Series 3 Episode 2: Will AI harm our critical thinking and automate away our jobs? Let’s put it to the test.

Transcripts are auto-generated and may contain errors.

0:00

Hey, I’m Rose.

And I’m Angie.

And this is Our Lives with Bots, the show where we ask important, timely questions about what it means to live with our bot counterparts.

And from time to time, we also dive deep into what an AI future might look like for us.

0:15

Sometimes we agree, sometimes we spiral, but we always go deep.

This is our second episode of our third series on AI in the workforce and a is impact on people’s relationship with work.

And today we have some more detailed information for you, a lot of research, a lot of blog posts and happy a little bit further into the new year to our viewers as it will be February by the time this episode of releases.

0:42

But know that we are recording this.

Should we date ourselves here?

We’re recording this on the 1st of January.

So I think it.

Shows our commitment.

Yeah, it does.

Like, don’t call us workaholics, OK?

But know that we are experts on talking about this topic because we work a lot.

0:59

Yeah, we like to say current and that we’re doing this on the first day of 2026.

So we’re getting in early.

We are going to be talking about things from the perspective of the people that created the product.

So really big tech, we’re going to be looking at the perspective of academia, so researchers perspective and we’re going to be talking from the perspective of those that use or rather those that sell the product.

1:22

So that’s industry.

And then we’re going to try and speak from the product perspective.

So we’re likely to hear from notebook LM.

And so it’s really those four different voices that we want to see how they are talking about AI in the workplace because we’ll see slightly different nuances that are coming through.

1:39

And today, from the academic perspective, we’re going to have researchers from the US and from the UK and from Europe and also from South Korea adding their voices to this conversation.

So we’re looking at some research studies from those individuals and from the big tech side of things.

1:55

We did cover a little bit of Anthropic in the last episode, in case you missed it.

But today we’ll be talking from Microsoft’s perspective.

So they put out some work on this recently.

And then within industry, we’re going to be focusing on Salesforce.

So they’ve been putting out regular blogs about the integration of AI into work.

2:13

And I know that they’re selling AI products to the companies that they consult for essentially, right?

So they are providers of technological products to different companies.

And so they have a really interesting perspective on this.

But part of the reason we’re disclosing what these perspectives are, where they’re coming from, is that you can take everything with a grain of salt because these are all, at least in industry and big tech individuals who are on the creation and selling side of these products.

2:39

Now, that doesn’t mean that academics are not part of that, but.

But worried I love that you mentioned Salesforce because I think 1 great quote that I’d love to take from Salesforce.

I know we’ll talk about their blog posts a bit later, but I thought this was a really great way to kick today.

I’ll just quote from their blog post.

2:55

As 2025 comes to a close, I’m struck by paradox.

The AI industry has never been more capable, yet the discourse has never been more confused and.

There we go, there’s the theme stuck right?

And just to give you a little bit of a sense of of what the theme is today beyond that starter point is it’s all about skills and specifically critical thinking skills.

3:18

So with AI in the workforce, everyone is talking about critical thinking skills and how they are the foundation for what we do next moving forward, retaining jobs, doing well at your job, and integrating AI productively into your workflow.

And this also rests upon this one key piece of advice, which is don’t be an uncritical adopter of AI.

3:41

With that in mind, let’s go into some of the research that we’ve looked at.

So let’s start off with the Harvard Youth Poll.

So Harvard regularly conducts polls of young people just to get a perspective of people’s perspective of things happening within the US.

So they surveyed over 2000 young Americans aging between 18 and 29.

4:01

And they collected this data in early November of 2025.

They basically found that AI is reshaping how young Americans think about work, and mostly in negative ways.

For some statistics, about 3 to one young Americans think that AI will take away opportunities rather than add to them.

4:20

So 44% think AI will take away, and only 14% say that AI will increase opportunities for work.

And over half see AI as a threat to their job prospects.

And AI is a bigger threat than, for example, immigration or outsourcing jobs to other countries.

4:39

And I thought what was interesting is that 41% thought that AI will make work less meaningful.

So right, there’s the job opportunity side of things, but then there’s also the relationship to your work.

And people aren’t super optimistic about that.

4:56

And does that surprise you?

Somewhat especially since there’s also this metric that they provided, which is that about half of all of these young Americans and 63% of college students specifically trust AI for school learning and tutoring, But then less than 1/4 of them trust it for medical advice, personal life advice, mental health support.

5:19

So it’s interesting because they trust AI to apply in these work esque tasks, but they also recognize or at least just like I guess are aware that this might make my work feel less meaningful and maybe just like in this immediate time period rather than in the long term, but.

5:36

They can see the consequence.

It’s interesting because in other conversations will then have real conversations about how they’re using it for companionship and the impact.

And so it’s like, I wonder if people are seeing AI in the same way across different contexts, you know, where they’re seeing it as all the same AI.

5:52

Do you know what I mean?

I guess everyone thinks about generative AI chat bots when they hear AI, But then I think when you talk about AI and work, people are thinking about these less social conversational applications of the algorithms or AI, just that you don’t interact with interpersonally in a sense, right?

6:13

So maybe when people are asked about AI and work, they’re thinking less about this interaction form of AI, which may be in the interaction form of AI, people feel a little bit more autonomy and control, and so they feel more involved in the process.

I think there is something about when it is the context of work, they’re seeing it more like like an agent that is there to come and do my job or display some of these skills that I could be doing.

6:38

With yeah, yeah, yeah.

And I mean, we’ll talk about that later in terms of collaboration, but I think I think that is right.

I think they are seeing it differently.

And I really love this other article that we read.

So it’s the future of work with AI agents.

And so researchers from Stanford University collected some data in the first half of 2025, and they basically wanted to map out, what is it that people want AI to automate for them in their work?

7:05

And do current AI capabilities meet demand?

I really like this article.

I think it’s so interesting.

This, I think, is a fascinating article.

I see this was a really, really big study.

I think they surveyed about 1500 workers, right?

7:20

But they also spoke to 52 AI experts and so they took a 2 pronged approach.

So they try to understand how people actually experience work and what their experience was and what they were looking forward.

To your point, what are they looking for from AI?

But also they try to understand what are AI’s actual capabilities.

7:37

So what can it actually do?

And what I thought was very interesting is they have this term for the Hass scale, so the human agency scale, trying to understand to what extent do we want automation from these AI agents or collaboration.

So I think they had a 5 point scale.

7:52

So do you want the task to be completely automated to the agent?

Like that would be a one.

