ChatGPT took our jerbs!

TLDR: If AI systems achieve Artificial General Intelligence, then losing our jobs will be the least of our worries. We’re a long way away from bowing down before our robot overlords.

There’s been a lot of panic around AI recently. Not so much that we’ll be getting blown away by Arnold Schwarzenegger clones but rather that we’ll be replaced at our desks by HAL 9000. It seems that the take over of middle level white collar jobs by AI software is imminent and we’ll be seeing white collar redundancies much in the way that many blue collar jobs were replaced by mechanical automation. Before we go too far down the headless chicken panic route, it makes sense to try to understand what the current iteration of AI or rather ML (Machine Learning) is and what its capabilities might be.

Generative AI like ChatGPT is basically autocomplete on steroids. It’s not magic and it’s certainly not true intelligence. LLM systems like ChatGPT have no understanding of what they are saying. Not only do they not know what they seem to know, they don’t even know what it is to know. There is no true understanding of knowledge to be found there. All they do is statistically guess which is the next word or token to apply after the current one, based on the input given to them. That’s it.

The wonder is that such a system is still able to generate coherent and fairly useful output. If there’s any magic here, it’s this.

Generative AI is certainly ground breaking and it will definitely be disruptive. Any kind of job which is scriptable was already at risk and will now be even further at risk. However, wherever there is any kind of human element involved; any kind of judgement call to be taken or social context to be applied, there will still be a need for human intervention. AI will do a better job of screening X-Ray plates for signs of cancer, but you will still need a human doctor to give it a final pass before treatment can begin.

The best way to think of the current state of AI is that it’s like working alongside a very helpful - and occasionally brilliant - junior assistant. He can do many straightforward jobs reasonably well but still requires supervision. The sheer quantity of data he can churn through is impressive, but don’t expect any leaps in understanding.

So where is the danger? If we achieve Artificial General Intelligence, which is true artificial intelligence, then all bets are off. If we can create an AI which is say 10% brighter than us, then that AI would itself be able to create another AI which would be 10% smarter than itself and so on. Within a very short period of time, we would be dealing with seemingly god-like intellects with capabilities we cannot even fathom.

At that point, the fact that the AI godling would be capable of replacing you at your desk would be the least of anyone’s worries.