In this compilation episode of The Parlor Room Presents: Hello AI, host and Harvard Business School Online Creative Director Chris Linnane explores one of the biggest challenges in artificial intelligence: knowing when—and when not—to trust AI.
Featuring HBS professors Linda Hill, Felix Oberholzer-Gee, and Jake Cook, the conversation examines AI adoption, organizational culture, and the critical role of human judgment. From AI hallucinations and overconfidence to experimentation and decision-making, this episode offers practical insights to avoid common AI mistakes in the workplace.
Resources
Catch up on previous episodes of The Parlor Room, featuring faculty from this compilation episode:
You can also watch The Parlor Room Presents: Hello AI on YouTube.
Transcript
Editor's Note: The following was prepared by a machine algorithm and may not perfectly reflect the interview's audio file.
Linda Hill: You have to have pretty good judgment about what really matters. And one of the things that executives are telling me who are deploying all of these different tools is—
Voiceover: The most convincing answer isn't always the correct one.
Jake Cook: I think this kind of comes down to a couple pieces primarily related to culture.
Voiceover: The tool isn't the hard part. Sometimes the people are.
Felix Oberholzer-Gee: If the claim is that it can perfectly do that, that claim is not right.
Chris Linnane: So that would be over-hyped if they're saying you can rely on this thing, no questions asked.
Voiceover: Three experts identify three blind spots. This is "Who Is Getting AI Wrong?" Welcome to The Parlor Room Presents: Hello AI.
Chris Linnane: Welcome to this compilation episode of The Parlor Room Presents: Hello AI. In this episode, we're taking a closer look at who's getting AI wrong and why they're getting it wrong.
What I've noticed is that AI doesn't seem to fail in the way that people expect it to. It doesn't crash, or break, or put up a giant error sign. It fails quietly, confidently, in imperfect, predictable sentences, and most people don't notice it until it's too late. The most dangerous thing about AI might not be what it gets wrong. It's how convincing it sounds while it's doing it. Confidence has a way of passing for truth. This is Linda Hill.
Linda Hill: When you're working with gen AI, or agentic AI too, the issue is the human is, at this point anyway, you're bringing the judgment to the equation, the solution. Because these are just mathematics. These are equations that are running, right? All of that. But to figure out the use case and whether it's how to design it in a way that it will solve the problem that you want it to solve, use case, and/or to look back and say, "Well, did it really do it, given that we know the technology is still evolving," you have to have pretty good judgment about what really matters.
And one of the things that executives are telling me who are deploying all of these different tools is you have to keep reminding people that it isn't human, because one of the things that sort of tricks you about it is it looks like it's human. It's speaking in natural language.
And so I was reading a study, and I don't have the statistics in front of me. It turns out that if you tell people, "This solution is the result of machine learning," some percentage trusts the solution, others don't. A large percentage don't. When gen AI does it, a larger percentage trusts it.
Chris Linnane: Why is that? Do we know?
Linda Hill: Well, because it's in natural language.
Chris Linnane: Okay.
Linda Hill: And you type the way...and it is not accurate, though, because in fact, I mean, there are two different sets of tools. But in fact, the machine learning answers and these people who do these studies are actually much more accurate than what's coming out of whatever a large language model has produced for you. It's just that it looks—we anthropomorphize. We sort of say, "Oh, no, not true."
So you actually need to have people who have good judgment to understand that this...I'm sure you've gotten some answers to some things, and you think they're true and they're not at all true, but it answers you in a way that makes you think, "Oh, this looks pretty authoritative, and it went through a computer."
Chris Linnane: Yeah. And I've talked about this before is it makes me think I'm a genius at the same time, because every idea I came up with is outstanding, every idea. And then it takes me a little while to realize that was not a very good idea.
Linda Hill: Total hallucination, right? Those facts, that place doesn't exist.
