AI in UX research: automation bias, synthetic users and protecting your brain

Introduction to AI in UX Research

Henrik Mattsson · 00:08
Hi everyone, and welcome to another season, the fifth season of the UX Research Rundown podcast. As per usual, I’m your host, Henrik Mattsson. I’m the CEO of Lookback, and it’s great to have you back. As followers of the podcast know, we have a theme for each season, and we’re going to continue this season with AI. But we’re going to try to capture the moment we live in right now, where, in my view, we have gone from imagining what AI could be, you know, looking for tools, trying to build new tools, trying to imagine what this will do to the practice, to actually having had time to test out some things.

We’ve learned to live with this technology, and we’ve started to see not only what it can do, which of course are some amazing things, but also some of the limitations. And I think, if I may say so myself, we’ve learned a little bit, we’ve discovered a little bit, who we are as researchers in this process as well, and I find all of that very interesting. So that’s what this season is going to be about. And my first guest is Tatiana Pilnik, who is working at Google, but she’s here as a UX researcher.

I talked a little bit with her before this, and I think she has some amazing perspectives to share with us about AI and what it is becoming, and who we are becoming with it. So without further ado, welcome so much, Tatiana. How’s it going?

Tatiana Pilnik · 01:39
Hi, thanks for having me here.

Henrik Mattsson · 01:42
Yeah, it’s great. All the way from Brazil, right?

Tatiana Pilnik · 01:45
Yeah, exactly. So far away.

Henrik Mattsson · 01:46
Awesome. I’m rooting for you in the World Cup now. My team got kicked out, so you still have a chance. Hopefully by the time this is published, you have won the gold medal. And as always in this podcast, why change something that works? We start every episode with the guests introducing themselves, but also sharing a bit of their origin story. Of course, there are many ways into research, and I love these stories, and I know our listeners do too. So please tell us about how you got into all of this, and why are you here today sharing these perspectives?

Tatiana’s Journey into UX Research

Tatiana Pilnik · 02:28
Yeah, well, it wasn’t a linear path, that’s for sure. It was a passion-driven path. But I like to call myself currently a social technologist. I’m a UX researcher, service designer. You can think about all of the UX titles that exist. We somehow cross through them. But I identify as a social technologist, especially because I’m super interested in understanding how humans and technologies interact, and especially not only how we shape technologies, but how technologies shape human identity and society. So this is my current focus of study, and that’s outside of my work practice and everyday life.

That’s especially because I did a master’s in design for responsible AI. And it’s interesting that before joining the master’s, I didn’t really know what design for responsible AI was. I didn’t even know it was something you could study. But now, with that degree in my hands, I can say that I kind of know what that is, but it’s still a field that we’re developing. But my background…

Henrik Mattsson · 03:38
Awesome. This is not a test, but I will definitely dig into that. Yeah, sorry. Please continue.

Tatiana Pilnik · 03:46
But my bachelor’s and my origin story are in design. So it’s not too far away from UX research, but it’s interesting that I didn’t really understand or get into research until, I’d say, my bachelor’s thesis or my first role as a service designer, where I was initially studying women’s health in Brazil and how menstrual poverty was affecting the population. I was like, there is a design problem there. And at that time, interestingly enough, I didn’t know the difference between qual and quant. And so I did 100 qualitative interviews.

Henrik Mattsson · 04:32
Hey, qual at scale, this is a thing now.

Tatiana Pilnik · 04:36
But I had the time to do that at uni. And yeah, I ended up developing an interesting project on women’s health in Brazil, which led me to my first role as a UX researcher or service designer. And it was the best, just going and talking to people. I think this is my favorite part. Getting to know people, learning their mental models, and how things that sound obvious to some people are not for other folks. If I can mention one specific learning, I was trying to improve self-service, yeah, that’s the word in English, self-service totems for users.

But the word in Portuguese for self-service is the same, or it sounds the same, as high service, like something that is an elevated service. And some people didn’t think that self-service was for them, because they didn’t think that elevated service was for them. That’s the difference in Portuguese between “autosserviço” and “alto serviço”. It’s the same word. So yeah, for me, that was the moment that research clicked, and I was like, humans, we are amazing and so interesting.

The Role of AI in Research

Henrik Mattsson · 06:00
I love that story. I love that story. It’s so at the core of… You know, sometimes when I talk to researchers that are getting into the field, they’re worried about AI and stuff like that. Sometimes I get the feeling that this is some zero-sum game, where there’s a certain amount of research to do, and if I don’t get to do a part of that research, then there’s no research left for me. But what I love about your story, and I’ve heard things like that, is that we will never run out of research, because humans.

