepts such as ‘Love Wins’, ‘Peace Rules’, and ‘Be Kind’. In the face of such ferocity, coming from a self-asserted place of love, many fear to protest lest their objection be seen as confirmation of the very prejudice the ideologues seek to expose. As one commentator recently put it, ‘only a witch would deny the existence of witchcraft’.
Yet it is a witchcraft that comes not just from the appropriation of words, but also from their absence. To return to the sphere of technology again, one of the challenges of data storage systems is space. All devices are limited, to solve for this problem computer scientists often turn to compression as a way to store more in less space. But compression does more than just save disk space. It also, according to researchers like Marcus Hutter, offers parts of the solution to achieving true AI. As Chiang observed:
If a compression program knows that force equals mass times acceleration, it can discard a lot of words when compressing the pages about physics because it will be able to reconstruct them. Likewise, the more the program knows about supply and demand, the more words it can discard when compressing the pages about economics, and so forth.
Ted Chiang
This process is what happens when we hear an individual speak and compress their speech to ‘free trade’ or ‘protectionist’, ‘climate change’ or ‘climate denier’. While there are considerable benefits to this in terms of functionality, there are also considerable drawbacks. The millennia of evolution that has brought us to the present day has favoured survival. This means we are more likely to perceive a threat, in something we cannot see or understand, than we are to perceive a benefit. While this was of great value when we roamed the savanna, and there was a potential predator lurking in the nearest bush, it is of quite the opposite value in the life of the mind where exposure to ‘risk’, in the sense of hearing something that contradicts our world view, is vital to gain an accurate understanding of the world. This is where the sources come in.
When I think about a topic, I recall theories and interactions, form analogies, and start throwing words on the page. But I always dive into my redoubtable zettelkasten for a higher level of resolution than I can recall. When my zettelkasten does not yield sufficient results, particularly counter arguments, it is a case of ‘to the library!’ Here I can sip from the cup of more than 2 million publications. While not an exhaustive catalogue of the world’s knowledge, there is enough range to form something approaching an authoritative argument on the topic.
The difficulty is the argument that emerges from a close reading of the sources can be — problematic. It can contradict our original thinking. It can take us down a line of thought we fear or may even hate. All of this can be triggering. But because this can be too much for some, much easier to continue to resample the image. To continue to redefine the language, reframe the argument, and reassert our outlook until the result is more harmonious to how we would like the world to be. This edited memory, the lossy jpeg to return to our earlier metaphor, is then resaved for future retrieval.
Though flawed, this nonetheless conforms to Foucault’s assertion that ‘the only valid tribute to thought… is precisely to use it, to deform it, to make it groan and protest.’ Unfortunately, this is to protest in the original sense of the word — not to speak against, but to speak for. As in ‘the lady doth protest too much, methinks.’ A process that attempts to assert something as true when the undeformed sources do not support this reading.
Where this leaves us in apprehending how much a topic matters to an individual. If they care little about the topic, they may retort when pressed for details ‘I don’t know, you’ll need to ask someone else’. But if the topic matters greatly, if they have a strong motive for you to accept their detail of the topic, then they will increasingly fill in the memory gaps with that which asserts their truth. Painting a picture for you, an impression of the world which may be a very long way from its reality. If you make a decision based on this picture, then your judgement will be flawed. But take a critical approach, reject the premise, and test the picture against a wider data set and different ideological biases, and you start to form a more accurate view of the topic.
At this point, I suspect you are thinking ‘what hope for the morrow?’ The hope is much, for having gotten this far is to be forewarned and thus forearmed. In that we do well to employ scepticism when listening to a human interlocutor. Because even the best of us are filling in the blanks in our memory.
Scepticism also means that instead of buying into the hallucination that is purely ‘lived experience’, we can check the people we interact with against third party sense data, against the physical libraries of other peoples’ thoughts and experiences. If their memory validates against objective truth, we can celebrate. If it is falsified, we are in a position to call the hallucination for what it is — Pseudo-Profound Bullshit.
Good night and good luck.
