Cheryl Strauss Einhorn | The Human Edge: Smarter Decisions in the Age of AI | Talks at Google

It's great to see you, Cheryl. How are you?

Good. So nice to be here with you, and so nice to be here with all of you.

So Cheryl, if you could summarize in just one sentence, just capture in one sentence The central idea behind this book, The Human Edge, and why did you write it now? Why did you feel compelled to write it now? Please.

So the one sentence about the book is, at a time when answers are very easy, we still need to understand and bring forward for ourselves what matters, what information do we trust, and what should we do next? And these are uniquely human questions.

And that's your thesis.

That's my thesis.

So if we wanted to go a little deeper, you use these fantastic scenarios about maybe people who have consulted with you and you're helping them make decisions. As you tell us more about the book, can you use those scenarios in painting and illustrating the real deep context of the book?

Sure. So I'll tell you what led me to write the book first, which is, as AI first started coming out and working with individuals and companies and students, I realized that so much of the conversation was on how to use the tools. But what people I found were really struggling with is, what does it mean for me To be in the room with the tool? What is my unique responsibility and what part of decision-making should I delegate and which parts should I keep? So in thinking about what is uniquely human, what I'm recommending to people is that for the very first time that we begin to think about our special sauce, We all have a way of making decisions. That's how we get to the future, those actions that we take. But most of us have not really identified what we do to get there or why we rely on the process that we use. And so if we can begin to think about our thinking, the metacognition, we can almost think of self-awareness as a competitive advantage. Right? If we can bring that forward to be relying on ourselves to think about what problem am I solving, why am I solving it, and what are those key contextual factors that AI can't know— if we can bring those things forward to ourselves and then use them with AI, instead of having answers that are general or that are highly available, we can get answers that are specific about what matters to us.

So, I mean, you emphasize AI as a collaborator as opposed to us placing it in the role of a decision maker. And I very much believe that. And as we go through, you know, more of the concepts in the book, let's talk about that because I'm a firm believer in that. And you paint so many scenarios to show the danger Almost, or the risk, I would say, of placing AI in a decision-making mode and, you know, us being very passive about it.

So, so when it comes to complex problem solving, there's sort of 8 critical steps. We don't all go through all 8 every time, but these are the areas where uniquely human judgment exists and where only we have the information. And so those would be problem definition, motivation, context, setting the research direction, conducting the analysis to find the meaning behind the data, identifying and controlling for biases, including our stakeholders, and coming to conviction. And at any of those moments, only we really know those answers, and AI doesn't. So what I offer in the book are 2 modalities And I gave them fun names just to help us be able to think of them as memorable. And I call them the surgeon and the Lamborghini driver. So the surgeon is when we come to AI, we know exactly what piece of information we're missing and we use it like an incision to extract that answer. For the Lamborghini driver, we're going through some number of those 8 steps. And in that instance, Only we can be sitting behind the wheel of this powerful machine and directing it, but getting quickly to the wrong place doesn't help us. So we need to know enough within the steps that we're using to be able to direct the machine to work for us specifically. So I'll just give you an example because you talked about so many of the things in the book are tied to different examples that I give So that readers can see themselves in it no matter where they are in their problem solving. So just thinking about, for example, giving an opportunity to somebody to, um, on your team to take on a new project.

Right. Right?

When you're working with an AI tool, if you're giving it some anonymized information about the different applicants, it may clearly say to you, oh, Marion, why don't you pick this person person who has the most experience, but only you know that a far more junior person on the team might be ready for that stretch opportunity or may look at that particular type of opportunity in a different way. And so you know best, even though the data may be giving you information that is actually different. And that gets to the point that we are not trying to always solve for for the best outcome possible, right? AI is going to give us an answer that could be, let's say, the most financially advantageous. But we don't make purely economic decisions. We make decisions based on our values, our emotions, our relationships. And these are things that AI just can't simply know or help us with unless we are able to bring that forward and then to be able to communicate it to the machine.

Yeah, that's very good. Yeah. I want to ask you something just a little personal. As you were writing the book and as you've gone about, you know, explaining the book to others, have you found that it's impacted your own personal or professional life?

