About fifteen years ago, I was leading tracks at Israel's national L&D and HR conferences, and later I served as chairperson of the L&D conference. I wanted session formats that were anything except a sage on the stage. We ran Pecha Kucha sessions, speed-dating between participants, a genius bar where people brought their questions to experts for 15 minutes of guidance, and round tables where participants built shared knowledge artifacts together. We called it the Crowd Space.
What can an ox teach us about crowds?
During the same period, in late 2011, Lior Zoref asked his Facebook and Twitter friends to help him build a TED talk about crowd wisdom. Lior is an Israeli speaker who later wrote the book Mindsharing and did his PhD on crowd wisdom. I was one of many people who contributed ideas. One of the best came from Or Sagy, then 16 years old, who suggested recreating the original crowd wisdom experiment. In February 2012 Lior did exactly that: he brought a live ox onto the TED stage in Long Beach and asked the audience to guess its weight. More than 500 people sent in a guess. Their average was 1,792 pounds, only three pounds away from the ox's real weight of 1,795 pounds.
That experiment goes back more than a century. In 1907, the statistician Francis Galton analyzed 787 guesses of an ox's weight, collected at a livestock exhibition in Plymouth, England. The guesses averaged 1,197 pounds, exactly what the ox weighed.
Crowd wisdom, or the wisdom of crowds, is the idea that the combined judgment of many independent people often has a better chance of being right than any single person in the group.
What does crowd wisdom mean for Are Right, A Lot?
Amazon's Are Right, A Lot leadership principle reads:
Leaders are right a lot. They have strong judgment and good instincts. They seek diverse perspectives and work to disconfirm their beliefs.
This is my third post on this principle, after When AI can't tell you you're wrong (issue #4) and Scenario planning for AI futures (issue #28). Here I focus on its last sentence: seeking diverse perspectives, and working to disconfirm your own beliefs.
Amazon's CEO, Andy Jassy, describes one way to practice it. In an Inside Amazon video on the leadership principles, he says that as leaders "we speak last in the room", because they want to hear everybody's input first.
Speaking last is a crowd wisdom practice. When the most senior person in the room speaks first, others may anchor on that view, and the group loses the independent perspectives that make a crowd wise. When you seek diverse perspectives and hold your own view until the end, the people around you can act more like Galton's crowd, and the group has a better chance of being right than any single person in it, including you.
What does a crowd need to be wise? In his 2004 book The Wisdom of Crowds, the journalist James Surowiecki collected stories like Galton's and described four conditions:
- Diversity of opinion - each person has some private information of their own.
- Independence - people's opinions aren't determined by the opinions of those around them.
- Decentralization - people can draw on their own local knowledge.
- Aggregation - some mechanism turns many private judgments into one collective decision.
Each of these conditions has a place in Are Right, A Lot. Seeking diverse perspectives brings in diversity and local knowledge, speaking last protects independence, and your own judgment does the aggregation when you turn what you heard into a decision.
What does this have to do with AI?
Lior's TED talk also showed people using their online crowd for much more than guessing weights. One of them, Francine, was raising her son with daily help from crowd wisdom, and said it felt like having a "super-nanny". Another, the American writer Deborah Copaken, posted photos on Facebook when her four-year-old son had a rash and a fever. Doctors took it for strep, but people from her network helped identify that he actually had Kawasaki disease, and they turned out to be right. She later told the story in Slate, in How Facebook saved my son's life.
Today, fourteen years later, many of these questions seem to go to a chatbot instead. A 2026 survey by Lurie Children's Hospital found that 81% of the US parents surveyed had used AI to help with parenting tasks, and 31% had followed AI advice over their own instincts.
One way to think of large language models is as a compressed version of that crowd. They are trained on, and can search, what many people wrote in books, articles, websites and forums, including many of the health and parenting questions people once asked their friends. That may be one reason the models are so often right on questions that many people have written about.
So how can you be right a lot with AI?
To get AI to help you seek diverse perspectives, I suggest treating it like a crowd that needs Surowiecki's conditions. Ask more than one model, or the same model in separate chats, so one answer doesn't shape the next. Share what you currently believe and ask what would make you wrong, not just what it thinks of your idea. When you get one confident answer, ask where people who know the domain would disagree, and then take that disagreement to some of those people. And when you discuss the decision with your own team, do what Jassy describes: share your view AFTER you have heard everyone else.
This issue's framework post turns these ideas into a step-by-step method, in Ten meetings in an afternoon. The term of the week, AI ensemble, names the practice of asking several models and combining their answers.
Go deeper
For a fun way into the topic, watch Lior Zoref's 2012 TED talk.
Your action step
Pick one decision you are working on this week. Ask two models about it in separate chats, tell each one what you currently believe, and ask what would make you wrong. Then, when you discuss it with your team, share your own view last.
When was the last time a colleague, or an AI model, disagreed with you and changed your mind about a decision?