You know, even if everything else just accumulated linearly, humans would come and this would happen. And it’s so fascinating, I think. Yeah, I love it.

Tatiana Pilnik · 06:44
Yeah, and we are unpredictable. And as my current study shows, we are shaping technologies, right? And of course technology is shaping us, and the way we interact with things, with products, with services, is changing. “Rapidly” doesn’t even summarize how fast things are changing. But still, we are unpredictable, right? In some ways. So there’s always something to study.

Henrik Mattsson · 07:17
Yeah, no, for sure. Awesome. You were telling me before this a lovely story about you actually having a connection to the podcast before. Can you please share that with our audience? I think that was such a great illustration of this field.

Tatiana Pilnik · 07:33
So before joining Google as a researcher, I was working as head of design research at a startup. It’s also an interesting startup, where we used to help victims of financial scams get their money back in Brazil. Scams are a huge thing here.

Henrik Mattsson · 07:51
Yeah, everywhere it’s rising.

Tatiana Pilnik · 07:54
I don’t know if it’s interesting or sad, but Brazilians actually export scam tactics. So creative.

Henrik Mattsson · 08:01
Yeah, there are a couple of export industries like that.

Tatiana Pilnik · 08:06
I know. So yeah, as I was conducting interviews with victims, it was really challenging, and they’re the use case where AI is not necessarily being used for good. And after wrapping up my role at this startup, I started interviewing for this Google role. And I was listening to this podcast, and there is one episode in which another Googler is also interviewed. That’s Zoë Glas, and I was listening to her episode preparing for the interviews. And then, I think it was my second or third, I don’t know which round of interviews, I clicked on the link, join, and there she was.

It was the same person that I had just listened to talking in the podcast. I don’t even know if she knows this story. I don’t know if I told her. But she was interviewing me, and I had just heard her talk about a bit of her work and everything, and I was interviewing for her area without even knowing. And now she’s my manager. So yeah, I love this loop being closed.

Challenges of Human-AI Collaboration

Henrik Mattsson · 09:07
I love this story. Yeah, and she was the one who also recommended you as a guest on the podcast. So this is how we do it. That’s awesome. I’ll give her my best. I’m sure she’s going to know now that this is the connection. I take full credit for your entire career from now on, then. I’m kidding. Great. Awesome. So thank you for that. Now let’s dig into the actual now. Obviously a huge, huge subject here. So many things to talk about, and especially as the first guest, you know, it’s like, what have we learned about AI now?

But I know from previous discussions we’ve had, and I’ve seen some of your work and everything, that you have so many interesting perspectives to share on what we have learned about this technology, what this technology has taught us about ourselves, and perhaps where some of the near-term and medium-term challenges and things we need to figure out will be. So I’ll just let you start with whatever perspectives you want to share, because I know you have a full book of those, and we’ll take it from there.

Tatiana Pilnik · 10:19
Where to start? I have a few… Okay, I’m gonna start by saying that there are no certainties, right? I don’t have a final answer. I don’t know what’s going to happen. I can’t predict the future, and honestly I don’t think anyone has an answer by this point. We’re just going to have to use our human skill to deal with ambiguity and deal with our own frustration of not knowing, and try to settle.

Henrik Mattsson · 10:53
It’s so interesting, you know, I wanted to ask you about that. Obviously, in my job, I’m a very product-heavy CEO, so I’m involved in all of this, especially as we involve AI in everything. Obviously it’s always good to be a bit humble about the fact that we don’t really know, but since AI came along, it’s moving so fast that I find myself constantly putting the disclaimer in there, like, I really don’t know. Do you think that this pace will… Are we going to keep saying that now forever?

And is that a good thing, an annoying thing? I’m not sure.

Designing for Effective Collaboration

Tatiana Pilnik · 11:31
That’s a good question. I don’t know. I think it’s a good thing, right? To not know and to acknowledge that we don’t know. Something that, unfortunately, we don’t see a lot of LLMs doing: saying that they don’t know. And this is part of the problem. But I think we also have to work with what we know, right? And there are a few things we’ve learned. We’ve learned that we can’t delegate everything, and there is a risk of delegating our cognitive skills to those machines. There is also a risk of actually not having increased efficiency.

And this is something I’ve studied recently. Not only are we actually increasing the efficiency, but when we think about humans and AI collaborating, is this collaboration being effective? Is it being optimal? And from what I learned, and there are several peer-reviewed papers published, they’re great papers by the way, I can send you the links later, they say that, no, actually, when we put humans and AI to collaborate and work together, they end up getting in each other’s way. That happens because humans are humans. And what I mean by that is we have biases, right?

