2X Founder & CEO @ Tensor Planet | Transforming Waste Management | MIT SMR Columnist | TEDx Speaker
When humans work alongside AI, how do you keep them motivated? Try designing a less-than-perfect AI! It’s not enough to build the best AI for a problem. Leaders must plan their collaboration with humans in the workplace. Susan Athey, a Stanford Professor talks about 4 scenarios of AI usage - few where it will be in control, and one where it will be paired up against humans! 1. Replacement AI: Tasks where AI is almost perfect and can take over a job. 2. Imperfect AI: Tasks where AI makes decisions but needs constant correction by humans. With known error rates, humans are in the loop, keeping them alert and motivated. 3. Augmentation AI: Tasks owned by humans. AI is pretty good, but plays the role of assistant. Ex. recommendation systems. 4. Antagonistic AI: Similar to #3, but the AI is designed to highlight decisions where it doesn't agree. Humans are in control, but they review and respond to conflicting suggestions. This is an interesting framework to designing systems that use AI with differing abilities and levels of control. What do you think of this approach? Which of these 4 AI designs is practical in your organization today? #data #analytics #business #artificialintelligence (The FastCo article link is in the comments)
Good strategy! Is assisted AI different from imperfect AI?
If historical experience with technology breakthroughs over the past several hundred years is any guide, the type of "AI" [sic -- I personally don't like that term] that will win out is "partial replacement", where human and machine work together. In other words automation will take over all operations that don't require intuition, emotion, or wisdom, tasks which will continue to be performed by humans. One consistent (human) flaw that I encounter is the belief that any job currently performed by humans can be 100% automated, perfectly. Following such a thought pattern leads systematically to "less than perfect AI" but /without/ the humans still in the loop.
Top Voice | Training Datasets and workflows for AI Agents
4 年Thanks for sharing this article Ganes Kesari . It makes a lot of sense for us where we are leveraging AI via AI assistant labeling framework!