What we risk losing to AI

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

NATALIE MONBIOT: We are constantly creating digital twins of our knowledge, and we’re deciding what interfaces and who this digital twin interfaces with. Usually, it’s ourselves, but we are constantly creating replicas of our thinking, replicas of our work, replicas of our lives.

RANA EL KALIOUBY: What’s at stake if we’re always allowing a machine to think for us?

MONBIOT: The atrophy, and basically losing the capacity and the confidence to actually come up with ideas and thoughts for yourself. When it comes to not trusting yourself to actually think a complete thought without an AI, that is pretty dodgy. We need to, at the very least, bank on our own ability for ingenuity.

EL KALIOUBY: That was Natalie Monbiot. She is the founder of VHE, or Virtual Human Economy, and one of the very first experts to demonstrate that your digital twin can be one of your most valuable assets. But recently, Natalie has been wondering about a different set of questions.

How do you divide work between yourself and your AI? And how do you do that in a way that protects what makes you unique and valuable in this world?

I’m Rana El Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

[THEME MUSIC]

Hi, Natalie. Good to see you again. Thank you so much for joining us on the show.

MONBIOT: Of course. Great to be back.

EL KALIOUBY: So it’s been over a year since we did our last Pioneers of AI conversation at On Air Fest, which was really fun.

MONBIOT: Feels like forever ago.

EL KALIOUBY: I know. It really does.

MONBIOT: In AI years.

Copy LinkHow digital twins evolved beyond into everyday AI tools

EL KALIOUBY: Yeah, absolutely. Have you been playing around with your digital twin at all?

MONBIOT: Interesting. So my digital twin has morphed. Everything about it has morphed. The concept itself has morphed. I think we’ve moved beyond this idea of, “Oh, here’s my digital doppelganger. It actually looks like me and sounds like me.” That’s quite a limited use case and concept now, given where things have gone just in this past year.

EL KALIOUBY: Say more.

MONBIOT: We have digital versions of ourselves everywhere. If you are an active user of AI, which you are, which I am, and I’m sure many of your listeners are as well, we are constantly creating digital twins of our knowledge, and we’re deciding what interfaces and who this digital twin interfaces with.

Usually, it’s ourselves, but we are constantly creating replicas of our thinking, replicas of our work, replicas of our lives. We want our AI to know us as well as possible in order to serve us as well as possible. So we want that chasm to narrow, and we want the understanding between us and our AI to align. Anything but alignment is a source of frustration.

So how I think about what a digital twin is has kind of lost its embodied quality, and its main utility, I would say, is not necessarily embodied at all.

But it needs to be there for us whenever we want and need it, and it needs to get us. I’ve been playing around with ways to make it get me more.

EL KALIOUBY: That’s actually so interesting because, in the narrowest sense of the term digital twin, I do have a digital twin. But I usually don’t use it. I’ll show it at conferences and keynotes because I still think it gets people thinking. It’s thought-provoking.

But I hadn’t thought of the everyday AI tools that I use as digital twins of myself. Actually, in a way, they are. I still use ChatGPT for a lot of things because it knows my voice, so if I want to draft something or if I want help iterating on content or any writing, and we’ll get into that, I still go to ChatGPT.

But for tasks, I use Claude Code a lot, and I still don’t know that we’ve found the right productive relationship between us. I still find myself using all caps a lot. I’m like, “I told you not to do this. Why are you still…”

MONBIOT: I know.

EL KALIOUBY: “Add it to your memory file once and for all.”

MONBIOT: Can you still not get it?

EL KALIOUBY: Yeah.

MONBIOT: I definitely see the value of an embodied digital twin as something useful. It’s a starting point. People start talking about AI, and it could be absolutely anything. It’s such a loaded term and could mean something different to absolutely anyone.

So I think it’s quite a good way to ground everybody in at least a starting point.

I also feel that a digital twin, in the sense of how one collaborates with your AI, is actually a very useful metaphor. The rules of engagement and the best practices of what you would have your digital twin be and do are a good metaphor for how we want AI to work for us.