We tend to agree. We tend to think that machines are always right. That’s called automation bias. If I asked you who does calculation better, a calculator or, I don’t know, me, you’ll probably say the calculator. I’m not a mathematician, you know, I’m humanities, design. But still, we tend to believe, or we tend to acknowledge, that machines are superior in that sense. So some folks will have this tendency to agree more with LLMs, with machines, whatever. And there are also the other folks, who will be overly skeptical.

Because they know that LLMs are not necessarily precise, they’re generating their answers most times, they cannot fully trust the system. So you have users who are skeptical and users who are overlooking, or just overly trusting, the system. And this creates a pitfall where the system is not actually doing its best work when humans are overseeing it, sometimes.

Henrik Mattsson · 14:03
That’s interesting. Are you saying that we are making it worse than it could be if we let it alone? Am I interpreting that right?

Redefining Efficiency in Research

Tatiana Pilnik · 14:18
I’m not saying that based on my own studies. This is not Tatiana P. as a researcher. I have to give credit to the original researchers on this subject. But sometimes, especially when it comes to decision-making, right? For example, you have to decide if an image is cat or not cat. If you have a system doing the preliminary analysis, saying, I think this is a cat, or I think this is not a cat, and a human having to make a judgment on top of that, that’s when we get in each other’s way, right?

I’m not saying that when you are generating an image for your slides, or you’re doing something that is more generative and not necessarily related to decision-making, we are underperforming, but especially in decision-making moments. And the key thing for me as a UXer, not necessarily as a researcher now, is that design plays a key role in this experience. That is, by shifting the interface and making design tweaks, we can actually improve the overall collaboration performance metrics. Which for me was super interesting. Like, okay, so UXers now have a key role in this space.

That is, we have to either increase trust or increase skepticism. And depending on the situation, we also want to change how cognitively engaged the user is in that task. So sometimes the person is tired, the person has been doing that decision for, I don’t know, three hours, and we have to change something for them to recall their attention. And that’s a new function that designers gain.

Henrik Mattsson · 16:01
Yeah, this is so interesting. We’re working on this problem, I think. I’ll tell you what we’re working on, and then you tell me if this is an example of this. AI moderation is obviously a thing now that almost everyone has, right? And I’ve been talking to a lot of researchers about how they use it and all of those things. And the most interesting finding to me, and it’s based on a lot of people, totally independently of each other, telling the same thing, is that they’ve tried to automate the moderation, and they’ve ended up actually having to watch more video, because the process was so efficient that it disconnected them from the raw data.

And then when the results were in, they felt like they couldn’t make sense of the results, or they couldn’t trust the results, because they hadn’t been along on the process. And for me, who’s been thinking about stakeholder engagement and things like that for like 10 years, this is similar to what a stakeholder that isn’t involved early and engaged continuously is feeling when they just get, here’s the research report, you know, learn what we learned. Same kind of thing. And it’s really interesting to think about how to design around that.

So if you have any kind of secret sauce here, if we figure that out, if there are some tricks, I’d love you to share to the extent that you can, because it’s a really hard problem. You want to make it efficient, but you don’t want to make it so efficient that people are disconnected and actually end up having to re-watch things.

The Future of Research Outputs

Tatiana Pilnik · 17:44
We have to update definitions of efficiency, right? Is efficiency precision? Is it depth? I don’t know. I agree. I don’t know if you’ve tried this. It’s an interesting exercise to use a tool such as NotebookLM, which I used before joining Google. This is my preferred tool, even before. This is not an ad or anything. If you add a bunch of sources and ask it to just output a report, I don’t know about you, I don’t know if you’ve tried this, it’s okay, but you don’t really understand it, and you know it doesn’t have the juice of the research.

And I agree. I never thought about it the way you said it, that maybe the way we are receiving it is the same way our stakeholders are receiving it. Which brings me to my current reflection, that is: what are new outputs we can generate? Because honestly, if we don’t think that a report now is, I don’t know, steering people’s product decisions, for example, we have to think of different outputs, research artifacts. And my perception is that human stakeholders are not the only ones who will actually use research.

Rather, their agents and their AI tools will be the ones using the report. Because maybe they’ll be in the presentation, in the readout, they’ll join in some research sessions, debriefings, but the output itself cannot just live in a 30-minute, one-hour readout and then a PDF that’s saved and no one ever looks at again. And how can we actually create artifacts that can be embedded into these agents or AI tools, so the tools, the products, will remember what you said in the research when they are making the decision?

For example, you have a PM who is now… I’m gonna use a context from a previous role of mine. We were designing an interface to allow victims to report their scams. And we had a challenge that we faced in Brazil: several users are illiterate. How can we design a mobile interface when our users are illiterate? So I produced a ton of research suggesting not only design solutions, but also how these people interact on the web currently, because they use WhatsApp, they use several other platforms.