Yeah, I think I had, as you might at any time when you're researching a topic, you come in with expectations and assumptions. And so one of the things that I really assumed about AI was that it wasn't gonna be that valuable when it comes to people. But I've actually found that one of the things that I really like about AI is that since I'm stuck inside my own perspective, that it can actually really widen my experience. So I can ask it to help me think about what would the biggest critic say? Or I know that I'm going into a meeting and somebody always pushes back and they have good points, but I'm not always ready for them. If I can share some of those details about the type of pushback, I can say to the tool, what might they come in with this time? How can I be better prepared and make them think that I am showing up ready to create greater trust and work in a better way with them related to decision-making? So I think that was unexpected for me. And then the other thing that I would say is obviously in writing the book and in testing the tool, I was using a lot of different tools very heavily. And now I really use AI with much greater intention. And I think there's something that I would want all of you to think about, which is just because AI can do something, here's the question to think about. Is there a purpose for me doing it on my own? Because a lot of times it can help us with things. But we're not trying to make everything easy. A lot of times where the discomfort is, is where the meaning is. It's telling us that something's important to us, that we're bumping up against a value or an emotion or something that matters to us.

I guess related to that, that's, that's a, you know, very interesting perspective. And I guess related to that, I would ask you, as As AI becomes more capable, and, you know, that's happening day by day, what human qualities become more valuable?

Yeah, so I think what becomes much more valuable is not only this invitation to get to know ourselves better, this metacognition and the idea of self-awareness as a competitive advantage, but certainly judgment. Because it used to be that people would move up the corporate ladder, for example, because they have experience, and that experience would give you judgment. Or think of it another way, right? We can learn theoretically how to merge onto a highway when we're driving, but that is nothing like us sitting behind the wheel during rush hour the first time we need to merge our car onto a highway. And so now we're having answers without experience. And so the question becomes, where are we going to get that judgment from? Because the judgment is generally telling us what's missing, what doesn't look right, and where should I press upon this answer. Now we all, because we're accessing answers about things we know nothing about oftentimes, we need to be identifying those questions. Where could it be wrong? How could it be biased? How could I bring my own assumption and judgment? How might algorithmic bias or availability or data bias be making that into a feedback loop? And how do I look for information that actually disconfirms this, which is far better than seeking to confirm a favored hypothesis? Because you could have a lot of information that could confirm something, but one insurmountable hurdle makes that an incorrect pathway forward.

So what you're saying is that as AI becomes more capable, our judgment becomes much more important.

Only we can decide what matters, what to trust, what to do next with the information, because when the hammer falls, it falls on all of us. It's still our work and our responsibility. We can't blame the tool for having done something. So these ideas where these questions become far more critical And I've been thinking a lot about what are other ways that we can help people build judgment, right? Can we be thinking about collecting examples where something has gone wrong when we're working with AI and use them as teaching tools where somebody who is an expert can say, this is where you might have looked for a sensitivity in the data, or here's where you might have identified that there was a gap. Or a credibility issue, because without having experience to really be able to check and challenge is not as easy as it sounds.

That's, that's what, what do you think is the biggest, you know, what you're basically saying is you have to be very proactive in your use of AI, and most people are kind of passive. You know, and they'll just let it guide them. And so what do you think are some of the misconceptions people have about competing with AI?

So first, it's not a competition. It's much better than us at collecting research faster and in much greater depth. And it also is better than us at analysis. We need to tell it what kind of analysis to do. So I think that is an area where we all can learn. And I go through and I give a worksheet on that, like what are key types of analyses when you're looking to make meaning from your data that you can ask AI and it will do it better, faster than we might. But it can't ever tell us why we're solving a problem. It can't tell us what problem we're solving. It can help us define a problem. We don't need all of these prompts anymore. It can help us us to develop the right prompt, but we need to be able to define the problem we're solving, know what that motivation is, communicate that to AI, and those key contextual factors that will make or break it. And so I really don't see it as a competition. One thing I would say that I see very often in the classroom and also in my work with leaders is Yeah. People think of interacting with AI as a transaction, and it shouldn't be. It should always be a conversation. We need to build a new habit of first turning to ourselves, and then when AI gives us back the answer, we have something to react to, and we always need to be pressing back on where could this be wrong and how. And I'll give you just one anecdote. I was looking for a citation, And I asked AI to help me find it. And it came back with a very credible-sounding journal by a researcher whose work I'm familiar with and who I really admire. And I thought, well, this is wonderful. I really— I'm so happy that I have something that, um, that I can use. And I thought, well, of course I'm going to check this. The journal didn't exist, and this researcher never wrote that article. So the problem is Almost everything sounds plausible. So if we don't have experience to think that something that looks so polished and sounds so plausible has limitations and flaws, we need to develop that judgment to know where to press on what could be wrong. And then we actually need to check the primary sources.