Chris Linnane: Yeah. Knowing AI makes mistakes is one thing, but knowing how to manage around them is completely different. It's a growing skill, and most organizations and individuals haven't figured it out yet. This is Jake Cook.
When you see teams struggle with AI early on, what questions do they usually start with that send them in the wrong direction?
Jake Cook: I think this kind of comes down to a couple pieces primarily related to culture. And so, if the culture doesn't reward experimentation, or it's okay to fail, if they sort of expect perfection, which no one wants to ship sloppy work, right? But inherently, AI is probabilistic versus deterministic, meaning that at its core, the way it's working is it's kind of guessing what's going to come, whether it's a pixel or a word, it's just making a probability of a guess, which means inherently it's going to be wrong, just like human beings are wrong, right?
And so culturally, if they're penalized for that or they give it a task that's too broad and complex, and so it's almost like they'll hand something maybe instead of approaching, I always say, treat it like an intern and maybe one that partied really hard last night.
Chris Linnane: Oh, okay. Limited expectations.
Jake Cook: Yeah, limited expectations. It might be a little bit fuzzy, but it wants to please you in a very big, big way. And so when you hand it a task like, "Come up with our go-to-market strategy for next year and what products we should develop," you would never in a million years deploy capital off of an intern's recommendation on that, right?
Chris Linnane: No.
Jake Cook: But we kind of treat it the same way, "Okay, well, tell me what to do." And then what happens internally with cultures is some people are really fearful of their jobs. "This is going to replace me." And when it fails, they cheer. "I told you it couldn't work; you need me."
And so that can be kind of this false negative with the AI, or a little bit of like they're still kind of building a sandcastle on the beach, hoping the wave's not going to come in. And so culture, I think, celebrating failure, building small micro experiments, and modeling that. One of our clients, the chief data officer, is wonderfully brilliant, very, very innovative, man, super great to work with, and he started going through in their all-hands and just being like, "Okay, I'm going to walk you through how I think we should do this. Here's some data, let's go through...let's do it live." And when people kind of see leaders or their managers kind of doing this and failing, I think that's huge.
Chris Linnane: The real risk isn't whether companies use AI. It's what happens when they trust it past the point it was built to go. Now this clip jumps right into a game I was playing with Felix Oberholzer-Gee.
AI helping leaders decide which projects to stop or prioritize. Is that over-hyped, under-hyped, or just about right right now?
Felix Oberholzer-Gee: So I'm not sure about—if the claim is that it can perfectly do that, that claim is not right.
Chris Linnane: So that would be overhyped if they're saying you can rely on this thing, no questions asked.
Felix Oberholzer-Gee: Yeah, for exactly the reasons that, I think, one good way to think about—two colleagues of mine, they had this interesting way of thinking about when should you rely on AI and when is it OK not to do so. So Andy Wu and Bharat Anand, they have—the source of information is explicit data that you have somewhere, versus tacit knowledge. The history of experiences that this particular executive has, that's a body of knowledge that is incredibly important in deciding, "Should I buy the company, should I not buy the company?" But it's not data that you can easily access. It's an intuition. It's a sense that you have. It's human judgment.
And then the other dimension that they talk about is like, "How costly is it if you're wrong?" And even though it's like a super simple two-by-two, anything that is sort of in that there's a lot of tacit knowledge, probably AI will perform not quite as well as the typical human person who has a lot of experience—which doesn't automatically mean you shouldn't use it because then you ask, "OK, so it's probably not going to be very good. Is it costly if I make a mistake?" And the answer is, "No, actually it'll be OK."
Then, of course, I can still use it if it's the acquisition of a company, and I get it wrong because I had AI decide, then that's probably not a place. So, this goes back a little bit to thinking about jobs as bundles of tasks. The more tasks rely on tacit knowledge, I think, the harder it will be to really take the human out of the equation.
Chris Linnane: As always, thank you so much for watching and listening. If you're enjoying the show, please take a quick moment to like and follow the show wherever you get your podcasts. I would greatly appreciate it. Thank you.