The great AI delusion is falling apart

I’m not sure if the whole AI thing is falling apart, but we are definitely in a phase where the hype doesn’t match what we are seeing in the real world.

This article provides some interesting food for thought:

https://www.msn.com/en-us/news/technology/the-great-ai-delusion-is-falling-apart/ar-AA1IyRwc

In a randomised controlled trial – the first of its kind – experienced computer programmers could use AI tools to help them write code. What the trial revealed was a vast amount of self-deception.

“The results surprised us,” research lab METR reported. “Developers thought they were 20pc faster with AI tools, but they were actually 19pc slower when they had access to AI than when they didn’t.”

This reminded me of something I’ve long wondered about some of the measurements provided by fans of AI – the processes that take so much less time than they used to. Are they looking at the whole picture? I know that sales demos do not. They’ll claim to have created an entire business plan in 15 minutes using AI, while waving away any questions about how much time it would take someone to review, edit, and correct the business plan in question. Creating a business plan the old-fashioned way requires research, writing, rewriting, confirming details, and so on. With a couple of prompts, I have one, but will I then put in the same effort to verify that the draft pushed out by my AI tool is accurate and workable for my business?

If so, where is the time saved?

To put it more simply, here’s an example from another article:

Does AI actually boost productivity? The evidence is murky.

The common narrative around AI and productivity is that AI automates mundane tasks, making us faster at doing things and giving us more time for creative pursuits. This, however, is a naive view of how work happens.

Just because you can deal with your inbox more quickly doesn’t mean you’ll spend your afternoon on the beach. The more emails you fire off, the more you’ll receive back, and the never-ending cycle continues.

This is the question. If AI helps you complete tasks faster, but those tasks generate more work, are you being more productive? How do we measure that? To continue the email example, if AI quickly drafted an email response to a question and I answered the question faster than usual, but the answer wasn’t as straightforward as it could have been, so there was an additional back-and-forth of 3-5 emails, did it make me more productive? Would taking a few moments to craft an email that answered the question without the need for further back and forth have been more productive?

Yes. That would be more productive. The larger task, answering the question, would have taken less time, even as the individual step of responding to the email was quicker. That’s what I want to hear. I don’t want to know about the individual processes that you’re saving time on. I want to see the whole picture. Please show me the process to take that AI-generated end-product and verify that it’s accurate before it goes in front of a client or is published online, or filed in a court.

That’s how we verify whether AI is making us more productive, and I suspect the lack of this is why we aren’t seeing the massive upticks in overall productivity that the AI fans have been promising.

Read both of the links above. There are areas where AI is helping people become more productive, but there are also many areas where the promised productivity has yet to materialize. When companies have invested hundreds of billions of dollars in AI infrastructure, small productivity gains are going to leave them holding a huge bag. I fear that is not going to end well if world-changing AI productivity doesn’t happen, and I fear what kind of world we’re creating for workers if it does.

Follow these topics: Artificial Intelligence