And then I guess as AI becomes more perfected, right, and these making mistakes or hallucinating, would you still say that our judgment is as valuable?

Yeah, I think I wrote the book thinking it's got to be agnostic of how the tools change. They're going to change a lot. What remains uniquely human, though, and figuring out what we care about and why and what information to trust, that's going to remain true throughout our lifetime. So as AI gets better, I think what we're facing, and this is certainly true with agentic and AI, is it's making a whole sequence of assumptions.

Mm-hmm.

Right? So we need to know even beyond what we've told it and where we might be making assumptions, how to check where it is making assumptions, because we don't know in the second step of what we asked us, it had some kind of mistake and now it's carrying it through the process. So I think the onus is always on us to know that we need to own the answers and decide what we're willing to accept.

Okay, that sounds good. And I guess related to that, is there something that humans do naturally that machines, like, fundamentally Well, I think this idea of knowing where and how to exercise our judgment, right?

The fact that only we know what problems we're solving, that is uniquely human. And then this idea that I was talking about that we don't make decisions for purely economic reasons. I may willingly and happily accept a suboptimal outcome if it produces a positive outcome. That preserves a relationship, without a doubt. And I think, I think all of us do things like that in our lives all the time, and those are calculus that only we can decide. And my decision on that, even if we were facing the same decision, could be different than yours because your motivation will be uniquely yours and mine will be uniquely mine.

Right, right. I can see that. Yeah, that's good. So in The Human Edge, you lay out a decision-making process that you created yourself years ago called AREA, and you rely on that in order to form the basic thesis of the book. Can you explain a little bit what EDGE is?

Sure.

And how it relates to AI and what, you know, just give us some examples.

So I'll give you a little background first in the ARIA method. So Marianne explained to you that I used to write for Barron's. It writes about publicly traded companies. And when those stories came out, I ended up specializing in what you'd call the bearish company stories. Stories that take a skeptical look at a company's finances or at their strategy. And when the articles came out, there was often a very outsized reaction. Often a stock exchange would halt the shares from trading because there wasn't an orderly order flow. At times enforcement agencies like the FBI would get involved, or regulators like the Securities and Exchange Commission. Over half a dozen companies went out of business. And for one company, the CEO was sentenced to 10 years in jail after a series of investigative articles. And although journalism really celebrates Truth to Power, it got me thinking, how do I know that I'm telling stories that are true, stories that should be told, that I'm marshaling the right evidence, and that I'm checking it for my own limitations as a thinker, these cognitive biases? And when I started to look for, are there processes out there that uniquely focus on these mental mistakes, this sort of dirty windshield that we look through where we bring our own experience but we may have unfounded assumptions and judgments. There was nothing out there. So I initially figured, with my background in research, maybe I'll put something together—

Right.