Do you want a three, which is like more collaboration?

So it’s almost equal kind of collaboration.

So we’re both working together, you know, me and the agent or A5 and and that’s kind of it’s entire reliant on the human.

8:10

They found that almost 50% was it like 46% were very positive towards having a degree of AI automation, even though it meant that they might have some job loss.

So when we talk about this 46%, this means that of the tasks that the workers do in their specific domain, they were OK with some level of AI automating that task.

8:35

So almost half of tasks, they were like, yeah, AI, it’s cool if it automates it to some degree.

I don’t have to be fully involved and I’m not exactly looking for AI to be not involved at all.

They looked at 844 different tasks across 104 occupations.

8:52

So general domains, there’s four domains, I think mathematical and computing arts, media, et cetera, management.

And then what was the other one?

There was one other business and financial operations.

Within those 4 domains are 104 occupations, and they looked at 844 tasks across all of these domains and occupations.

9:13

From this human agency scale.

They’ve asked people to what degree they would want AI to automate certain tasks and their job, and people were OK with some level of AI automation for 46% of of all of the tasks across all of these jobs and all of these domains.

9:29

And why?

Well, they also asked why it is that workers wanted this degree of automation for the specific task across all of these tasks.

And the number one selected reason was that automating the task would free up my time for high value work.

9:46

So maybe things like more creative work, more enjoyable work.

And then I think this is helpful to think about in contrast with like the next highest selected responses.

So it’s like, you know, freeing up time for high value work is the number one reason.

And the second reason is the task is repetitive or tedious.

10:04

Makes sense.

Automating this task then would improve the quality of my work.

The task is stressful or mentally draining or the task is complicated or difficult.

So the number one is high value work and then the next we’re like, it’s just tedious and challenging.

Not worth my time perhaps?

Difficult.

10:20

That’s concerning.

Yeah, but that was a lower percentage.

So if you look at the discrepancies between those regions, you look at the bar and I’ll put it up on the screen.

The bar for automating the task for free of time for high value work is like this.

Is this Y?

And then the task is complicated or difficult is about 1/3 of that.

10:39

And I guess the question is because we hear this a lot anecdotally, you know, we want automation so that it can free up time.

And I’m just curious to see whether that actually turns into real time for better work or what what the time gets used for.

10:54

And, and I think that like that’s the study of the decade that needs to get done.

I.

Think really keen to see that.

I can see that worry where it’s like, yeah, I really want to have all this time for high value work.

But then when it comes down to it and people are offloading all of these tasks on to AI, they realize first of all that they’re like, this is great, not working as much.

11:13

But second of all, that also just leads to a decay in certain cognitive skills due to habitually relying on a tool.

So it’s like their motivation and skills to do the other tasks may decrease.

So they wanted to look at the landscape in terms of workers preferences, but then actual AI capabilities.

11:31

So they asked the question where workers want automation, do current AI capabilities based on 52 AI experts reviews and AI capabilities with these jobs and skills and tasks, Does AI meet the preference?

Does AI meet demand?

11:48

And I thought was fascinating.

So they have what’s called a workbank database.

So this is preferences from domain workers and AI capability assessments from AI experts.

And they have 4 different zones within the work bank database.

So they have automation green light, which is people have high desire for automation for this domain and AI is capable of meeting that preference demand.

12:11

Then there’s the red light zone, which means people have very low desire for AI automation in this domain, but AI is very capable of doing so.

So it’s like, OK, I can do it, but people really don’t want it for that.

Then there’s research and development application zone, which is people have a high desire for automation in this area.

12:29

The AI capabilities just don’t meet that threshold yet.

They don’t have the capability to meet those demands.

And then there’s a low priority zone, which just means low on both.

People don’t want it and AI can’t do it.

So whatever.

If you look at the spread across these four different areas from this Workbank database, based on their data of of preferences and capabilities, there’s a lot of mismatch, is what these researchers are saying between what people really want and what AI can do.

12:58

What I thought was also interesting, what they picked up there is that where the current investment is going, if you look at it like through a proxy of where Y Combinator companies are, is also not aligned to those 4 quadrants.

So I think they found that 41% of Y Combinator companies are concentrated in the low priority zone and the automation red light zone.

13:21

So 41% of capital is being invested into places that are not going to generate the return because actually we either don’t want it or it’s a red light zone.

It can’t be done.

We don’t want it to be done.

So that doesn’t.

Make paper encompasses the landscape so well, right?

13:38

Oh well, so well.

Interestingly, they do reflect that the research is being done on the right areas, which is great.

Which again, is a nice way of pointing to the gap between academia and and industry and.

One last thing about this.

Of the four domains, people who were workers in the computer and mathematics domain, they expressed the largest amount of tasks for AI automation versus people in the arts, design and media domain expressed the lowest amount of tasks that they wanted to be automated.

14:09

So of the mathematics and computer domain workers, they were OK with about over half of the tasks they do for the work to be automated to some degree by AI versus those in arts, design and media only were OK with AI automation for about 17% of the tasks that they do.

14:27

So once again, why is AI creating videos, art, whatever when people in the creative industry are like no.

Well, I well, exactly, except I think what was also shown there is that’s what the workers were saying, whereas the experts were saying, but you can and you should, the alignment wasn’t there, right.

14:51

I think what the paper was also pushing towards is that this collaboration zone that’s coming through, there isn’t an alignment.

And so you’re going to have experts saying the AI is going to be able to take over.

Workers are saying we don’t need that.

We don’t need that, like you say in certain domains more so than others.

15:09

And so is that going to create friction as the AI capabilities exceed what people desires are for beyond collaboration?

Yeah, and I think the findings from people in arts, design and media, in the actual industry, I think of course this is something that we’ve seen before with the whole question of creativity in the age of generative AI, is that people who are actively involved in this creative process value the creative process itself and value their ability to do it.

15:39

So when the door opens for all of these other people who aren’t creatives to be creative with AI, there is this threat to kind of the uniqueness of the group of creatives.

So I think that’s also.

A play definitely.

So just to share an experience, this is a paper that I thought this would be a really interesting opportunity to see how would I use AI in our work.

16:05

So Rose, you’re familiar with the AI tool Notebook LM and I didn’t realize just how busy they’ve been at developing that product.

So I first used Notebook LM.

This was probably about a year ago.

Maybe you put an article or you know, piece of writing into it and it would pop out a podcast.

16:22

So 2 podcasters like us having a chat and basically it would take a long article and put it into like a 10 minute podcast which we can listen to in a moment.