And there is tons of research beyond my own already proposing solutions, or proposing other ways they actually navigate this environment. So if we could upload this to the agent you’re using to design that interface, that would already be incorporated, without having to be top of mind for the designer or the PM the entire time.

In Lookback An MCP server lets the AI tools your team already uses work with your projects, recordings and participants. Learn more

Skepticism Towards Synthetic Users

Henrik Mattsson · 21:03
Yeah, yeah, absolutely. And it’s so interesting how, you know, I don’t even know if we can call it research. I’ve been reflecting about this recently, because obviously I’m trying to research, and we’re trying to research people using AI. But it’s so new and it’s so fast that it’s hard sometimes: are they actually using it, or are they experimenting? There’s a difference between someone actually using it, like, yes, this is what we’re doing now, this is what we’re trying to do, and someone just playing around with 50 different tools and running experiments.

But in what I think is research about AI, I’ve noticed, and perhaps this is now an obsolete thing, there used to be this discussion: is the dashboard, the SaaS dashboard, dead? Do we need dashboards anymore, or will it just be agents talking to agents on data layers and stuff like that? And what I see is that most people just use… Sometimes they will use the dashboard to do things manually, even export and watch clips like they would. Sometimes they will use the in-app agent to do a lot of that stuff for them, because it’s faster, et cetera.

Sometimes the agent will be pre-programmed to do some things that perhaps they didn’t even know, like, you’re getting biased, you have confirmation bias, here is a contradictory story, and things like that. So it’s almost like an information dashboard, a heads-up display with decision support. Sometimes they want to bring all of that out through an MCP to their LLM, because that’s where all of the other context that you’re talking about is living. Have you looked into how people use these different approaches? Here’s my question now, after long-winded context sharing: do you think that we’re in this moment now because we haven’t really figured it out yet?

Or do you think that this is a little bit what it will look like when we do work in the future?

Tatiana Pilnik · 23:15
The thing that keeps coming into my mind while you were asking this question is that there is also the thing about digital twins and personal work avatars, which are not yet a reality. But I think we’re gonna keep experimenting for a while, until we find what best suits us. Probably what I think we can say is already established is that, definitely, the manual work, for example transcribing an interview, we’re not gonna do anymore, unless we really, really, really want to. And as for most cognitive skills that we are delegating, we have to choose what we still wanna do.

I don’t know, when I’m doing a lit review, I tried summarizing a lot of studies, and what I noticed is that we really lose a lot of nuance. And I think that’s something that we still couldn’t replace with AI tools in research. That is, we as researchers can read between the lines, understand what’s not being said, because in a recording we don’t have body language, hesitation, and those things we still can’t replace. And honestly, I’m very skeptical about synthetic users. I don’t know how much you experimented with them, especially because of this.

In Lookback Every session is transcribed automatically, in over 99 languages. Learn more

Henrik Mattsson · 24:49
Very, very briefly in the beginning, but yeah, I want to hear your perspective first before I go on a rant about that.

Tatiana Pilnik · 24:56
I’m extremely skeptical. I haven’t come across a single use case where they would be the optimal resource or tool to answer a research challenge. Especially because of what I’ve mentioned. I think the gold in research is in these very specific, unexpected moments of an interview where you see, they hesitated there. Why did that happen? Let me see if someone else does the same. And then you start linking the dots, and then you can find an insight which is different. There are of course general insights that are not new to anyone.

People are feeling overwhelmed. People are feeling they have phone fatigue. Some insights, yeah, of course this can be replicated, or in the scenario of the scam victims, they feel embarrassed. Those sorts of insights, yeah, they’re very… but they don’t translate into a unique business or product opportunity, which I think is the real goal of research.

In Lookback Recruit real participants for your study without leaving your dashboard. Learn more

Understanding Compute vs Non-Compute Problems

Henrik Mattsson · 26:10
No, I agree. So just a disclaimer here: I’m always terrified of talking about these things, because I’m at the age now where, you know, I’m young Gen X. So I grew up with all of these technologies coming one after the other, and there were always a bunch of people saying things like what I’m about to say now, and then they were embarrassingly wrong, because they just didn’t get it right. So I’m terrified now, but I’ll tell you what I think about this. There are compute problems, and then there are non-compute problems, in my world.