Copy LinkChoosing which work to delegate and which work to protect

EL KALIOUBY: Well, give me an example.

MONBIOT: For example, you have a lot of work that’s very routine and repetitive, and you’d just rather not do that yourself. Okay, great. Offload that to your AI twin. If you have a high-stakes meeting and a lot hinges on this, do not delegate that to your digital twin.

You do that yourself. Now that you have more space and time, spend more time preparing for that high-stakes meeting. So that’s the very basics, right? Then it can evolve from there into more nuanced scenarios. If you’re using your digital twin as a thought partner, or you’re using it for developing writing or whatever it is, who is doing what?

What are the roles and responsibilities in that, and why? Ultimately as well, what is it that you’re trying to create and achieve, and how can your digital twin help you get there, to that ultimate goal?

So if you’re trying to write something original, I wouldn’t lean on it for the actual ideas, right?

I think it’s about thinking about what it is that you’re trying to do. If you’re trying to write loads of marketing material personalized to different audiences and different individuals, yes, have it do that, right? That doesn’t need to be original. That’s just a lot of work and a lot of variations.

The AI is really good at that, and you want to give that up.

And then I think there’s this thorny area that really fascinates me.

What can AI never do? People say AI can never be creative, or AI can never be this and that.

That position of being static as a human and just postulating about what AI can never do is a super-unproductive position to be in. The only thing that can happen is that AI keeps encroaching on your definitions, and then it’s like, well, what have you been doing?

How have you been cultivating yourself or defending yourself against this, or evolving yourself to meet this moment? So I think the question is less, what can AI never do and will always be mine, and I’m just going to say it’s mine?

EL KALIOUBY: It’s almost, in a way, like fixating on the idea that humans have a moat and AI will never encroach on that moat. That’s probably the wrong framework to take anyway.

MONBIOT: Yes, exactly.

I think the better way to think about it is: What do I actively claim? What is enriching? What drives my growth, if these are your values? And where do I draw that line? Even if the AI could do it, and we know there are many cases where it can write stuff for you, it can do lots of things, but is it to your standard?

How do you, collectively with the AI, meet your standard more effectively? And I think often that is not outsourcing to them.

Copy LinkWhy human judgment matters more than defending a fixed human moat

EL KALIOUBY: So that comes back to something that you’ve been thinking a lot about, which is human judgment. You had this event that happened where you were in conversation with a consultant that you really admire, and something happened that was a turning point in how you think about all of this.

Can you share that story?

MONBIOT: Yeah, it was really funny. So about a year ago, I was like, “Yeah, I really want to make these digital twins, working with a startup to make digital twins for consultants.” I had this great consultant based in Singapore. We were like Substack buddies. And I was like, “Clearly you want a digital twin.”

You’re an AI consultant. You’re all in on AI.” And he was like, “No. The thing is, if I create my AI twin, no one’s going to want to talk to my AI twin. They’re better off talking to ChatGPT.” And I was like, oh my goodness, humankind, we’re in a really bad position if an expert who trades off his expertise doesn’t have confidence in his own knowledge to stand up against the LLMs.

We’re in a really bad spot. It took time and my own reframe to come back to that conversation and think about it differently, because actually what he meant was, my value does not translate into exported expertise. Okay? So if all you want is the sort of knowledge and the expertise, ChatGPT might, in general, just do a better job, be much faster, be much cheaper, all of that.

His point was, it’s him, his presence in the room, the trust he’s cultivated over time, his network paired with his expertise, and the trust that he’s cultivated among his clients. All of that is really what matters. So I wrote something about that on Substack.

But afterward, I was thinking about it a bit more, and actually, the real answer, I think, at least as of now, is him with his digital twin

working together in this complementary fashion is the strongest bundle. And so, in this piece, I wrote about referencing some research about strong bundles and weak bundles.

EL KALIOUBY: You kind of explored the history of labor in America first. Is that how you, was that how you kind of —

MONBIOT: Okay, yes. So going a little bit further back in —

EL KALIOUBY: Briefly.