Now it’s got loads of things which we’ll share with everybody.

You can turn into a presentation quiz, flash cards, infographics, mind maps, all sorts of things.

16:45

I think it would be worth sharing with our viewers and listeners the presentation that it created, because it shows you just how we’ve still got a little way to go in terms of recreating some of the images that are actually provided to the notebook LM, which is disappointing.

17:00

Within this workflow, did you use notebook LM and listen to the podcast it put out first or did you read the paper first?

How did you go?

Yes.

So let me tell you what my process was.

First of all, I put the paper in, but I didn’t listen to it.

I mean, you sent me the paper about a week ago when we first said we’re going to do this one.

17:18

And I quickly wanted to put it in and I sent it to you.

But I was like, I am not listening to this until I’ve read the paper myself.

Because I was concerned that I wouldn’t be able to judge whether or not it was hallucinating with there were errors in what it was saying and how I would be able to consume that information.

17:37

Because it feels like once you’ve heard something, it’s hard to kind of unhear it and go, well, that’s correct.

That’s not correct.

So I was very concerned that I wouldn’t be able to test the veracity of what I was hearing if I hadn’t read the paper first myself.

And we’ll talk about it with something else that I tried a bit later.

17:52

That’s how I approach these things.

I would imagine that most people would probably listen to the podcast first if they’re going to use the podcast.

What do you think?

Well.

This is what I want to get on to.

We’re talking about AI in work, and I actually found that particular paper really valuable from a practitioner point of view.

18:13

As you know, I’m a practitioner, right?

So I’m often curious as to how we can get some of this insight to practitioners and how they can make use of this in industry.

And I don’t mean big tech necessarily because they’ve got researchers, they hire researchers, but certainly to like other big companies or small companies.

18:32

How can they learn from this research?

Because let’s be honest, you’re not both, but you’re in academia.

I’ve written academic papers.

We don’t write for your average person who’s got a corporate job.

It’s just not our target audience.

To be fair, it’s it’s not the plan.

18:48

So then I thought, actually something like notebook LM is a very good way to help business people consume the information.

So to your point, do I think somebody in industry or in business is going to listen to the podcast and then read it?

19:05

They’re not going to read the paper at all.

They may go to it after that.

They really want to get to the detail.

But actually the hope is, would this be a way to disseminate really valuable insights from a paper to people that could most use it?

And that’s the question, did you?

19:22

Find Yeah.

Did you find that the Notebook LM podcast was accurate to your understanding and reading of the paper?

Were there any areas where it wasn’t?

I felt it was thin.

It was accurate enough.

Putting on a business hat, would I be able to take out enough value that I would get enough right for business, bearing in mind I’m not having to cite it for a paper?

19:45

I guess it’s like if you want to get maybe the gist of a paper, just get the overview notebook LM can be helpful.

Or maybe it can be helpful for figuring out whether you think it’s valuable to go deeper and get more insights out of it.

20:01

I think that would be a decent use of it, especially if you are in industry.

And bearing in mind it’s 14 minutes, you’ve got somebody like on the train, you’re listening to it as a podcast.

Great.

Would I trust it?

Like if it take 46% of tasks versus people, would I trust it?

20:19

And like if somebody said I read this paper, I’m like, no, you didn’t.

No, no you didn’t.

You know what I want?

I want an AI that will be able to read APDF of a research article out loud to me and skip all of the parentheses, citations and the page page numbers.

20:39

Because PDF readers.

They do not skip over the unimportant things on the margin of the page.

They will tell you what’s on the margin of the page and they will tell you what’s in every citation after a sentence or within a sentence.

So you’ll be hearing AI is changing work ( 15 name of researcher’s date, name of researcher’s date.

21:01

And you’re like God.

Exactly.

We shall we listen to a bit so you can get a sense of what it’s like because it’s, it’s quite weird.

Their latest version.

You can also interrupt it so we can talk to it.

Very odd.

We won’t do that.

We’re also going to point out any inaccuracies that came from the Notebook LM podcast version of that study.

21:23

In just a few moments, you’ll hear a few clips from a 15 minute AI generated podcast about the article we just covered.

To generate this audio content, all Angie had to do was upload APDF of the paper into Notebook LM.

Between these clips, Angie and I will also provide commentary on what we found misleading and accurate or strange about this generated podcast.

21:45

If you want to listen to the full podcast, just click on the Notebook LM link in our show notes.

If you’re listening to this, chances are that within the last, I don’t know, few days you’ve interacted with a sophisticated goal directed AI agent.

Oh, for sure.

22:01

And we’re not talking about simple chat bots anymore.

We’re really talking about these compound systems that can execute multi step plans.

Exactly.

Systems that draft complex emails that can summarize dense legal documents or even, you know, generate entire blocks of code.

22:18

They’re.

Already everywhere.

And they’re, I mean, they’re fundamentally reshaping the labor market.

And that reshaping is happening so fast.

I mean, prior research, it really sounded the alarm.

It suggested that roughly 80% of US workers could see large language models LLMS impact at least 10% of their daily tasks. 10% that’s that’s an enormous segment of the economy facing this rapid tangible change.

22:39

It really is.

Absolutely.

But when you hear a statistic like that, the big sort of looming question everyone keeps asking is, are these AI agents automating jobs away, or are they augmenting what humans do?

And, you know, general exposure analysis can only tell us so much about that.

22:55

That’s why we’re diving into a landmark new study from Stanford, which uses the the newly created Work Bank database.

And the core mission of this research was while it was critical to move beyond the high level fear and the hype, they performed a really rigorous task level audit to capture the one thing that has been missing from almost all AI analysis, which is the nuanced perspective of the worker.

23:19

The worker themselves.

Yes.

I mean, when we talk about AI integration, we often use this very binary language.

It’s either full automation or no automation at all.

Right, it’s one or the other.

But the reality is a spectrum, and if we want to successfully integrate these agents, we need a shared language, one that defines collaboration, starts and where it ends.

23:38

That’s why they introduced the Human Agency Scale, or HASTE.

OK, so this is the central architectural piece of this whole research project.

It is.

It’s defined by 5 levels, H1 through H5, and it’s a tool that quantifies the required degree of human involvement in a task, deliberately centering human agency rather than just the machines capability.

23:59

Right.

So if you look at the extremes, H1 sits at one end of the scale.

This is full automation.

The AI agent handles the task entirely on its own, with absolutely no need for a human.

OK, so total hands off.

Total hands off.

