Is Generative AI Becoming a Religion?

Lack of certainty about the future of this technology is dividing the world into zealots and sceptics. Moderates seem harder to come by nowadays.

My position on generative AI is best described as moderate. I have no doubt that recent developments and releases have transformed what it is possible to create without any form of human involvement, but I am also acutely aware of the current shortcomings of generative AI, the risks it poses and the uncertainties surrounding its future development.

As someone who posts frequently on social media about analytics, data science and AI topics, I am noticing a trend when I post about generative AI which I don’t see when I post about other topics like statistics or methods. I am seeing a greater level of anger and intolerance in response to any suggestion that models like chatGPT may not ascend to be the only dominant technology of the next 30 years or so. Why are such strong feelings being evoked on this topic?

My moderate position on generative AI

I was in the earliest group of people to try out chatGPT when it was released last year. I was amazed. Having experimented a lot with prior incarnations like GPT-3, there was a huge uplift in the quality of the response of the model to prompts. There was also a massive improvement in the model’s ability to handle multiple languages, including programming languages. It was, without a doubt, a revolutionary moment in the development of generative AI.

But I was never under any illusions about what models like chatGPT actually do: they generate language and that’s all they do. What really impresses people about these models is the quality of the natural language they generate. ChatGPT was the first model that was able to produce reliable, readable and often stylish and engaging natural language at scale. This has a ton of applications. It can save people time and effort in writing, in summarization or translation, in idea generation and in basic programming.

Some people may be fooled into thinking that generative AI is capable of actual reasoning. It is not. If you give it a problem to solve, the quality of the solution depends…