The AI and I

Author's Note: I think this article is the topic I'm asked about most, so I pulled one together. I hope you enjoy it.

Those close to me will know that I'm generally a tech sceptic. Even more so with the so-called “Gen AI” that we have now. This scepticism in AI technology was grounded in my research for my undergraduate dissertation on Artificial Intelligence and Technological Unemployment.

I had chosen this topic during the previous summer - summer 2020. I had been speaking with my dissertation tutor about my topic; I was interested in dystopia and exploring humanity through the medium of science fiction (that’s a whole other story). She had advised me that it would be good to choose a different adjacent topic for the dissertation, so I settled on the Apocalypse - naturally. I had spent that notorious summer researching different apocalypses and deciding which one I felt most likely and most drawn to write about.

I landed on AI majorly disrupting the jobs market - I guess I wasn’t far off the mark on that one.

However, the types of AI I was writing about at that time were not necessarily the large language models (LLMs) that we see everywhere today. I was writing about the autonomously self-improving algorithms (defined by J Fletcher 2018) already employed by social media, predictive analytics and surveillance capitalism. By this point, we have been surrounded by predictive algorithms for the better part of 20 years. Machine Learning (ML), neural networks and deep learning (DL) have been in development since the 1970s. It isn't exactly a new story.

I have several major concerns/issues with what is branded as AI.

Magic Economics.

As far as I can see, the novelty of ChatGPT, Claude, Copilot, etc. is the scale at which they can be deployed. Before 2022, predictive algorithms were expensive and/or inaccessible due to the hardware and computing requirements. This confined their use to large corporations - big tech companies and asset-intensive industries - where the return for implementing these technologies could offset the huge costs of training and maintaining such models.

We now have a supposedly much more financially accessible set of predictive algorithms. They are affordable, fast and easy to deploy, because the computing is done remotely. I am again sceptical about this. In design, we are taught that something can be cheap, fast or high quality, but it can only ever be 2 of the 3. If what companies pushing these technologies tell us is true, they would have us believe these ‘miraculous machines’ are fast, cheap, and improving in quality exponentially. I suspect someone somewhere is lying.

Mis-labelling and Overselling

This is one of my big issues with what is sold as AI today: some present it as a matrix-style doomsday, while others frame it as the turning point that moves the whole world toward a futurama-esque utopia. It will either change the world for the better or the worse, but in an unrecoverable, inevitable and unavoidable way.

First, nothing here is inevitable. That’s not to say it is going to disappear, but it isn’t a guarantee that LLMs are here to stay in their current forms and use cases.

Second, LLMs are really good at some things. I am using Grammarly Go to help me draft this blog. I have dyslexia and believe me, this would not be an easy read if I did it unassisted. It is much better than I am at spotting my spelling and grammar mistakes and giving me reader feedback. It makes my writing at least average - I’m hoping you are here for the witty commentary rather than my SPAG accuracy. However, the ideas you are reading about, the style I’m writing in and the turn of phrase are all mine. I wouldn’t trust a machine to write an opinion piece, because, to start with, it doesn’t have an opinion, and it would always be regurgitating someone else’s stance.

Let's be real: these machines can do some helpful functions that add value, but they are not a panacea for intelligence/creativity-baseded task automation. As far as I know that doesn’t exist - it certainly isn’t going to be done by a machine that would know if creativity hit it with a sledgehammer - because, you know, it isn’t self-aware ;)

AI is a catch-all term for specific types of machine learning. What is pitched as Gen AI is an LLM. It is AI in the sense that it is an autonomous, self-improving algorithm. But it isn’t going to become a C3PO or Cortana. It's a text predictor (or image, video or audio predictor). It doesn’t understand what it produces. Just today, I had to correct Google AI’s maths. Why? Because it just generates statistically likely word sequences. That is why you see so many memes on the internet about some LLM failing to count the number of r’s in strawberry - because it can’t count, and it doesn’t know what an r is.

When you view it like this, it is clear that it isn’t going to autonomously design the next version of the International Space Station, or create world peace. I don’t even know if it is possible for it to become ‘smarter’ than humans - for starters, what does that mean? Computers and humans learn and process information really differently, so it isn’t an easy apples-to-apples comparison.

Deception by Design

Your glorified chatbot isn’t sentient, all-knowing, or even intelligent. The way they are programmed to speak to their users is designed so that we imbue them with human-like status. The mind is easily deceived. We naturally want to anthropomorphise. We do with pets, cars, and toys. If it speaks like a person, it must be a person and have agency (the ability to make decisions based on wants and desires). That is why so many people believe these algorithms are sentient. Our brains miscategorise these systems because they behave in ways analogous to humans - particularly the way they speak. This is why very intelligent people in the field can be misled. I believe this design feature of these algorithms and their interfaces to be the single biggest danger “AI” poses to people. To such an extent that if I could bring in one piece of legislation overnight, it would be to make it illegal for machines to attempt to imitate human-to-human communication or to present themselves in human likeness.

There are other areas that I could write about (and I may do at a later date), but I’d be interested to know your thoughts on the topics above. Any rebuttals you may have, questions or experiences you wish to share.