And then on the other end, H5 is where the AI agent just cannot function without continuous human involvement.

24:17

The human is essential for, you know, steering, correcting, completing the task.

And the level that that kept popping up in the data, the real sweet spot that sort of defines this future of augmented work is right there in the middle H3.

H3 It’s defined as an equal partnership, and this isn’t just handing off a task.

24:36

It’s the AI agent and the human collaborating closely to perform a task.

And crucially, they outperform either 1 operating alone.

So it was a synergistic relationship.

It is.

That’s where the maximum value is created, and this is what truly stands out in the data.

They said they found this, but they all just they just inferred this.

24:55

So that’s the one piece, and I think the other big piece is what I noticed is that the female voice, because not a person slips into.

We found this and we found that a misattribution.

We didn’t do anything.

We saw a really clear disconnect between worker pain points and existing usage patterns.

25:14

Current usage lags dramatically behind actual worker needs.

And that usage disconnect is mirrored in the investment world, isn’t it?

It is.

We used Y Combinator companies as a proxy for startup investment concentration, and the findings show a massive misalignment.

The researchers did it.

25:30

So what are you talking about?

We.

That was weird.

I think at the start the female voice was talking as if she was a podcast host.

But then without identifying, oh, I’m now the researcher of this study, it started to go into we.

25:48

Yeah.

And that hasn’t happened before on something I’ve listened to that they’ve done.

No, no.

OK, that’s interesting.

Yeah.

But anyway, you’ll be able to take a look at that and I think this is really a relevant time to point out some interesting perspective articles that have come from the academic side of things about using AI in the workflow.

26:09

And specifically, we’ll start with one about using AI in scientific research.

This paper was conducted by Lisa Masseri and Crockett in 2024, so it’s about over a year ago now.

And it’s called artificial intelligence and Illusions of Understanding and Scientific research.

26:26

And I think the clue there is and the illusions of understanding, which I think is very, very valuable, kind of signposting what they’re trying to talk about here.

What they were concerned was specifically was how the widespread adoption of AI tools might impact scientific knowledge production and understanding.

26:41

What they point to is that almost as the more we think we know, the less we might know, which is kind of like this, this paradox that happens.

They created quite an interesting taxonomy, but within this AI space, kind of looking at the different roles that AI might play.

Of the ways that people are applying AI and scientific research that kind of fit into these 4 categories.

27:01

So this is the taxonomy of four categories of how people apply AI in scientific research.

Exactly.

And I thought that was quite cool because I had this taxonomy where they called the AI as Oracle, AI as surrogate, AI as quantum, AI as arbiter, AI as Oracle is, You would be able to go to AI and you would be able to find out anything because there was basically too much knowledge to get in science.

27:25

And AI would be able to kind of disseminate and aggregate all about knowledge and you could go and find out the answer from AI.

Great.

Right.

So it’s like instead of doing a literature review on your own of seeking out, reading, analyzing, synthesizing a ton of scientific literature and making hypothesis about the next study that you might do what you might find.

27:46

AI is the Oracle that can do it for you.

All sorted and then data collection can be sorted too.

So collecting data is just too difficult.

It’s time consuming.

No need, AI is surrogate.

We’ve got that covered.

So AI can create data for you, proxy data, creating synthetic data essentially.

28:04

Instead of getting and a bunch of humans to respond to your survey, you can just make up a bunch of AI agents that are supposed to be representative of certain groups of people.

For example, middle-aged women.

This AI is operating as a middle-aged woman and is answering the survey in line with that identity.

28:22

There’s been some interesting research on trying to apply this and especially minorities are not accurately represented and very homogenized.

So if you want to imagine, I wanna get the perspective of people who are middle-aged in India.

And then the survey that you would get from AI as the surrogate is extremely flattened like you’ve said in many of our episodes, extremely homogenized and is very stereotypical.

28:48

So.

Yeah, and you can see why people might do this.

Like if you’re in a rush to get data, if you don’t have the funding, this is very appealing.

But the risk, and we’ll talk about what the risk is in terms of creating what they talk about monocultures, but we’ll come on to the So what?

29:04

Then there’s AI as quant.

So data analysis, data sets are too large.

You know, we’re just fallible human beings.

So if we can use AI to crunch all the data, then there’s AI as a quantitative partner, right?

And in fact, not really a partner.

It’ll just do it.

Right, the article also talks about it’s not just the AI doing data analysis but also making inferences about the data so pulling out trends that the, like you said, fallible human person simply cannot find.

29:31

Like, we just haven’t been able to do this before.

You know how we’ve gotten to this point that we ever developed AI as a mystery?

Truly.

OK, so what’s the last one?

AI is arbiter.

So again, we’ve created a word for our own back.

There are too many papers and proposals to review because AI is creating them now, and the reviewers are biased, again, fallible humans.

29:53

And so we’ve got tools that can objectively do this work for us.

So AI can come as arbiter to kind of settle these problems that we can’t seem to solve for ourselves after creating ethics and frameworks like that as human beings.

But now we have AI to solve these problems for us.

30:10

It’s like kind of circular, right?

So we’re creating problems by using AI.

So we’re using AI to solve these problems.

And what the researchers of this article say is that so much of this is about saving time and reducing the fallibility of human bias.

30:27

I understands time, but time is so often used as an excuse to apply shortcuts.

So I like that they brought that up.

OK.

So with this application, what are some of the risks on the scientific process of using AI in these ways?

30:43

So their big sort of take home and what they say is the big risk of this is the risk of creating monocultures.

So scientific monocultures, and I really love this.

A friend of mine did a lot of work around monocultures within AI.

When you talk about monocultures, it’s kind of creating very narrow views.

30:59

And the risks that can happen here, they build on monocultures from agriculture.

But the risk of kind of marginalizing and narrowing either the questions we ask or the people that get to ask the questions, the methodologies we use, we don’t even use real people, right?

31:15

So it’s not like we’re even gathering data from people anymore.

So it kind of makes it very, very narrow.

And therefore we don’t get this innovation anymore.

So we narrow our methodologies, we narrow the kind of knowledge we can create.

And so it’s almost like an echo Chamber of the kind of scientific knowledge that we’re able to generate.

31:33

And I think what they’re talking about there is that the risk is that we don’t even get to see that we’re doing that.

That’s the illusion part, right?

So we think we’re creating this almost better knowledge because now we’re using AI, we’re not even involved in that knowledge creation anymore.

So we don’t realize that there’s a flattening and a dumbing down of the knowledge that’s being created.

