Generative AI fails to be truly creative, but it's also utterly unneccessary
Modern AI software has been promised as a transformative technology. And while it's improved in the last few years, it has failed so far to deliver on its lofty goals. Why is that?
When we say AI, we usually mean modern generative AI technologies. Trained on huge volumes of images, writing, or computer code, they can spit out their own images, words, or code almost instantly. The rate of improvement has been staggering in a short time.
Sure, the people in AI video still seem like they might pull off their skin to reveal the steel skulls of a Terminator, but they can pass for real for a minute or two.
Yet it's still possible (even now that they don't feature extra fingers) to identify most AI-generated content. The images have a false, glossy sheen, the text is obsequious and annoyingly ingratiating, like bad marketing copy. Code has to be carefully watched to ensure it isn't buggy.
Maybe it will all improve, maybe it will be indistinguishable from human-produced writing, images, and video in a decade. Or maybe not - a recent study by OpenAI itself suggests that the "hallucinations" of bad info that text generators are prone to is baked in, mathematically.
The computer code seems to be the most useful application. But a recent study found that while 90 per cent of people in the industry, only 24 per cent trusted it, using it mostly for short snippets of code or tests. The consensus seems to be that it's useful, but only if overseen by someone with experience.
As for the rest, how is a flood of (mediocre) text and images supposed to change the world, exactly?
There was already a surfeit of that stuff! People love doing creative work so much they do it for starvation wages or as a hobby. The internet is flooded with entirely human-created artwork, book, and magazine publishers have more submissions than they know what to do with, and young would-be filmmakers are always trying to break into the industry.
At the end of the day, businesses have to accomplish things in the real world. Creating a flood of images, email spam, and marketing copy falls short of that. Endless data slop is not a revolution.
Generative AI can't generate electricity, it can't grow food, it can't treat a disease.
I'm not being cynical, there are real, incredible breakthroughs in those areas either happening now, or on the horizon!
Solar power and batteries are getting cheaper and better every day.
Fields like genetic engineering and cellular agriculture promise to do everything from produce proteins straight from water, air, and trace nutrients, to boosting the growth and carbon capture efficiency of plants.
In medicine, we've seen new treatments or preventative therapies announced for everything from Huntingdon's disease to HIV to simple baldness, just this year.
Future versions of AI will prove useful. One day, they'll build robo-taxis that can handle snow, androids that can sew and do carpentry, computers that can research new medicines with minimal human input.
But automating creative work? It's a dead end, and far from a revolution.