Just to do a more ethical job at work. And that became the AREA method. So AREA is an acronym for the steps of the process that is meant to uniquely control for and counter cognitive biases so that we expand our knowledge while improving our judgment. And I'll tell you what's very different from classical problem solving. Most of the time when we are starting to solve a problem and we're conducting research, we start looking for outside sources, right? What do the experts say? What can I learn from the outside world? What ARIA basically says is that misses out on you developing your internal compass as a researcher. So the first letter in ARIA is absolute information. It's primary source information up close on the target of your decision. So if you're investigating, for example, um, what's going on with AI in the New York City public school system, which just had a big announcement that they are limiting it to certain grades and only in certain ways, you might look and gather just the facts on what is the policy, what does it say for K through 3rd grade versus what it says for the higher levels and where it may be implemented. Now you are not assuming you know the problem. You're not jumping into options or jumping over the problem. You're actually defining it. Here's what the actual primary source information says in absolute. Now you can move into the R of area, relative information, which is where we generally start. And what is so nice about differentiating is that in that second perspective where you gain information from outside sources, now you can vet that information against what you've learned from the primary sources. What's relevant, what isn't, how does the narrative change? And then the E of AREA, I consider the twin engines of creativity, exploration and exploitation. Exploration gets beyond documents, So you learn about lived experience. What can the numbers not possibly tell you? So for example, something can be 10 miles away, but it takes you 30 minutes, right? That's something that you can't know from just looking at the information. It's much more useful to actually hear that from somebody who's dealt with that problem. Then area exploitation is a brand new step of problem solving where you look at those assumptions and judgments again, against the evidence you've collected in those earlier steps. And then the final A of analysis puts the process back together and helps you to come to conviction. So the idea of ARIA is that we're always making decisions amid uncertainty, and what we need is confidence and conviction that we've gone through a quality due diligence research process that arrives at a decision that we can really stand by. And that we're not assuming is true. And so in using AI, the reason why I think something like ARIA is so useful is first you're starting in all perspectives at once, right? So you're starting in that relative area. So if you use the actual letters of the ARIA method, you're not skipping out on holding yourself accountable for making sure that you're moving through the different perspectives and you're checking and challenging your assumptions. And this really becomes so important because we use AI like we are, not as it is. And my third book is on decision-making archetypes and the 5 different ways that people solve problems. So each person has a data preference, right? So some people like to gain information about the wisdom of others, and they'll come to AI asking about what the outside world has to say. Some people just want the numbers, and they'll focus on that. So a system like ARIA is making sure that you go through even those steps that are less comfortable to make sure you get a fulsome understanding and you're always checking the integrity. of the information that you have, which I think is really important with an AI tool.

So you see them as going together?

Yes.

But you created ARIA independent of AI, and you— do you think AI adds a richness to the information that you have?

Well, I think that what AI can do is it can help you with all of the steps of the process. But you won't necessarily be collecting information from all these different perspectives unless you use a process like AREA. So a process is an equitable tool, it's free, and it gives you these nice guardrails so you can bump up against the outer limits of it and still ensure that you go through a system that Looks at how the different ways you could interpret data, right? Because each piece of data can be interpreted through multiple lenses. So understanding where the diagnosticity is and how it's relevant and how it could be incorrectly understood for the process that you are using it for becomes really important.

Yes, and I imagine the, the role that you used to have, it would have been tremendously Yeah. Yeah. That was an interesting job.

It was a very interesting job.

Yeah. I can see why you went into teaching after that. Yes. That's good. So, I wanted to know, like, you talked a little bit about this just now, but you've done research, as many of us have, on the different ways that people make decisions and solve problems. And, and I was wondering how you think, you know, AI in general is going to impact diverse teamwork.