31:53

Isn’t it so ironic because generative AI, generative in the name, is this idea of creating new ideas rather than just being at baseline totally a predictive algorithm that completes a sentence for you.

It’ll generate new ideas.

And it’s actually reducing more, but with less creativity, less innovativeness, more errors, and very systemically predictable errors, right?

32:19

Predictable hallucinations, predictable, whatever.

OK, something I also want to point out, I know that the whole scientific monoculture is much deeper than this, but I have to say that there is a somewhat scientific monoculture, at least with the introduction paragraphs of research about human eye interaction, including this paper.

32:39

All of the papers that I read start with AI is becoming more ubiquitous, more people are using it.

And this may sound like science fiction, but every single paper starts off like that.

And I’m, I’ve done that before too, before I noticed that everyone else was doing it.

32:56

So we all think we have an original idea, but we’re actually just all homogenized when we’re writing about human AI interaction.

And we have to fight against it.

I thought it was really interesting actually that they talk about pre AI like, so if you talk about like the history of science, almost like this umm, journey that we’ve been on to move away from monoculture within scientific research.

33:17

And I don’t think it’s only in science, but kind of moving and trying to get more diversity of voices within academia generally.

And the risk that this, as you say, generative AI should take us into generating new stuff and the risk of taking us straight back down there just seems so ironic.

33:32

And also one last thing that I want to mention about this paper is I love this paper.

I think it’s great really for mapping out the space and specifically within scientific research and AI.

But I was asking the question when I started reading this, what is responsible AI use look like then?

33:48

Are the researchers going to talk about this?

And they say that they’re going to talk about it in the discussion section at the end of the paper.

So they say the challenge is remaining aware that increasing productivity does not guarantee an improved understanding of the world.

And they also say we conclude with several suggestions for navigating this complex trade off.

34:09

No real suggestions are provided other than reiterations of what to worry about.

OK, then what is not quite the answer, Yeah.

No, no, there’s a lot of like.

Just don’t get it wrong.

Yeah, and there’s.

Let me get.

It wrong, but how do you get it right?

34:26

Yeah, but don’t, just don’t do that.

Yeah, just throw away AI out and don’t like that and use it at all, Yeah.

Yeah, right.

I I agree that that was a little bit short on solutions, but but short.

But also points to a really interesting article that just came out as a preprint in 2025.

34:42

And this is from university researchers in Italy, Slovenia, and South Korea.

They coined a new term called epistemia.

And this is the idea that AI agents, so they use linguistic probability to provide you with outputs.

And they say that linguistic probability substitutes for epistemic evaluation.

35:01

So critical thinking evaluation, deeper evaluation, so that essentially people who use AI can produce the feeling of knowing without the labor of judgement.

So this is their new term that they’re coining within the age of human interaction and using it for certain tasks.

35:18

They said epistemia is not reduce full to user naivety.

It is an architectural and socio technical phenomenon that arises when generative systems deliver finalized fluent answers in context where justification, uncertainty and revisability are essential.

35:33

As generative models scale and become more persuasive, the risk is not only that errors persist, but that evaluation is systematically displaced.

The epistemic workload is shifted from the system to the user, and more importantly, made easier to omit.

35:50

I like it because it’s pointing out that this is not a user problem, it’s a model design problem and it requires A socio technical approach to solve.

It.

Agrees, and it mentions how we need more than just better models, and it requires maintaining the social and institutional conditions under which reasons can be demanded, errors can be owned, and belief remains answerable to evidence.

36:16

So yeah, don’t take AI outputs at face value, no matter how tailored, beautiful, whatever accurate they seem, put in some extra effort to judge those outputs.

Don’t be biased by how pretty it looks.

Yeah, it goes back to what you said in the beginning, right?

36:32

Skills, skills, skills.

Critical thinking is what’s going to matter.

I read this and did some work on it.

So here is the other place where I kind of thought, all right, let’s put on a business hat.

This is a business person who’s got very little time.

36:48

Imagine my good friend Rose has sent me this article.

It’s like this will help you in your business context because this is our work series, right?

You’re doing the ethnography part of things.

I love that you did this.

Meanwhile, I’m like, yeah.

I mean exactly So what would somebody who is time pressured, not reading like 20 syllable words with 16 line sentences wanting to make sense out of this because I actually think there is a lot of value for work in this article.

37:19

But like what?

What is it?

So I didn’t even look at notebook LM because I thought these days many, many companies, people at work are looking to ChatGPT.

So I thought, let me look at ChatGPT.

So having read the article as well and like you have a strong interest in cognitive psychology, I was particularly interested in epistemia and said to Chachi PT.

37:41

Now let’s talk about epistemia and how it’s different from intuition.

And it gave me a mouthful on how it’s different from intuition, which was quite interesting.

And it tried to say true to the author’s point of view.

And it turned this article and tried to give me insights from this article that are relevant to work.

38:01

So this article talks about the seven fault lines and talks about about how you know, you have the grounding fault, you have pausing fault.

So we won’t go to all the different faults.

You know, if we talk about the grounding fault, for instance, it very quickly gave the example about like humans draw unlived social and situational contexts, whereas AI only processes text.

38:20

So at work, AI would miss tone and power dynamics and situational context, for example.

So it’s able to very quickly help kind of situate this article in a work context.

But from epistemia point of view, specifically when you think about work in organizations, the risk is people will hear all of the stuff.

38:41

And you know, this term work slop that we’ve heard before, it didn’t talk about work slop, thank goodness.

But actually when you think about the risk at work is that AI will just give you something.

So it I asked for a specific example of epistemia in the workplace context and it understood the term exactly based on what I had read it to be.

39:00

And it’s quite a nuanced term.

It was able to simplify it in very, very good ways.

It gave me a very specific example.

So I’m going to try and look for it and share it with you because I thought this is quite useful for people that are in the workplace that might be like listening to us and still be like, so just say so it says, I love this, a clean example of epistemia.

39:21

I did feel a little bit patronized to be honest, but anyway I dealt with it.

Such as scenario a manager preparing for difficult people decision.

A manager types into an AI system.

Summarize the key considerations when deciding whether to place a high performing but disruptive employee on a performance improvement plan.

39:40

The AI returns a fluent, structured answer.

Balance individual performance with team impact.

Consider prior feedback.

Document behaviours.

Ensure fairness and consistency.

Align with company values, right?

Yeah.

The answer is coherent, sensible, not obviously wrong, even ethically framed.

40:03

The manager reads it and thinks, yes, that’s exactly it.

