Thoughts on AI: 2024 Edition

AI 😠

My criticism for generative AI is relatively simple: whenever I have tried to use AI to accomplish a complex task, it’s failings become painfully obvious. It is still very narrow technology.

I know AI Bros usually revert to playing their trump card; “This is the worst the tech will ever be” but their logic is flawed. We’re not saying complex tasks will “never” be possible with AI - never is a long time - we are saying “not now” and “not soon”.

It would save me a lot of time if I could feed blog subjects into Chat GPT and have it spit out blog posts for me. But it doesn’t work that way. You get a result, sure. Someone who has never heard of Chat GPT would believe a person wrote it. But it’s formulaic and dull. It is either boringly correct (like Quora bot answers) or the AI hallucinates and gives you something totally made up (like searching for a song using Chat GPT only for it to invent a song that doesn’t exist).

If your goal is to replace a writer with Chat GPT, you already know the limitations, even if you can’t list them as bullet points. It boils down to: It doesn’t work.

Apple and Uber gave up developing driverless cars with AI because it couldn't be done. Amazon gave up developing "just walk out" shops run by AI because it couldn't be done.

The machine learning models are amazing, they are a leap forward, they are changing the world, but clearly there’s a pretty hard limit to what the models are capable of; they are not bringing about the End of Days like AI bros claim they are.

When I asked Benjamin Field, expert in AI regulation, how big a piece of regurgitated material is, his response was “recognisably big”. For example, an AI that had been trained on Shutterstock data was asked to create a tree. It produced a tree with a fragment of the Shutterstock watermark on it.

I’m sure newer models will use smaller pieces. But they remain pieces of existing material. There’s no getting around that.

AI ☺️

There are some uses of AI which are fantastic. Some which might disrupt a few jobs.

For instance, being able to much more quickly remove a green screen from a video, or even a background not shot on a green screen. This is disruptive, but it’s only as disruptive as other advances we lived through and survived, such as high-quality cameras on iPhones. You can pick one specific job in one specific business and argue iPhones having high quality cameras was a disaster (the person who developed holiday photos at your local shop is out of a job) but clearly the economy coped with the transition.

Creatives are using AI to do things which previously couldn’t be done, such as this Coca-Cola ad (which still had a cast, crew, et al, just advance CGI to bring the paintings to life).

So, while generative AI is an impressive, useful, innovative and even revolutionary technology, it is not a replacement for human creativity. It won't replace artists any more than computer animation replaced artists. AI’s limitations are hard-baked into it; Generative AI cannot, by definition, produce truly original work from scratch. But all this discussion takes oxygen away from the really good things AI can do well, across a broad range of industries.