So because different people look at problem solving differently, they look at different aspects of a problem. So let me just give you an example. People think about speed differently. Some people see it as an asset and some people see it as a detriment. Some people, as I said, have different information preferences, right? Some people like to collect the data, some like to jump right to the options, some like to gather the wisdom of others, some like something that is big and creative and outside of the box, and some people just want the next step in the logical progression. And then different people also need different quantities of that, and they think about including stakeholders at different places at different times. So I'll just tell you very briefly about the 5 problem solver profiles. I gave them fun names because we have never had a language to think about decision-making archetypes. So they're the adventurer, detective, listener, thinker, and visionary, and each of these will use AI very differently. The adventurer is somebody who is very intuitive and has a very strong gut. And as a result, they have almost an emotional visceral reaction to information that gives them very quickly a direction to go in of what they think is the next logical step when solving a problem. And that has a beautiful optimism bias to it in the sense that if they make a decision and it starts to go awry, they always have the next decision ahead. And there's something very comforting about that. So when they're using a tool, they're the most likely to sort of look to extract information in the surgeon mode, take it and go. Right. But our detective— and I'm a detective— we're people who really like research. We like to get the hard evidence. And that means that we really can fall prey to a confirmation bias, right? We know how to do research. We seek to confirm a favored hypothesis. We talked about how that's not as good as looking for disconfirming data. So I could come to AI looking for the numbers and for evidence, and I'm gonna think that that's how you should be thinking about it too. The detective has a real projection bias that other people should also see the data as having a lot of diagnostic value. And then there's the listener, and this is our most collaborative, cooperative, trusting decision maker. This is the person who has the least developed gut, and as a result, they can fall prey to a social proof bias where they go along with the rest of the team, or a liking bias—they can overweight information from somebody that they have an affinity for. So when using an AI tool. It may be very useful to gather other people's opinions, but they really want to know what that trusted group of advisors has to say. And then you have your thinker, and this is most of the people in my data set that I have, almost half of them actually. These are our most cautious, thoughtful decision makers. These are the people who end up in a real relativity bias because they want to look at the options and they can too narrowly circumscribe the decision as they look at one option against the other. They often jump over to finding the problem, and when using AI, they want all the information. They want to know what the data says. They want to know what the different experts have to say and so on. When they actually arrive at a decision, though, this is the group that can uniquely explain to you why they took the path that they took. And why they think it's going to succeed. And then the last problem solver profile is our visionary. And this is the person who wants something that is outside of the ordinary. And so you can see when they're coming to AI, they may know to ask for lesser-known information beyond what is available. Where else should I be looking for voices that may not easily rise to the top in my first AI usage? So you can see that each of these profiles are looking for something very important, and they're most comfortable staying within the data type and the data quantity that has really served their purpose in the past. So when working with a team, unless you have the real intellectual diversity, what's beautiful about AI is you can ask it to play the role of these different profiles to make sure that you're bringing in data from all these different perspectives to give you a fulsome understanding of the problem you're solving.

I like that. I like that. Using AI to recreate the diversity of teams and their thought processes.

And it's beautiful also to help you with playing devil's advocate. And, you know, it can come up with scenario analyses and give real pushback on ideas that teams would normally have to bring forward and interrogate on their own. And it allows teams, I think, to sit on the same side of the table then because you don't have to be the person looking to poke a hole in an idea that the team wants to investigate. You can let the tool tirelessly do that while everybody else is thinking about solutions together.

Yeah, that's interesting. That's great. Yeah, I like that. I hadn't thought about that. One thing is, what can the audience do tomorrow if they wanted to strengthen their human edge?

Yes, I think there's 2 things you can do. And I think I've mentioned them both, but I think they are worth repeating. The first one is asking yourself, just because AI can do it? Is there a purpose for me to be doing this on my own? So what part of decision-making is uniquely mine that I don't want to be delegating? And then the second thing is to create this new habit. We're used to sitting down at our technology and just starting typing, but now with AI, since we're going to be seeking an answer of some kind, always start with yourself first and ask yourself those 3 very early parts of complex problem solving. What is the problem I'm solving? Why am I solving it? My motivation. And what are those key contextual factors that are going to make or break it? So that you have something to react to when the machine gives you an answer and you have brought that critical thinking forward.

Oh, interesting. Good. All right. I'm going to I'm going to give it a try. If you could leave the audience with one challenge, what would it be?

I think it would be to really take up the invitation to get to know your own thinking better. I think it's a really interesting moment when people are criticizing that this is the first tool asking for some of our cognitive load, right? You get to decide that, and you can use it to really strengthen your thinking by really getting to better understand what goes into solving a complex problem and then making sure that you are better identifying in yourself what are your values, what are your morals as it relates to this, why does the outcome matter to you. And that way you can continue to learn more about yourself, you can express it to other people when you're not on the machine, and And you also can get the machine to work not just more efficiently for you, but more effectively.

It's so interesting. Many, many people, especially outside of, say, the technology areas, you know, see AI as diminishing human value, where you're saying just the opposite.

I'm saying that that really is the invitation here because we've really never thought about these problem solver profiles, why we solve a problem in a specific way. What does that say about our underlying value? Right? The listener values other people, right? And the detective values data. And each of us have something that is uniquely important to us about the way that we get to our future. But only we should get to determine what our future looks like and how to open up that That real possibility. So I do think that we could really be strengthening our thinking if we accept this challenge.

Yeah, yeah, that's a fascinating perspective and I love it. So thank you, Cheryl.

Thank you, this has been such a pleasure.

Thank you.