They proceed, feeling clearer and more confident.

Where epistemia enters.

Nothing failed.

Technically, nothing was hallucinated, no misinformation occurred.

Yet epistemia has happened.

40:19

Why?

Because the manager now feels they’ve understood the situation, has experienced epistemic closure without engaging in the labor of judgment the decision requires.

That’s good.

It’s not bad.

Very flowery.

Maybe I’m just, maybe I’m just attached to the floweriness and the beauty of that sentence.

40:37

But it tells you what epistemic work was bypassed.

The AI did not experience interpersonal history, that sense team dynamics, way emotional consequences, feel the moral weight of harming or protecting someone, anticipate downstream relational fallout, calibrate uncertainty, or say this depends in ways I can’t see.

40:55

The manager meanwhile, may now substitute the air fluency, et cetera.

So I guess my point there is it shows well in a work context, the risk of how AI creating all of this stuff takes away.

And it’s a little about that cognitive deskilling, which I think you’ll share in a moment, how that cognitive deskilling or at least maybe not taking away the skills completely, but kind of like absent cognitive and judgment making and us thinking that we’ve created knowledge or thinking that we’ve created some sort of judgment just isn’t there.

41:24

And there isn’t that epistemic scaffold that’s required in a workplace context.

Yes.

And as we’ll talk about critical skills are what industry is looking for, especially for entry level jobs.

We’ll get to that in a second, But looking at the cognitive deskilling part of things, so a researcher of the Swiss Business School in Zurich conducted a study of UK participants.

41:46

So over 650 UK participants were surveyed and did 50 interviews to understand how AI use related to cognitive offloading.

And I’ll define that in a second and critical thinking skills.

So cognitive offloading is when you engage with your external environment or an external tool to help alleviate your working memory.

42:08

I’m engaging in cognitive offloading when I have a bunch of my passwords saved on my computer instead of in my head because in my working memory, my working memory is what I’m able to remember at this moment in time.

I can’t remember all of the passwords at this moment in time as well as go about daily life.

42:26

So in any case, I’m cognitively offloading the task of memory onto an external agent, in this case a computer.

So what this researcher found, and this researcher collected this data sometime in 2024, but the paper is published in 2025.

So it looked at AI tool use and cognitive offloading and critical thinking, right?

42:46

And what they found is that younger participants, if they had higher AI tool use and they cognitively offloaded onto AI, they had lower critical thinking skills.

And that was just the majority of younger participants, whereas the majority of older participants, they had lower AI tool use, less cognitive offloading, and higher critical thinking skills.

43:10

And you might think, OK, people who are older have had more time to develop critical thinking skills, so they’re just going to higher critical thinking skills.

And this researcher did find that age as well as education level was related to critical thinking.

But when you look deeper into the data, what this researcher found is that the strongest predictor of critical thinking was AI use.

43:31

Even if you looked at the influence of age and education, AI tool use was the strongest predictor of people’s cognitive skill.

And this researcher also found that it really requires people to cognitively offload.

So if people are using AI, but they’re not really cognitively offloading on AI, they’re still retaining their own cognitive processes.

43:49

They’re still engaging in those cognitive processes.

They’re not going to have lower critical thinking skills.

But if people use AI and cognitively offload more onto AI, then they will have lower critical thinking skills.

So long as people are actively engaged and still cognitively invested that that kind of drop off isn’t happening.

44:08

Is that kind of what you’re saying?

Yes.

So that is what I’m saying.

And I think the findings from Big Tech really under score this.

So, Angie, do you want to talk about the Microsoft Research?

Yeah, it was majority Microsoft and again this was done earlier in 2025 which last year and I think they looked at about 320 knowledge workers from within that domain and they had participants share their first hand experience of using genitive AI in work tasks specifically.

44:36

And I think they gathered about just over 900 examples of them using Gen.

AI in those tasks.

Both the quantitative and qualitative data suggests that people that had like higher confidence in genitive AI was associated with less critical thinking.

They weren’t looking at critical thinking in terms of the scale that measures people’s critical thinking.

44:55

They were asking workers to indicate how much effort they had to put into critical thinking for their workflow.

So if you have a higher confidence in the tools themselves, then it’s almost like those people weren’t doing as much critical thinking themselves.

45:11

It’s as if the tools reduce the perceived effort required for those critical thinking tasks amongst those knowledge workers.

From an effort perspective, it’s almost like if you have higher confidence, then you think it’s less effort because you think, well, I don’t have to put in as much effort because I, I kind of think that the tool is doing it for me.

45:31

Great.

But conversely, uses self-confidence is a subjective measure of their knowledge and their experiences.

So this is kind of different.

So what they found there is that the higher their self-confidence that they had that was associated with more critical thinking.

45:48

And again, to your point, critical thinking being effort, right.

So if they had more self-confidence, they thought that they were having to put in more effort, even though, yeah, they were confident in their own skills.

They perceived they had to put more effort into the tasks.

46:04

But is that then separated out from their belief in the AI?

How do you see?

That it’s like, OK, if I have more self-confidence, I perceive that I put more effort into critical thinking for the tasks that I’m doing.

And then when people doubted their ability to perform their work tasks independently, they tended to accept generative AI outputs by default for the reason of, I don’t think I’m able to engage critically with this at all, even on my own.

46:35

So I’m going to take the generative AI outputs at face value.

And similarly, a lot of other people within this study perceived AI as always truthful and of high quality, which I was like, Oh no, it’s not.

Is that a sample bias?

So the people that they sampled, they sampled knowledge workers, but they sampled them from Prolific, from a crowdsourcing platform.

46:57

I don’t know that it was with, but also I don’t quite know what their definition of knowledge workers is and who gets qualified as a knowledge worker.

They did require if someone wanted to take the survey, the person had to use Gen.

AI at work at least once a week.

47:13

So there’s a little bit of a sample bias.

Right.

Yeah.

But back to your point.

So I mean, it was about task confidence.

So if people were confident in the task itself, that did influence how the users would engage with the AI tools and especially with regards to how much they were willing to collaborate with the AI.

47:32

If somebody has high task confidence, that was associated with users feeling willing or able to delegate tasks effectively.

And it’s a little bit like what we were talking about.

If I’ve read the article and I feel like I know what’s in the article, then I’m much more willing and I I feel that I can maintain some accountability.

47:51

Then when I’m working with ChatGPT, I feel like I can maintain that accountability when I’m using the tool because like, I know that’s not right, that’s right, that’s not right.

Whereas if I haven’t read the article, I don’t know what that domain is.