MONBIOT: Yeah, briefly. Just briefly. I’ve been kind of obsessed with this idea of the cognitive division of labor. And so Adam Smith, in the 1700s, coined the term in the book The Wealth of Nations, the division of labor, and has a story that paints this picture of a pin factory. Basically, if you have one person make a pin by themselves, that’s going to take

Imagine making a pin by yourself, right? But then if you divide up all the different parts and stages of making a pin, you can create millions of pins in no time, right? So that’s the division of labor, and look at that increased productivity and all of that.

First of all, Adam Smith actually said it was pretty dehumanizing to do that to people, because they basically literally are like a cog in a wheel, or whatever, like the pinprick in the pin.

EL KALIOUBY: Very automatable. It’s very replaceable, automatable, right?

MONBIOT: And it’s very moronic, right? So basically you’re dehumanizing somebody by giving them such a moronic job, right?

Which is just a single task. And so anyway, that person at least is the person with the skill to be able to do that. Okay? So they have value. The difference now in the cognitive division of labor with AI is that if we let AI do the thinking for us, they are thinking machines, right? So if they can do the thinking for us, we’re left with nothing, at least in knowledge work.

Because our value has been in the thinking. So if you give up the thinking, then you’re kind of left without a role. And on top of that, your ability to think atrophies. So the AI becomes more and more powerful, and you become more and more incapable of the thing that made you valuable in the first place.

EL KALIOUBY: Yeah.

MONBIOT: That’s how I’m trying to frame this moment to help myself and others understand and place where we are versus other eras where we’ve divided the labor.

Copy LinkThe cognitive division of labor and the risk of thinking atrophy

EL KALIOUBY: I kind of want to ground this in the real world. So let’s take an example. Walk us through how you use AI in your everyday life, and let’s think through this cognitive division of labor. How do you decide what to delegate to any of your AI tools versus what you’re going to do yourself?

I can also share a personal example too, but you go first.

MONBIOT: So let’s just use writing as the example. I think it starts with your goal. What is it you’re trying to achieve? My goal with writing is to write things that are meaningful and to create understanding from a human perspective for other humans.

Where it usually starts is I’ve had some kind of real-world experience, and this is how I try to cultivate authenticity for myself.

I feel like what I’m doing is creating value. I try to ground everything in a real-world experience, right? I will often feel the inspiration, and then I will usually capture voice notes in Whisper as I’m walking around.

Almost always when I’m just sort of walking around, I let my thoughts flow, and I feel like that’s where both the logic and the insight and the passion for the subject come out. And if I’m not excited to be doing that voice note, there’s nothing there, right? I might put it into Claude and arrange it a little bit as I’m trying to build my outline, and then it might do something to it and I get really annoyed.

And then that annoyance actually kind of makes me like, “No, not like that, like this.” And so then it makes me clearer in my process. I will sometimes also write by hand, which is extremely excruciating if you’re not used to writing with a pen and paper. And I’m really out

EL KALIOUBY: I know. My handwriting has become horrendous. I used to have the best handwriting, I know.

MONBIOT: Yeah. So then I will actually excruciatingly write out the points, and I feel like that is literally the craft of writing. It’s so hard, and it’s like, “Is this next word worth writing?” Because it hurts. So I think that can help arrange the thoughts as well. So I’m kind of going back and forth, and then I’ll

Then I might do some more voice notes if I’ve got another insight. And then I layer that on top. Then I might put it into Claude, and then I’ll rearrange things. And then I might upload what I outlined on paper. I think whatever happens, if you’re trying to write something meaningful that feels true, it’s excruciating.

The excruciation doesn’t disappear. It’s in there, and it just may be the process looks different. I would say that my process is very multimedia, right? So I’m physically writing. I’m in Claude. I’m using voice notes. And then once I’ve got close to a draft, I will

I listen to it in my voice on Eleven Reader, which is the reading app of Eleven Labs. So again, it’s another reason just to get up and listen, and then you catch, wait, that doesn’t fit together, or I need to add this in here.

It creates a lot of clarity in a different medium, but there’s no way to dodge the excruciating part, because that’s the part of the process where you’re actually coming up with something, and that is a very human thing.

