Worker-in-the-Loop: A Retrospective

I managed an AI engineering team with HITL (in this case WITL) in place and my experience mirrors the article. A worthy read for anyone considering entering the space.

We all want the efficiency of AI, but it does take a lot of human effort to achieve the required accuracy first.

Re: Worker-in-the-Loop: A Retrospective

#5

Post by gumby » Tue, Jan 07, 2020, 7:51 PM UTC

everdev wrote: post #4

I managed an AI engineering team with HITL (in this case WITL) in place and my experience mirrors the article. A worthy read for anyone considering entering the space. We all want the efficiency of AI, but it does take a lot of human effort to achieve the required accuracy first.

This is true of human intelligences as well. Faster than ML but doesn't really scale. HITL is a way to make it scale.

Re: Worker-in-the-Loop: A Retrospective

#6

Post by akilian » Tue, Jan 07, 2020, 8:41 PM UTC

dmessing wrote: post #3

Great resource thanks for sharing.

Appreciate it!

Re: Worker-in-the-Loop: A Retrospective

#7

Post by akilian » Tue, Jan 07, 2020, 8:42 PM UTC

gumby wrote: post #5

everdev wrote: post #4

I managed an AI engineering team with HITL (in this case WITL) in place and my experience mirrors the article. A worthy read for anyone considering entering the space. We all want the efficiency of AI, but it does take a lot of human effort to achieve the required accuracy first.

This is true of human intelligences as well. Faster than ML but doesn't really scale. HITL is a way to make it scale.

This is right on; but it is still difficult to pull off in practice. I expect that 5-10 years from now it will be somewhat standard/turnkey.