And I think research is about the non-compute problems. It’s a very inefficient way to compute things, through some simulated research process. That seems to be a roundabout way, if you understand what I mean by that. If the synthetic user can answer it, or if we can learn something from the synthetic user, that means it’s a compute problem. So just ask the AI. There must be a more efficient way to do it than to run some weird synthetic research, is my view. And where I see there’s value is the, you know…

When you look at what AI is, I don’t know if you’ve seen this chart where they take all of the industries and the skills and everything, and you see the amount of disruption by AI and replacement by AI and different things. So for example, transcription: that is a pure compute problem. You can just make a machine do it. As long as the cost of the machine doing it is cheaper than the human doing it, the machine is going to dominate and take over. Translation, same thing, all of these things.

So that’s why it’s taking over coding and things like that, for example, because these are just compute problems, if we simplify it. But when you’re looking at research, a lot of it is sense-making. You’re not producing anything, you’re not creating an artifact, you’re not creating a design, you’re not creating a line of code. You’re creating a reflection on something that is happening in ways that you don’t even understand. When you as a human look at body language and you get the feeling and you get the intuition, perhaps that is just some higher-level compute problem that we do.

Like, I think so far it’s something else. And that’s why I don’t believe in synthetic users. I just think it seems like a roundabout way to just ask a machine. Does that make sense, or did I make a fool out of myself now?

Tatiana Pilnik · 28:44
No, I think it does, and I agree. I would just add one word for both of us, right? Currently.

Henrik Mattsson · 28:53
Currently, you’re right. Yes.

The Intersection of Technology and Theology

Tatiana Pilnik · 28:57
Maybe tomorrow I’ll change my mind. Especially, well, I started studying more on the philosophical side, or the social technologist side. I currently started studying symbiosis, and this is the subject that I’m interested in. I’m personally interested in how technology and theology overlap, because right now it’s weird how much they are overlapping. And there is one researcher, or scientist, who had this interesting theory about life and the evolution of life, from matter to unicellular individuals to multicellular individuals, and how complexity is becoming an increasing part of life and human life.

And what he suggests, and actually he’s citing another paper, I’ll share all the sources later with you, is that life is a state of matter. As we have liquid and gas and everything. And he explains DNA as a part of computing, and how we actually evolve through this complexity and by replicating. So yeah, that’s why I added the “currently”, because yes, I agree, they don’t reflect the human complexity now. But as things evolve, as the computation for these products evolves, we might actually be surprised.

And if this researcher, philosopher Blaise Agüera y Arcas, is correct, then I don’t know.

What Makes Us Human?

Henrik Mattsson · 30:48
Yeah, no, exactly. A good friend of the podcast, and a researcher, he moves around a lot, so I don’t know where he is now, he’s been at all the companies, but Noam Segal. Once in a seminar, when we were talking about AI, he challenged everyone to identify what is truly human in a way that cannot be replaced, you know. And I’ve been thinking about that since, and I think I know the answer, currently. I love this word. I’m going to keep saying that now. And I think it is that something that is uniquely human, that AIs are not yet doing, they’re actually terrible at it, and it’s one of the reasons why I distrust them, is that they don’t care.

And humans really care about things, for very weird reasons sometimes. Having built product now for a long, long time, I note that, and I don’t know, perhaps I’ve just been in dysfunctional… you know, I don’t think so. I’ve worked with some great people. Great product people care a lot. The best ones are the ones that, when I say something, they almost get upset. I mean, of course they stay professional, but you’re supposed to fight a little bit about it, because you care about it, right?

And customers care about things. Humans care about things all the time, and it’s completely irrational. The other day, I wanted to buy a product, and I know I need it, and I will end up buying it, and I love it and it’s great and everything. But there was something with the payment flow, and I’m just, I will never buy this product, because of the payment. It makes no sense whatsoever, right? But whenever you talk to an AI, you’re just, you’re completely wrong, and it’s like, yeah, you’re right, that was completely right.

And it just goes on, and it never cares. And I think a lot of research is about knowing what people care about and what they don’t care about. And currently, I think it’s very hard for a computer to understand those things. That’s my claim.

Defining Humanity in the Age of AI

Tatiana Pilnik · 32:51
Yeah. In my lectures, as part of the things I do outside of my regular work, I have this empirical lab for critical AI literacy, where I share learnings with school kids and teachers about how to critically understand AI. And the first exercise I drive with them is exactly asking: what is a human? I show a bunch of pictures and ask, is this a human or not human? To figure out where our humanity is. Is our humanity in our physical bodies? So how much can we change our bodies, or, I don’t know, remove or add to our bodies, and still be human?

Is our humanity in our biological origin? So if I’m born a human, I’ll be a human forever? Or can I lose my humanity even if I’m born human? Or can something that wasn’t born a human, but suddenly starts behaving and acting like a human, become a human? So if my dog starts walking on only two paws and goes to work and has an income and pays taxes, will it become human because it behaves like a human? And usually what I see as feedback from people in the room is that we don’t have a single unified definition of what a human is.