I don’t have that sense of mastery over that task.

48:08

And so I can’t maintain that accountability.

And I think that’s one of the things they found.

So the people with low self-confidence, as you said, they do rely on it, but it diminished their critical engagement.

To your point, they’re like, well, I’ll leave it to you, you know better.

Anecdotally, in my own experience, I don’t use AI that much now.

48:27

I use a lot at the beginning when it first came out to see how it worked.

And I think I was more impressed with it when it first came out.

You know, novelty effects and all that makes sense.

I found it especially useful for coding because I struggle a little bit with dyslexia.

48:43

And so typing out code, seeing parentheses, seeing commas, very difficult for me.

So ChatGPT was very helpful for coding initially.

But when it comes to using it, for example, to explore and another domain, right?

I’m an expert in psychology and some ethics and human air interaction, but there were times where I needed to have some level of knowledge or understanding of different areas.

49:09

So for example, clinical psychology or health and well-being research.

So I had to look at a lot of that for some of my chat bots as social companions research paper.

And when I was doing a follow up to that study, I was looking deeper into this health and well-being literature.

49:26

And what was difficult for me is I really didn’t know which journals were considered reputable by that domain.

And I wasn’t sure what sorts of methodologies were considered reputable.

So when I was using AI to try to map out this space, I really didn’t feel confident in knowing whether or not what the AI agent telling me was accurate.

49:45

And so it felt very uncomfortable.

And I just ended up not using it for that because I just, I wasn’t able to vet its outputs.

And I think that.

Not many people do that, and I think that’s one of the things I’ve heard people say.

So again, anecdotally, is that it is most powerful in the hands of experts, you know?

50:01

So if you know what you want out of it, you can really use it.

But if you don’t, then you’re more likely to make more of a mess or you’re just not able to get stuff out of it that you want.

You can’t trust it and I think that’s the risk.

That leads us right into some of the research by Salesforce on needing AI training.

50:20

So one of their blog posts, this came out in September 18th, 2025.

It’s all about AI training within the workforce and workforces readiness.

They surveyed 14,000 adults from 13 countries across the world and they found that most people in the workforce support investment in AI related training, but only a third.

50:40

I think their workplace will get that done, get that training done within the next 12 months.

And only 29% think their workplace invests enough in AI training.

And less than half of all workers say their current workplace is prepared to use AI tools in daily work.

50:56

However, India and Saudi Arabia are leaders in this respect.

So over 3/4 of workers within these countries said that actually our workplace is prepared to use AI tools in daily work.

So there is a discrepancy across this.

But I mean, AI training is needed.

51:12

And I get it.

AI is so new, but right right now it’s a situation where everyone’s being told to use it up front.

That’s step one.

And then step 7 is like, oh, we need training for this.

I expect to see these results when we do our minute study, but it is worth reminding ourselves of the Providence of this research, you know.

51:35

So Salesforce, I think we’ll sell this training or at least we’ll sell these AI tools just to know our mind.

You know you all, you all say that you need AI training, but your workplace isn’t giving it to.

You but.

I.

51:52

Do think this is likely to be the truth or at least to be a strong representation of what is actually happening out there?

What did you think of the five trends that they put out?

Because I thought that was interesting and obviously we’ll share it in the show notes.

So Salesforce also put out a blog post December 18th.

52:07

This is five trends that will transform business in 2026.

So they’re looking at what they predict will be AI design advancement changes that will transform business, and one of them is what’s called ambient intelligence.

It’s just rise.

Love the language, rise of ambient intelligence.

52:26

And ambient intelligence is intelligence that is always on.

It knows when to act.

So it’s the idea that AI will be able to act without needing a prompt to do so.

So it’ll shift from ask and receive, user asks and receives from the AI to AI anticipates and delivers.

52:47

So instead of having to deliberately talk to the AI or choose to consult it, it’ll automatically give you a suggestion.

It will already be there.

Which, sure, you know, if you’ve got agents and you’ve got a customer service center or you’ve got a customer support model and you’ve bought us and Salesforce will be selling these kinds of tools and you have ambient intelligence that is with you all the time and is proactive, not reactive.

53:13

Because I mean, these are just marketing words, just dollars worth marketing words.

That’s great, right?

It’s responding without you even having to ask.

It’s literally making money for you without you having to ask.

It’s generating customer value.

But my goodness, that sounds terrifying.

53:30

And I love the Ambient Intelligence title that they give this trend, and I would like to retitle that to human agency not required.

What do you?

Use, yes.

I mean, that’s entirely it, especially when you look at one of their other trends, which is around agent to agent orchestration.

53:48

If you want to buy a car, for example, your AI agent can negotiate with the car dealership AI agent, which will negotiate with the sales agent and the whatever agent and like you can just sit back and let it happen.

So again, we have agents, other people who have either sales force or let’s say third parties that have an SDK to sales force, so we can integrate with sales force, no doubt great.

54:16

And they have an agent.

They have an agent that integrates with our agents, and so their agents, because agents need to sleep, would be able to do work while you’re sleeping and you’re getting time back.

Remember that time for more value, things like spending time with your family and sleep and so we can do all of this work on your behalf, which is great between businesses without you human agent ever being involved.

54:42

What I did love is that they then used agent for the computer agent and the human agent interchangeably.

So at times it was hard to figure out which agent they were talking about.

It’s so funny.

Again, the title.

It’s the semantic layer for agent collaboration.

54:57

So agents will interact with agents.

And I would again like to retitle this, which is human not required for collaboration.

Normally it’s called human AI collaboration, but this is AI AI collaboration.

Human is no longer in the loop.

You don’t know what’s happening.

55:13

How do you make sure that your values and your needs and your desires are being implemented?

And how like do you oversee it?

Or you’re just like feel like most people do not oversee it.

They’ll be like, that’s good enough.

Well, bearing in mind you’re not going to drive the car, So what values are important here?

Because not going to be.

That’s true probability.

55:30

My AI will drive the car for me as well.

And then it’s like, oh, and then the whole fear of the robot revolution will occur and some robot will tweak the system so that everyone’s AI bot that’s driving their car that was bought by an AI bot, negotiated by another AI bot, it’s going to run right off the road.

55:48

But maybe we get to choose a color and because, I mean, that’s important.

What I did think was really positive is a simulation environment.

So this sounds cool.

So you have a simulation environment that agents will need to show they’ve been tested in.

So it’s almost like I like.

I think this is good.

56:04

Ideas like AI is only occasionally great.

