Ethan Mollick is one of the most provocative researchers on how AI will transform our day-to-day jobs, and his new book 𝐶𝑜-𝐼𝑛𝑡𝑒𝑙𝑙𝑖𝑔𝑒𝑛𝑐𝑒 rounds up his best thinking. Here are 3 striking quotes – and my quick reflections on what they imply for healthcare.
𝟏. “𝐆𝐞𝐧𝐞𝐫𝐚𝐥 𝐏𝐮𝐫𝐩𝐨𝐬𝐞 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐢𝐞𝐬 [𝐬𝐮𝐜𝐡 𝐚𝐬 𝐀𝐈] 𝐭𝐲𝐩𝐢𝐜𝐚𝐥𝐥𝐲 𝐡𝐚𝐯𝐞 𝐬𝐥𝐨𝐰 𝐚𝐝𝐨𝐩𝐭𝐢𝐨𝐧, 𝐚𝐬 𝐭𝐡𝐞𝐲 𝐫𝐞𝐪𝐮𝐢𝐫𝐞 𝐦𝐚𝐧𝐲 𝐨𝐭𝐡𝐞𝐫 𝐭𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐢𝐞𝐬 𝐭𝐨 𝐰𝐨𝐫𝐤 𝐰𝐞𝐥𝐥.”
Consider the internet, which took decades to transform our world as individuals, organizations, and markets figured out how to use it effectively. Similarly, healthcare will need many years to adapt our practices for AI. We'll need to do the hard work of implementing AI in every exam room, every back office, and every C-suite.
But when we're done ... wow, what a different world we'll live in. As Amara's Law puts it, "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."
𝟐. “𝐖𝐡𝐞𝐧 𝐰𝐞 𝐮𝐬𝐞 𝐀𝐈 𝐭𝐨 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞 𝐨𝐮𝐫 𝐟𝐢𝐫𝐬𝐭 𝐝𝐫𝐚𝐟𝐭𝐬, 𝐰𝐞 𝐭𝐞𝐧𝐝 𝐭𝐨 𝐚𝐧𝐜𝐡𝐨𝐫 𝐨𝐧 𝐭𝐡𝐞 𝐟𝐢𝐫𝐬𝐭 𝐢𝐝𝐞𝐚 𝐭𝐡𝐚𝐭 𝐭𝐡𝐞 𝐦𝐚𝐜𝐡𝐢𝐧𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐞𝐬, 𝐰𝐡𝐢𝐜𝐡 𝐢𝐧𝐟𝐥𝐮𝐞𝐧𝐜𝐞𝐬 𝐨𝐮𝐫 𝐟𝐮𝐭𝐮𝐫𝐞 𝐰𝐨𝐫𝐤.”
This is an example of automation bias: deferring to automated decision-making systems, rather than using our own judgment. It's particularly hazardous in healthcare, where, for example, a radiologist who’s presented with an AI-generated diagnosis might not bother to second-guess the AI’s output.
When I use AI tools such as ChatGPT in my own work, I try to avoid automation bias in a few ways: (1) by pausing to consider my own perspective before asking the AI to weigh in; (2) by asking the AI to help me think through a topic, rather than offering its own solution; and (3) by asking the AI to give me 5-10 diverse perspectives to consider, rather than just one.
𝟑. “𝐀 𝐥𝐨𝐭 𝐨𝐟 𝐰𝐨𝐫𝐤 𝐢𝐬 𝐭𝐢𝐦𝐞-𝐜𝐨𝐧𝐬𝐮𝐦𝐢𝐧𝐠 𝐛𝐲 𝐝𝐞𝐬𝐢𝐠𝐧…. [𝐂𝐨𝐧𝐬𝐢𝐝𝐞𝐫] 𝐭𝐚𝐬𝐤𝐬 𝐰𝐡𝐨𝐬𝐞 𝐟𝐢𝐧𝐚𝐥 𝐰𝐫𝐢𝐭𝐭𝐞𝐧 𝐨𝐮𝐭𝐩𝐮𝐭 𝐢𝐬 𝐢𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭 𝐛𝐞𝐜𝐚𝐮𝐬𝐞 𝐢𝐭 𝐢𝐬 𝐚 𝐬𝐢𝐠𝐧𝐚𝐥 𝐨𝐟 𝐭𝐡𝐞 𝐭𝐢𝐦𝐞 𝐬𝐩𝐞𝐧𝐭 𝐨𝐧 𝐭𝐡𝐞 𝐭𝐚𝐬𝐤 𝐚𝐧𝐝 𝐨𝐟 𝐭𝐡𝐞 𝐭𝐡𝐨𝐮𝐠𝐡𝐭𝐟𝐮𝐥𝐧𝐞𝐬𝐬 𝐭𝐡𝐚𝐭 𝐰𝐞𝐧𝐭 𝐢𝐧𝐭𝐨 𝐢𝐭 – 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐫𝐞𝐯𝐢𝐞𝐰𝐬, 𝐬𝐭𝐫𝐚𝐭𝐞𝐠𝐢𝐜 𝐦𝐞𝐦𝐨𝐬, 𝐜𝐨𝐥𝐥𝐞𝐠𝐞 𝐞𝐬𝐬𝐚𝐲𝐬, 𝐠𝐫𝐚𝐧𝐭 𝐚𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬, 𝐬𝐩𝐞𝐞𝐜𝐡𝐞𝐬, 𝐜𝐨𝐦𝐦𝐞𝐧𝐭𝐬 𝐨𝐧 𝐩𝐚𝐩𝐞𝐫𝐬.”
One such task in healthcare might be prior authorization, which physicians have long claimed is used to gatekeep expensive care. In this view, if AI made it easier for physicians to file prior authorization requests, insurers might replace them with a different hurdle – so the payoff from automation might be lower than anticipated.
What other intriguing books or articles on AI have you been reading lately? Any quotes stand out as particularly provocative?