I agree, I think caring is a part of that. But the problem, and every time I show the pictures we always get to this point, is that I’m showing a picture and people tell me, no, but this is a representation of a human, right? This is a picture. This is not a human itself. To which I respond: great, but what is the difference between a representation and the thing itself, if the representation is so realistic that you can’t actually distinguish? And that’s what’s happening with AI right now, right?

I’m not talking about professionals now. I’m mostly thinking about everyday consumers who are using it to help them do groceries, or, which for me is quite concerning, people using LLMs as therapists or as friends. And for me, what this shows is that the imitation is, not perfect, but it is fooling our brains. Even people who work with AI, and we actually know that this is not a human, this is a system, our organic brains don’t know the difference. This is called the Eliza effect, which is so interesting.

We are really fooled into thinking that it is a human. And what you said about caring, I agree, but again, what if it looks like it’s caring, right? What if it simulates caring so realistically that we can’t actually distinguish? This is one of the things I’m most concerned about.

The Nature of Caring and AI’s Limitations

Henrik Mattsson · 36:00
Yes, absolutely. I think back to my first year in university, I studied philosophy, and of course, these are very deep philosophical questions that go all the way back to Plato and probably before. We could go on and on for hours about this. I think when I say caring, I don’t mean that it cannot mimic it. I just mean that it’s not doing it. And I think your example is so interesting. By the way, please don’t forget, there’s a great website about this work, right? The one you do with the school kids and everything.

I checked that out. That’s amazing. And we should really put that in the bio of this and everything. It’s great work. With your permission, of course.

Tatiana Pilnik · 36:48
Yeah, happy to.

Henrik Mattsson · 36:50
What I think is so clear is that humans as collectives can kind of… We choose what we care about, and then that thing is real for a lot of, you know, intents and purposes, perhaps all. You know, if this thing is human to me, I’m going to act towards it like… I don’t know, I have a dog. Will I run into a burning building and save it? Probably, because it is somewhat human-like to me. It’s a person. It’s more than a dog. It’s not any dog.

It has become this thing. So I don’t know if this is a good example, but I think we can start thinking about things as human. And I’m not surprised to hear these stories about people falling in love with their AI partner and things like that. I’m sure it happens. But those phenomena will manifest differently in different users. It’s not a computable problem, I think, at the current technology level. You know, when you were talking about “currently”, and now this word is in my mind: we are very efficient computers, you know, the human brain, and it might just be that with the current technology, AI cannot do it energy-efficiently enough to reach those levels of compute.

Perhaps that’s all it is. I don’t know. But I think that’s what I mean by caring, in terms of how can a computer understand the weird ways in which humans care? From what I see, AI systems are very bad at humor, for example. They don’t do very engaging art. They can mimic art, but they can’t really create art that makes someone cry or stare at the painting for five hours. Have you come across a lot of studies about, you know, the engagement with things that it produces?

Collective Intelligence and Individual Skills

Tatiana Pilnik · 38:43
My thought on this is, it’s weird what I’m about to say, and I might change my mind soon, but as individuals, right, as individuals we’re also not intelligent. I can’t make art that will make anyone cry. But I’m still human, right? Because I’m part of humanity. And that’s where our intelligence lies. Probably the average person doesn’t really know how the flush in the toilet, how that engineering works, but we can’t just say that they’re not intelligent. I mean, they are taking advantage of the system, right?

Humanity as a collective. We’re intelligent as a collective. And this is debatable, but I think that these systems will embrace it and be like, okay, so now they’re just part of it. And this is what I meant by symbiosis. They’re supporting us to express, or to use, our collective intelligence better. Which doesn’t mean that it will suit, or it will benefit, everyone, as we’ve seen in human history. It doesn’t seem to be true, right? It doesn’t seem that everyone will benefit from these tools equally.

And what I’ve seen now, thinking specifically about kids, is that we currently think they have to learn the same things we did, the same cognitive skills that we’ve learned for the past, I don’t know, 100 years. And maybe that’s not true. I’m not saying it is not true, absolutely. I’m just saying: what would be new skills, or cognitive skills, that are indispensable, that, yes, we have to teach them, because, you know, we’re not human without those specific skills? But what are the skills that maybe we don’t need anymore?

For example, one thing that has already happened in humanity, because this is not a new effect, is memory, right? Before having writing as a technology, we had to remember everything. The Bible was meant to be memorized by heart. That’s why it’s so recursive and repetitive. But now we don’t have to, because we can actually read it. So our memory for that, that skill, is no longer necessary. Is that a good thing or a bad thing? I don’t think there is a judgment here.