So in order for AI to be integrated into business for certain applications, the AI system has to go through experiential simulations to prove that it’s able to do the task it’s designed to do, rather than just be this general purpose tool that’s not vetted.

56:24

Like if you’re putting out software, you have production environments and you have testing environments.

So you would need to have new software tested before you put into production, before you put it out into the real world.

So this feels good because you’d have like sandbox environments and you’d have to show that it’s been tested.

56:39

I think that’s good.

What I wonder is whether we could create an entire separate sandbox of the Internet for all these agents to play with each other.

Because my concern would be like my agent may be tested in its sandbox as it’s always been tested in its sandbox with mine and do they play nicely together?

56:57

To like a standardized benchmark.

Yeah, Agent, agent benchmark, yeah.

OK, so I think buying a car, do they play nice Yeah.

So the last one is spatial intelligence.

So AI will be able to understand and interact with 3D environments, for example, like in shopping online environments.

57:14

It’s not just a static interface.

I’d like to retitle this one as well.

That AI will grow eyes, limbs and hands because touches mentioned here also isn’t that fun.

That’s like a real I.

Think, and it was interesting because I was looking because one of the podcasts I really enjoy is AI for humans.

57:30

And they were making like 3 predictions for 2026 and they were saying like robotics, this will be the robotics kicking off.

And I think it’s, you know, to some extent it’s really like, how do you embody AI?

Because that’s got to be the next big frontier.

And I think the part from like a psychology perspective is just how do you create more stickiness?

57:48

How do you create more desirability within marketing spaces?

Because that’s what companies are trying to do in terms of sell more, right?

Create more value.

I think that’s the, you wouldn’t say risk, but that’s the use case for these things is creating more value through embodiment and through kind of presence.

58:05

Because I think that’s part of the problem is they can’t chase after you.

Like use me, use me because you can turn it off at words.

I think it’s quite different if it’s physically present.

Definitely adds the anthropomorphic tension of I feel like I’m turning off a living being.

But the five trends were interesting because in the conclusion remarks of these five trends on the Salesforce blog post, this person says the common thread of these five trends requires humans to remain firmly at the helm.

58:34

To me, this mostly sounds like humans don’t have to be in the loop at all.

It certainly feels like it’s designed for humans not to need to be as involved.

But a lot of these things require like, some serious oversight, which is interesting because I think a lot of the jobs that we were talking about in our last episode are talking about a very different type of role.

58:55

You know, whether it’s about throwing water on it, kill it, whether it’s about kind of like oversee it, whether it’s about chief air officer.

It does seem to feel that these kinds of trends will require very different types of roles for the humans that are maybe not in the loop, but watching the loop, holding the loop.

59:14

I don’t know what they’re doing with the loop, but they don’t feel like they’re in the loop.

Well, it’s interesting because the survey of the future of work with AI agents that study that we talked about a little bit earlier, it also mentions that the skills needed are transitioning from informational skills to interpersonal skills.

59:31

And I think, yes, if we trend toward that, great, AI can’t be a manager.

But when we are using AI as our first stop at work to ask the tricky questions, blah, blah, blah, you’re detracting from the interpersonal engagement and communication skills.

59:49

So I don’t know that this is really the transition that’s occurring.

I think there is a transition simply to each of us at our desk monitoring what the AI is doing, interacting with the AI, using the AI to do work, and not necessarily branching out to the ecosystem of our workplace.

1:00:09

It goes back to one of the tech giants last week that we were talking about who is kind of saying recommendation that he would give to kids these days is don’t study something different.

Don’t try and go into different jobs, but do figure out how you’re going to do this differently.

Do make sure you are looking at AI, but and to your point, like keep developing these critical thinking skills.

1:00:29

You know, it’s not not about actively go out there and try and do something different.

Right.

So just to wrap up, Salesforce also had a November 26 blog post and they’re saying that executives actually are increasing the amount of entry level jobs that they are looking for.

But what’s key is that people who are in the humanities etcetera versus those who are in computer science are outperforming those in computer science in terms of job prospects and unemployment rates.

1:00:56

And what people are saying is that this is because critical thinking skills are the skill of the century, essentially.

They also looked at some Slack workforce surveys of 5000 global desk workers, and they found that people who are daily users of AI that had 64% greater productivity than other people who use AI had more strong critical thinking evaluation skills.

1:01:22

So the idea is if you’re going to use AI, well, you have to engage critically with output and critical thinking skills are the skills that people are seeking for these entry level jobs.

And so that’s why we’re seeing more people from art history, humanities, nutrition sciences, etcetera.

1:01:39

All of the ones that were mentioned in this blog post, they’re outperforming people in computer science because they’re saying that prompting is engineering now.

And those with the critical thinking background from their degrees, we’re going to be able to apply skills across the board and not just specifically to creating AI.

1:01:57

Instead, it’s engaging with AI critically.

That is what is going to be valued.

It’s quite interesting when we talk about the amount of work that’s going into creating artificial general intelligence and moving from narrow to general.

And that’s actually exactly what’s being prized.

1:02:14

And the human type of intelligence is more general applied intelligence.

Interesting.

Yeah.

And I mean, I just feel like we’re coming full circle plugging Sherry Turkle again, which is don’t try to solve problems created by technology with technology.

1:02:30

Retain the human in the loop.

Build and practice using those critical thinking skills and just as you would with prior tools, don’t get fooled by the fact that AI can talk to you.

Be as critical with AI output as you are with other technological outputs.

1:02:45

Have a keen discerning eye with outputs and don’t just think it’s always accurate.

I can’t.

Believe people still think that scary thought, right?

Scary.

So Rose, that’s been a really interesting conversation going through all of the different perspectives, all the different voices.

Thank you for being patient with my business hat today.

1:03:03

Well, thank you for putting on your business hat.

So thank you so much for listening to episodes.

So two of our exclusive third series on AI in the workforce and people’s relationship with work in the age of AI.

And we can’t wait for you to hear from our interviewees this time with this series.

1:03:20

Very excited to have some high profile names joining us for this conversation looking at AI and hiring and AI in the workforce from an institutional perspective.

So stay tuned for two weeks from now when you’ll see our first interview for this series.

And our slightly different profiled ChatGPT and and I friends who are colleagues to many people out there.

1:03:41

High profile in a different way, but just as high profile in the sense of how many people engage with them.

So we’re looking forward to hearing from them as well.

And this is Bye from Rose.

And bye from Angie.

We’ll see you soon.

We’ll see you next time.