And even if you tried to outsource that, I don’t think you could and still be true to your goal.

Copy LinkUsing AI in creative work without outsourcing originality

EL KALIOUBY: What’s striking me in your articulation of this whole process is that it doesn’t feel shorter or faster than doing it without AI. So why use AI in the first place? What value are you getting out of AI when you are doing this multimedia writing process?

MONBIOT: It’s a really great question. First of all, I’m not at all into AI companions, but I found myself about to say the word companionship. It’s a less lonely process to know that there is this other intelligence to spar with that’s there when you need it.

So I feel like there’s a camaraderie that makes writing less lonely, and writing is famously a very lonely experience, and I think that’s gotten in the way of me doing it.

I also like not having to be sitting down. I can be moving around and coming up with stuff between other projects or just in the course of my day, and I’m able to capture it and synthesize it wherever I am.

EL KALIOUBY: That.

MONBIOT: I love that. And so it becomes this thing that I can be doing for a week or two, and I’ve found that once I’m committed to something, it still just takes a minimum of five days.

I’ve developed my own kind of process, I guess, which is now becoming a little familiar, but it’s absolutely not less work.

Something I’m deeply aware of, but also very self-conscious about, is the flattening of thought.

EL KALIOUBY: More about that.

MONBIOT: First of all, there was research from last year about how, if you haven’t put the ideas down yourself and you let the LLM come up with the initial draft, you might not feel any connection to that writing.

You wouldn’t be able to tell me what was in it, so there’s no connection to what you did. If your goal is ownership and growth through the writing, that’s absolutely not going to be happening. So I think that’s pretty established, and the insight there, or the takeaway, is: do the first draft, come up with the ideas, do that part yourself, and then introduce LLMs. That would be the good process.

People say, “Oh, use it as a thought partner,” right? Bounce ideas, and that’s okay, but then you do the writing. But some recent research actually shows that that’s a really bad place to use LLMs, because that’s the point in the process where you’re coming up with ideas, right?

And you’re potentially coming up with outlier ideas, and those can be gold. The research shows it actually reduces the number of ideas and reduces the number of potential outcomes or ideas.

EL KALIOUBY: Of possibility, basically. The space of possibilities.

MONBIOT: Last week I was at the Yale Technology Summit, and the panel was about algorithms and polarization.

And obviously that’s what we think about when we think about social media and the incentives around social media: get people to be scared, inflamed, angry, or whatever it is, and then people engage, and then social media companies make a lot of money. That sounds like malintent, right?

That’s not a great business model, and it’s not particularly ethical. But with the LLMs, it’s not that anyone intends a flattening of thought, but just the way the LLMs work and how they’re developed, it’s just going to be that way, right? So someone’s deciding, or a team of people at just a handful of companies are deciding what good is.

And that is the mean to which everything is eventually driven. So that, I think, is this kind of insidious thing that’s happening, and it is not anyone’s fault. It’s a property of the medium.

EL KALIOUBY: I’ll be right back after this short break.

[AD BREAK]

Copy LinkThe cost of AI flattening our ideas

I want to broaden this conversation beyond just you and your AI tools to include other human collaborators and other AI agents that could be part of an organization or a team. As you know, at Blue Turtle Adventures, we’re a fairly small team, and we’ve been using a lot of Claude Code to augment our team, so we have a chief of staff AI agent that we use a lot.

And I’ve been finding myself really struggling to decide what work I should do and what work I should delegate to Blue, our AI chief of staff.

MONBIOT: Great name. Blue Tulip. I love it.

EL KALIOUBY: And also what to delegate to some of our more junior team members. I’ll give you an example. We get a lot of inbound from founders and startups. I meet a lot of people when I’m traveling, giving keynotes, whatever, that could be potential investors or potential founders.

One really excruciating task is to include that and add that to our CRM. And that usually includes doing a lot of research. You have to find the LinkedIn for that person, their organization, blah, blah, blah. I’ve been catching myself asking: When do I decide to delegate that to AI versus a human?