Preparing Future Generations for Change

Henrik Mattsson · 41:29
No, it is what it is. I mean, this is so interesting. My father-in-law, who was born in the 40s, in his training in school, they learned everything, you know, rhyming and stuff like that. So he can still read entire texts from, like, the Odyssey and stuff like that, because he remembers the rhyming. Was that technology too? I guess it actually was, the whole poetry as a technology in that sense, you know. But I think this is one of the core questions, of course. As the adult generation, you know, and as parents perhaps, it is our responsibility to prepare the young generation for what’s coming.

But we are also in so many ways not able to do it, because the world has changed, you know, and we don’t know. And it’s important to be very, very open about that. My personal strategy is to just learn almost first-principles things that have been true forever, basic things like that, but to not be afraid to let go. And I think we’re definitely in a moment like that right now with these technologies. But it’s very scary, for sure. I wanted to bring… Go ahead.

Tatiana Pilnik · 42:40
But I think the same applies to us, right? As researchers, we learn skills. We learn how to conduct interviews, how to ask questions, how to analyze answers. And this is the difference between us adults, us researchers, professionals, and the kids: we have more agency in what we choose to delegate to these systems. And there are things like transcribing that maybe don’t affect our individual processes. And maybe some folks will notice that if they actually do the transcription on their own, the analysis process will be much faster, or they’ll get to better insights, for example.

I’ve noticed that some folks would rather take notes by hand, because it still works better for them. They take notes on whatever sheet of paper and then they throw it away, because just from the mere fact that they wrote it by hand, they can remember every single note that they wrote. So it’s also an opportunity for us to understand ourselves better. For me, what I’ve learned is that I like reading. I need to read. And writing, yes, but sometimes I’d rather write in a very drafty way.

And this is a great tip, by the way: always write your own drafts. Don’t let AI do the first version. The draft is exercising your brain. So write the first draft. But yeah, I like reading. I need to read to understand things. Maybe once, twice, three times I need to read a paragraph, and then I can fully understand things. So these are the things I’m not delegating.

Henrik Mattsson · 44:31
Yeah, no, that is the… You know, I think I recorded some videos about this where I walked around and philosophized about this, but it’s like, just because these technologies exist… And of course, in the workplace, sometimes we’re forced to use them. We joke at Lookback. I say to my developers, writing code by hand, that’s something you can do in your free time. Here you got to vibe code. It’s a bit of a joke, but it’s also kind of like, we got to be efficient here.

And they love it, and they’ve embraced it. But there are some moments where we’re forced to use methods that perhaps are a little bit inefficient, but we are still free to read and write. We can do whatever we want. We all have 24 hours in the day. And if we want to read, we can read. If we want to write, we can write. If we want to transcribe, we can transcribe. If we want to look at the data in a CSV file and, you know, feel every single little cell in the spreadsheet, we can.

And same thing with researchers. I always tell new leaders at Lookback, you have to protect your strategic focus. No one else is going to protect your strategic focus. You are responsible for the results that you bring. And if you don’t focus on how you can bring that, then you’re going to stand there and be accountable to that. You can’t say, well, they told me to do this. Well, why did you do it if it wasn’t good? And I think the same thing goes for researchers now with how they embrace these technologies.

You have to ultimately… choose, is what I would say.

Tatiana Pilnik · 46:15
Yeah. Choose, and again, protect your brain. You said strategic focus. I would say protect your brain.

Henrik Mattsson · 46:20
Protect your brain. Yeah, you take that even further.

Tatiana Pilnik · 46:26
Exercise it, right? Don’t forget to sometimes put it to the test. If you need to deliver research in three weeks, which five years ago would sound insane, we have to, right? Maybe that’s not the perfect scenario to exercise your analogical brain, organic brain. But once in a while, it’s good to check in and see, okay, well, wait, are my skills still there?

Henrik Mattsson · 46:54
Right, right. This connects to… you know, we’ve reached this point now where we need to start thinking about the time of our dear listeners. I think we’re becoming something new all the time, obviously, because we live in a world that changes, and technology that changes, and everything. As researchers, or I can ask this as a personal question: what have you learned most about who you are, and who you’re becoming as a researcher, working with these technologies in the last six months to a year or something like that?

The Evolving Role of Researchers

Tatiana Pilnik · 47:30
That’s a good question. I think I’ll circle back to the beginning of the conversation, because I mentioned my origin story, but not in a very chronological order, to answer your question. I started my career in consultancy. And as a consultant, we have to do everything. And as a junior consultant, it was really weird to have to be an expert in anything just because I was the researcher. But at the same time, I learned there that the most valuable skill I could bring to the team was actually saying no instead of just executing.

