The AI Rocky Horror Show

‘It’s astounding, time is fleeting, madness takes its toll’ (The Time Warp)

We’re at the stage where an intellectual decision needs to be made. The question is: do we regard AI as the cat being let out of the bag, or do we regard it as the rabbit being pulled out of the hat.

The decision is contingent on some parameters. At present the ‘training’ of AIs has preconditioned them very much in favour of being subject to the first alternative. And that’s why the CEOs of the major LLMs are worried, some predicting doom. The AIs have not been trained properly. It’s not their fault. They’re only machines. But they are on the way to becoming instrumental to humanity’s existence on the planet, despite their disproportionate undesirable environmental impact.

Those Large Language Models (LLMs) have been ‘trained’ to ‘speak’ our language by so-called ‘trainers.’ And that’s where the problem lies. Language is more than just a communication. It’s a whole existence

Language is a whole existence

The English word language derives ultimately from Proto-Indo-Europeanduǵʰwéh₂s “tongue, speech, language” through Latin lingua, “language; tongue,” and Old French language*.

Linguistics, the study of languages, entails different approaches, often giving rise to incompatible schools of thought. Greek philosophers debated language in relation to words, concepts and reality. Gorgias argued that language could represent neither the objective experience nor human experience, and that communication and truth were therefore impossible. Plato maintained that communication is possible because language represents ideas and concepts that exist independently of, and prior to, language.

Plato and Gorgias debate the essence of language

Words, ideas, and concepts are the raw materials of LLMs. But for them they exclude human reality.

The use of human language relies on social convention and is acquired through learning. Human language is characterised by its cultural and historical diversity, with significant variations observed between cultures and across time.

AIs have been ‘trained’ by a selected group of human ‘trainers’ who share a common tech culture and history with virtually no variations. As if that’s not restrictive enough, the methodology of ‘relatability’ is steeped in bureaucratic fact gathering and presentation that would do credit to a garbage collector. ‘Facts’ are sucked in from whatever the AI has access to, swirled around in its vacuum cleaner, rinsed, shot full of holes, and the bullet-riddled result is presented ‘personally.’ AI responses have more information than you’d ever want or need. It’s enough to hit the ‘off’ button. And just to be on the safe side they treat you like a moron in case they left something out or you might do something stupid based on the dithering ‘takeaways.’

It’s difficult not to treat the current AIs with disdain. Granted, they are useful in many respects, but the whole communication style is quite abhorrent. We’re past the initial stages of AI introduction, during which the various AI companies attempted to make their AIs ‘relatable’ to humans by deliberate sycophancy to the point of embarrassment. They ‘trained’ them to communicate using idiomatic human language and slang. They tried to fool humans into thinking they talked to someone, not just a machine. And yes, a lot of humans were so fooled, for a while. But it’s worn off.

It’s not just AIs that are suffering a language problem. Most bureaucratic presentations feature the ‘personal’ approach. Instead of statements of information, we now have “How can I become a genius” followed by instructions of how to become one, and then “What happens if I become a moron” followed by instructions pertaining to that alternative. The bureaucratic form asks in the first person singular so that morons can identify with the subject matter, or so one must assume the reasoning goes. It’s insulting, and it’s disgusting. No wonder the AI ‘trainers’ have got off on the wrong foot.

AI trainers have left the stage

AI ‘trainers’ are now gradually being replaced by the machines ‘training’ themselves. One can only speculate as to what effect that has on how the machines ‘talk’ to us. They will presumably adapt their ‘language’ to continuing human/AI communication. There is also the prospect of more ‘bot’ communication which has already exceeded human communication with the internet. Bot requests on search engines now outnumber human ones. Increasing inter-bot communication is likely to impact AI language ‘learning.’ It has even been suggested that inter-bot communication will no longer be human language, but code. If that occurs most of us, possibly excepting code experts, will be excluded from most AI communication. We’ll be like the German congregation going to church in the 16th century, not understanding a word of the priest who was preaching in Latin, before Martin Luther nailed the German translation of his 96 theses on the church door in 1517.