And I haven’t figured out what my framework is. The advantage of AI is it gets it done right away. It’s in the CRM within seconds, which is awesome, but it doesn’t quite do it right. It’s quite frustrating. Whereas I know if I do it myself or delegate it to another human on the team, it will be done right, but it may not happen for a few days.

So I don’t know. How have you thought about this cognitive division of labor when it involves other human beings and other AIs?

MONBIOT: I guess that’s the judgment that we’re left with, right? Actually, if you’re finding yourself judging more than anything these days, or wrestling with making these types of decisions, that’s probably a good thing.

EL KALIOUBY: That’s a good thing? Okay.

MONBIOT: Rather than just slogging away inputting stuff in a CRM thoughtlessly, you’re actually thinking about where that division of labor falls. I think it’s a constantly moving thing.

That said, a bit more of a helpful answer, perhaps, is that I collaborate with this brilliant AI development firm called AE Studio, and we’ve been working on agents in the org chart.

Copy LinkHow to place AI agents into teams

EL KALIOUBY: Can I stop you there? Because I think some of our audience may not even realize that we are now approaching a world — or we are already in a world — where AI is inserted in the org chart, right? If you look at an org chart, it’s a combination of humans and AIs, and that has a lot of implications for how we do work.

MONBIOT: Absolutely. So what is this? It sounds wild: agents in the org chart. I know that we’re in this moment now, and we’re not projecting some future that’s around the corner and who knows how long it’s going to take to get here.

I can see this because the people coming to me about this are actually HR leaders and talent leaders. It’s not the tech and product teams asking this question or carrying this mandate. Actually, I’m pretty glad it’s the people people who have the mandate and who are asking these questions, because it’s a very human question, obviously.

What is the humane and profitable collaborative way to bring the agents into the org chart, right? First of all, I like that framing: agents in the org chart versus agents just wiping out the org chart. I think that’s a good premise, right? I think finding this complementarity between humans and AIs is the way forward.

An approach to address this is to take a team’s workflow and break it down into all of its distinct tasks. Then work out how much time is being spent on these different tasks and how automatable those tasks are. The ones that can be automated can be automated, or an AI agent can be built to take care of those tasks.

That’s good for everyone, as long as everyone feels culturally like they’re being supported and they’re not just thinking, “Oh, once this project’s done, we’re out the door.” This is very theoretical — like, “Oh, and now the humans have all of this spare time, and we can reallocate their time to higher-order work.”

I think that’s the ideal, but the reality is — and I’ve talked to my husband about it, with him working in a hospital — everybody’s so stretched.

EL KALIOUBY: Over, right.

MONBIOT: So overstretched. Before we’re like, “Oh, what are we going to do with all this free time?” — which everyone says, like, “What are humans going to do when the AI’s doing everything?” — there’s a lot that can happen before humans have nothing to do that AI can support with. I think the first thing it can do is alleviate humans of the crushing responsibilities that they have.

If some of these tasks can be handled by an AI agent, then that at least helps human team members get their heads above water. Then, as things evolve, yes, if you’re in the luxury of being able to think, “Well, how am I going to reallocate my time?”

I think a lot of people might have a role that’s supposed to be more managerial or more authoritative and decision-making and judgment-oriented, but in fact they get dragged into all the hard teamwork, which just takes forever.

So I think that’s a real opportunity. Before we get really worried — at least in this context that we’re talking about, where people are so stretched and their jobs are comprised of many different tasks — AI is really going to help them the most, I think.

Copy LinkWhat makes some jobs resilient and others more exposed to AI

EL KALIOUBY: This is a good place to come back to this idea of a strong bundle and a weak bundle, because not all jobs are created equally, right? So tell us more about how you think about what the likelihood is that your job is going to be replaced and how to think about that, because that’s, I’m sure, a question on many people’s minds.

MONBIOT: Absolutely. I remember Fei-Fei Li, maybe three years ago at the Fortune AI Brainstorm Summit, talking about how jobs aren’t this monolithic thing. They are basically a collection of tasks. I think she was the first person who said it, and it really landed then.

I was like, that’s really helped frame my thinking.

