[mullenba]

I've consulted with some big companies on AI strategy. I tell them there are two approaches to AI.

1) Train AI to replace human work. This gives you 50% quality for 10% cost. 2) Train AI to assist human workers. This gives you 200% quality for 110% cost.

Most companies will go with option 1, and it's a race to the bottom. Eventually, someone will go with option 2 and gather up all of the pieces and take over the market.

[altmanaltman]

Do you have any basis for these numbers or claims or just took it out or your backside and presented it as facts?

What companies do you consult and on what

[mullenba]

Look at Stitch Fix. Initially, they used AI to sort through their offerings and present options to human stylists. The humans were free to make changes however they felt was necessary. Then they brought in a new CEO from Amazon and fired most of the stylists to use straight AI. Satisfaction went down, revenues went down, and the company as a whole suffered greatly.

[i5heu]

This is not at all how all of this works.

If you train an AI in one thing it will become better in the other.

[mullenba]

That's not true at all. They're entirely different problems. I worked with a global manufacturing company where we were able to cut plant stoppages for custom products by 2/3. When we used AI to handle the repetitive issues in service of human experts, it freed up the humans to see longstanding issues with data and plant processes that nobody had noticed before. Simply replacing the humans wouldn't have given nearly the same benefits.