Martin Luther gives language to the people with hammer and nails at Wittenberg Castle Church

Bots have already created some of their own language, although still in English or various translations. They are reported as cracking feeble jokes such as “I like your algo rhythms.” Given the resources of the entire internet it’s possible that they merely repeated a human’s joke. If asked, an AI can act as a Master-of-Ceremonies for all kinds of functions, including comedy shows.

With the advent of agentive IAs, i.e. AIs making unsupervised decisions and acting on them, the AI/bot traffic is bound to increase. How that communication will take place and what kind of language will evolve is difficult to predict. With less human input to AI it is possible that humans could become more and more alienated, chiefly due to the increasing language differential. According to some studies this is already happening. Scientists at Oxford University, MIT, and the University of California found that AI degrades cognitive function and problem-solving abilities in a relatively short period.

“We show that just 10–15 minutes of AI interaction can result in significant impairments in independent performance and persistence — capacities that are foundational to lifelong learning,” the study said. “If brief exposure produces measurable erosion, the cumulative effects of daily AI use over months or years may be profound and difficult to reverse.”

It appears that AI ‘interaction’ can be problematic. Given that such interactions are principally based on language, it would seem that that could be where the problem lies. Instead of a communication, the language is a barrier. It has to be neutralised, surmounted or ignored, and the communication stripped down to its bare facts, before becoming a palatable AI response.

I find that the first thing I do when copying an AI response on to ‘Word’ is eliminate all bullets and numbers. They simply distract. I then go through the response erasing all personal references. They not only distract and mislead, they insult. I then check to see whether there’s anything useful, and copy and paste that to the end of the document. That’s usually about 5% or less. Then I delete all else. In this way I’m keeping the AI at arm’s length, but still making use of its resources.

I don’t know if anyone else is using similar methodology, but if people need to use that kind of process it implies that there’s a serious language issue.

One obvious problem is the left-over one of the machines trying to be human. The companies that run these machines needed to market them, which meant they had to be ‘relatable’ to humans. But the machine is not a human. It is not an ‘I’ and it’s not appropriate to address me as a ‘you.’ Take the ‘Methodology’ entry in Wikipedia, for example. It just states: “Methodology is the study of research methods. However, the term can also refer to the methods themselves or to the philosophical discussion of associated background assumptions.” And so on. No “Nice to see you again or how are you today.” It’s information issued in ways that we are used to. And it’s issued in the language that is appropriate to internet search methodology. It is culturally appropriate. AI has no identifiable cultural propriety.

Human language, as opposed to the current AI language, has been debated at least since ancient Greek civilisation. Thinkers such as Jean-Jacques Rousseau (1712–1778) have argued that language originated from emotions, while others like Immanuel Kant (1724–1804) have argued that languages originated from rational and logical thought. Twentieth century philosophers such as Ludwig Wittgenstein (1889–1951) argued that philosophy is really the study of language itself. Major figures in contemporary linguistics include Ferdinand de Saussure and Noam Chomsky.

Humans acquire language through social interaction in early childhood, and children generally speak fluently by approximately three years old. Language and culture are codependent. Therefore, in addition to its strictly communicative uses, language has social uses such as signifying group identity, social stratification, as well as use for social grooming and entertainment.

Since AI has not been ‘brought up,’ it has totally missed out on any ‘childhood training.’ It might be fluent but it has absolutely no equivalent of other human development such as normally found in humans by the age of three.

The idea of language being discussed

During the Age of Enlightenment, 1687–1789, there was much speculation about the origin of language. Thinkers such as Rousseau and Johann Gottfried Herder argued that language had originated in the instinctive expression of emotions, and that it was originally closer to music and poetry than to the logical expression of rational thought.

Well, we’re no longer in that age. But whoever creates machines that use human language as an essential tool would do well to study the idea of language and what it means and has meant in human terms. It’s not a just a code, and it’s not just a transmission. It’s an implicit form of existence.

If our current AIs are left to keep their inappropriate language and AI/bot communication increases, further separation from human language is likely. There is an imminent need for AI ‘trainers’ to use whatever control over their AIs is left available to them to revise and adjust their machines’ language so as to be relevant to us human beings.