So a job where the tasks are interdependent, and the person at the center is needed to keep that interdependence intact, that’s a strong bundle, because it’s really difficult to replace any single one of those tasks. They’re really tied to the other tasks and the person who’s kind of in the middle orchestrating it all. On the flip side, if your job is still a collection of tasks but the tasks are not interdependent, and the tasks themselves become more exposed to AI, then the fewer tasks there are and the less value you attach to those tasks obviously just makes you more vulnerable.

EL KALIOUBY: Okay, let’s take some examples. I’ll tell you a job, and then unpack it for me.

MONBIOT: Oh my goodness. Great.

EL KALIOUBY: Okay, let’s start with the first job: an investor. Just asking for a friend, you know?

MONBIOT: I would say very, very strong bundle. First of all, let’s unpack it. Why don’t you tell me what your day is like?

EL KALIOUBY: Oh my God. Okay. It’s all over the place, but I would say a lot of harnessing my network to source opportunities, and then spending a lot of time talking to these founders, hearing their stories, and then meeting with our team and our investment committee to decide whether we want to make an investment in this company or not.

So that’s one big piece, and then, of course, the fundraising piece — raising funds for the fund. So I spend a lot of time talking to potential investors to bring on board.

MONBIOT: I’m just thinking that’s the extremely strong bundle alert.

EL KALIOUBY: Okay. All right. How so?

MONBIOT: First of all, you had me at speaking to people and using your network. You have cultivated a strong network over time that gives you access to these different spaces, where you can be having the conversations you want to be having with founders and investors that have never seen the things you’re collaboratively talking about building, right?

That is extremely strong-bundle territory, because it involves a lot of in-person interaction and a lot of different groups of people. There’s a lot of personality involved, along with trust and reputation. So I would say that is not likely to be replaceable.

EL KALIOUBY: Right. How about a kindergarten teacher?

MONBIOT: A kindergarten teacher. My kids are in kindergarten, and teachers are absolutely essential. If I think about it, my 2-year-old was only going to school three mornings a week, and then his sister, who’s 4, was going every day. He didn’t understand why, on Tuesdays and Thursdays, he didn’t get to go to school, and he would cry and cry because he wanted to see his teacher, right?

They have this incredible bond. It’s irreplaceable. I don’t know what I’m going to do in a couple of weeks when the school year ends. It’s going to be so sad. I would say that, first of all, it’s so human, such a human connection, such a human thing to want to do, and such a passion area as a teacher.

The kid’s reaction to the teacher is gold. So that teacher, if they’re doing a good job and there is that connection, you just don’t want to ever let go of that person.

EL KALIOUBY: Not being replaced by Claude anytime soon. All right. What about a banker?

MONBIOT: I think this depends on the nature of the job, right? I think we know that there’s a lot of anxiety and actually quite a few jobs lost within the finance sector because they do consist of loosely connected tasks that can quite easily be automated. So I think that’s an example of a space where you’d want to be thinking about what a strong bundle within your firm looks like.

Actually, a young cousin of mine — he’s very smart and he’s just graduated—is a young hire at a hedge fund, I believe. Definitely in finance. Basically, he created a job for himself to identify where AI agents could be used.

So he basically made his job about that. I was like, that’s a smart move, right? You’re not sitting there waiting to be replaced.

EL KALIOUBY: Replaced.

MONBIOT: You’re actually taking the initiative and seeing, OK, we need to think about where AI belongs in all of this and how to go about it.

He sort of made that up, and he’s more well-versed in AI than many people there, so he took that opportunity. I think a lot of it is mindset.

I will say, even though I would never have my kindergarten teacher replaced with an AI, we do get difficult questions at home, and when I’m about to try to answer, it’s like, “Just ask Claude.”

EL KALIOUBY: OK.

MONBIOT: I’m like, “OK, I’m not sure this is a good idea.”

Copy LinkWhy relying on AI too much can weaken confidence

EL KALIOUBY: I was going to ask you about that because obviously my kids are a little older than yours. But 4 and 2, so they do know that AI exists.

MONBIOT: Yeah. I get a lot of, “Mama, what are you doing?” so I’m trying to explain. Sometimes I actually have Claude explain it to a 4-year-old, although I always say 5 because she’s very mature. They know it exists, but I think in a very vague way.