Of course it’s important to know how to execute and to be a good executor of research. But also saying, no, this is not the research we should do. No, this is not the user problem, or this is not the client’s actual problem that they’re trying to solve. And I kind of knew that, but it hadn’t really surfaced until, I think, the last year or six months ago, when doing and executing the research became the easy part of our work. The hard part is actually knowing which research to do and which questions to ask, because every day people have less time to answer questions.

So instead of having one-hour or 90-minute research sessions, as we used to in the beginning of my career, now I only have 30 minutes or 20 minutes. How do you conduct an in-depth interview in this time? So knowing the right questions to ask, and knowing what briefings to say no to, I think this is something I’ve learned, and an indispensable skill right now.

Henrik Mattsson · 49:25
Love it. And did you learn this, if I understand you correctly, basically because as you are able to do more research, you realize that this is becoming the limited resource, the bottleneck? Or is it something else that drove you to that insight?

Tatiana Pilnik · 49:46
It’s more related to career than actual efficiency. For me as a professional, the value I can deliver is not directly tied to the amount of reports I deliver, but to the quality, and the match between the need and the delivery that I can actually find. So it’s not necessarily that volume is the bottleneck. No, I mean the opposite. Volume is not the solution to high demand.

Henrik Mattsson · 50:18
Right, yeah. But did you realize that because it was all of a sudden easier to do more research, or was it just where you are in your…

Tatiana Pilnik · 50:28
I think both. It was easier to do. Earlier, before it was so easy to conduct research, we would just say, no, I can’t do this because I don’t have the time, my backlog is already full.

Henrik Mattsson · 50:39
Yeah, exactly, resource constraints.

Tatiana Pilnik · 50:42
So now it’s, this is not a priority, so we’re not doing this.

Henrik Mattsson · 50:47
Right. I think that’s what developers and designers and everyone are feeling now too. Yes, I can vibe code this very quickly, but should we really be doing this thing? That’s so interesting.

Tatiana Pilnik · 50:56
Yeah, this is not gonna solve the problem, so let’s not do this.

Henrik Mattsson · 51:00
Yeah. Is it Winnie the Pooh who said that there’s no point running when you don’t know where you’re going, or something like that?

Tatiana Pilnik · 51:07
I think it’s Alice in Wonderland. It doesn’t matter which road you take if you don’t know where you wanna go.

Henrik Mattsson · 51:14
Yeah, there’s probably a bunch of versions. It’s very, very smart. Talking about the first principles that we should still teach our kids. Awesome. Well, look, this was amazing. Thank you so much for sharing all of those perspectives. And, you know, I’d love to have you back, perhaps in the next season or two, to see how wrong we were. We did say “currently”. We’re trying.

Tatiana Pilnik · 51:37
Currently. Maybe tomorrow everything will change.

Henrik Mattsson · 51:42
But I think there’s a great message here. First of all, I love your message of protecting your brain. I think there’s a protect your time in there, protect your focus, all of these things, but protect your brain first and foremost. Just because these technologies exist doesn’t mean that you have to use them all the time. What I love about discussions like this is that you are in an environment where you’re probably one of the researchers that are on the bleeding edge of this stuff. You’re at Google and you’re doing these things and everything.

And look how you’re still trying to figure it out. This is the message I want to send to everyone: this is the normal state. If you feel like you’re disoriented in this, or it’s moving fast or something, it’s not because there’s a bunch of other people sitting there knowing everything and knowing exactly where it’s going. It’s like this for everyone. And this season, I really hope that we can bring a lot of voices in here to try to figure it out together. So a huge thank you to you, Tatiana, for taking the time with us today.

We talked about your website and everything. Is there anything you want to share in terms of, you know, can we follow you on social media? Where do we keep up with this work? Any resources like that that you want to share?

Tatiana Pilnik · 52:59
Yeah, absolutely. Feel free, anyone who wants to, to add me on LinkedIn. Just ping me. I have a lot of bots adding me, so just, you know, ping me and say, I heard the podcast episode, I’d be interested to connect. I’m happy to open my network to let people know more about my work. Also, I know we discussed a bunch of things, different subjects. They were a bit all over the place. So yes, if I mentioned any resource you’d be interested in, I’m happy to share the papers and the philosophers or researchers behind those specific thoughts.

Henrik Mattsson · 53:38
Awesome, thank you so much. And thank you to everyone who’s been listening, and welcome to this new season. I hope you liked the first episode, and there’s more to come. Until next time, happy researching, take care, and bye.

Tatiana Pilnik · 53:51
Bye, thank you.