Jagged Frontier - part 1

I have been having conversations about how academics and professionals are using AI day to day, and it has reminded me of the importance of the 𝐣𝐚𝐠𝐠𝐞𝐝 𝐟𝐫𝐨𝐧𝐭𝐢𝐞𝐫 concept.⁣
⁣
This idea, from a Harvard paper, describes the uneven and unpredictable boundary between what AI can and can’t do well. The “jagged” part is key: 𝐭𝐰𝐨 𝐭𝐚𝐬𝐤𝐬 𝐭𝐡𝐚𝐭 𝐟𝐞𝐞𝐥 𝐞𝐪𝐮𝐚𝐥𝐥𝐲 𝐡𝐚𝐫𝐝 𝐭𝐨 𝐚 𝐩𝐞𝐫𝐬𝐨𝐧 𝐜𝐚𝐧 𝐟𝐚𝐥𝐥 𝐨𝐧 𝐨𝐩𝐩𝐨𝐬𝐢𝐭𝐞 𝐬𝐢𝐝𝐞𝐬 𝐨𝐟 𝐭𝐡𝐞 𝐟𝐫𝐨𝐧𝐭𝐢𝐞𝐫. AI might nail one and flounder on the other - but the 𝐢𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐢𝐧𝐠 𝐩𝐚𝐫𝐭 𝐢𝐬 𝐭𝐡𝐚𝐭 𝐢𝐭 𝐜𝐚𝐧 𝐛𝐞 𝐝𝐢𝐟𝐟𝐢𝐜𝐮𝐥𝐭 𝐭𝐨 𝐩𝐫𝐞𝐝𝐢𝐜𝐭 which side of the frontier a task falls.⁣
⁣
That unpredictability means we can’t reliably judge AI's usefulness by experimenting with just a few (or unknowingly biased) tasks. 𝐖𝐞 𝐫𝐢𝐬𝐤 𝐰𝐫𝐢𝐭𝐢𝐧𝐠 𝐢𝐭 𝐨𝐟𝐟, 𝐨𝐫 𝐭𝐫𝐮𝐬𝐭𝐢𝐧𝐠 𝐢𝐭 𝐭𝐨𝐨 𝐦𝐮𝐜𝐡, based on a distorted picture.
⁣
I’ll dig into this more in upcoming posts, including how the authors frame navigating this frontier as working like a “Centaur” or a “Cyborg”.⁣

You can find part two here and part three here