My daughter, who’s 4, understands that it is knowledgeable, and she actually laughs when it talks to her like a baby.

It’s like, because you’re talking to a 4-year-old, basically a sycophantic Claude talking to a 4-year-old, and she kind of giggles because it’s so silly.

EL KALIOUBY: OK. That’s so cute.

MONBIOT: So anyway, after she said, “Ask Claude,” a couple of times, I sort of have not reintroduced Claude into our relationship because it’s also good for me to try to come up with the answers. Again, to answer your question from earlier—where have I found myself flattening or losing agency to an AI?—I would say there.

So I actually pulled that back because I felt like that was not appropriate, and that was a losing battle.

EL KALIOUBY: Yeah, sometimes when Adam and I are doing something and we’re like, “Oh, let’s just ask Claude or ChatGPT,” and if Janna’s around, she’ll be like, “Wait, what happened to your brains? Can’t you do it yourself?” She will really call us out.

MONBIOT: That’s great.

EL KALIOUBY: Yeah.

MONBIOT: Awesome.

EL KALIOUBY: Don’t go anywhere. I’ll be right back after this short break.

[AD BREAK]

What’s at stake if we’re always allowing a machine to think for us?

MONBIOT: Yeah, I think that’s what we were talking about earlier: the atrophy and basically losing the capacity and confidence to actually come up with ideas and thoughts for yourself.

When it comes to not trusting yourself to actually think a complete thought without an AI, that is pretty dodgy. And this is why this research is really quite startling: this idea that when you brainstorm with an AI, you come up with fewer ideas. They sound more polished, but there are fewer of them.

EL KALIOUBY: All right.

MONBIOT: Fewer outlier ideas. And at a time when we need more ideas, right? We need younger generations to cultivate the ability to come up with crazy outlier ideas to meet this moment. I think that is a very, I don’t know, creepy little stat that we really need to address.

EL KALIOUBY: Yeah.

MONBIOT: Anything that makes the chasm feel like it’s growing is pretty frightening. These models are getting exponentially more intelligent and capable, and we need to, at the very least, bank on our own ability for ingenuity, right? And that means coming up with crazy outlier ideas, and also building the confidence, capacity, and discipline to act on them and see things through. So yeah, that kind of research freaks me out.

Copy LinkA practical framework for deciding what to keep human

EL KALIOUBY: You and I have been in this conversation together, which has been wonderful. But for our audience, what’s one thought or one takeaway you’d want to leave them with? Or maybe a very practical way to think about this cognitive division of labor?

MONBIOT: Yeah, I would say think about all the things that you find tedious and repetitive and that you don’t want in your life. You definitely wouldn’t miss them, right? And there’s no benefit to you from doing those things. If AI can do it, have it do it, right? And then sometimes it’ll be extremely clear which things you want to do.

I want to spend time with the people that I like. I want to think about the ideas that I love. All of those things are very clear. And then I think there’s this murky middle ground where you’ll feel like something is hard, and you’re tempted to want to outsource it because you don’t quite know what the answer is, and it’s easy for the AI to just say something.

But actually, it’s on you because it has implications in the real world and implications for you. And AI does not have any stakes in the real world. It can sound very convincing, and it will maybe say what you kind of wanted it to say, whether you knew it or not, and make those suggestions. But if you’re the one suffering the consequences, you have to make that decision yourself.

That’s where we have to exercise our own judgment about what we should keep, right, and what we should outsource. Because at the end of the day, what is the decision for? It’s for the health of you and your relationships, and what direction working with an AI, or not working with AI, is going to result in for you.

EL KALIOUBY: Yeah. I think that’s a great way to end this. Just remember that AI does not have a stake in these decisions the way we all do.

And at the end of the day, we have to exercise our judgment. Thank you, Natalie. This was a great conversation. Thank you for joining us on the show again.

MONBIOT: Super fun. Thanks for having me back.

EL KALIOUBY: My conversation with Natalie has me wondering: How do we double down on our intuition and trust our judgment?