AI and UX research: world designing, design fiction and working with AI engineers

Henrik Mattsson · 00:09
Hi everyone, and welcome to another episode of the UX Research Rundown podcast. My name is Henrik. I am the CEO of Lookback and your host for this show, which is now in our third season, and we have changed the subject. This used to be about how you get your team along on the journey that is research. That’s still super relevant, but this season we’re talking about AI in all its forms in research. And as always, we’re going to have some great guests, and today is no exception.

With me today, I have Savina Hawkins, who is a researcher at Meta, working a lot with AI in all its different forms. So welcome to the podcast. Thanks for being here.

Savina Hawkins · 00:51
Yeah. Thank you. Excited to be here.

Henrik Mattsson · 00:53
Awesome. Cool. So, per ancient tradition, now in our third season, we always start with the origin story of our guest. So if you could please just tell us a little bit about how you got into research and how you ended up in AI, and, you know, what has led to us being here today?

Savina’s path from psychology research to AI

Savina Hawkins · 01:13
Absolutely. Honestly, it’s a little bit of a hard story to tell, because it’s really not linear. I’ve just been an interdisciplinary person and a lover of learning my whole life. Even when I was a kid, I was doing things like learning how to code programs in BASIC, getting really into videography, creative writing. I had all these different passions that were kind of pulling me in different directions. By the time I got to undergrad, I actually got my start in research working in an academic psychology lab.

I designed research projects using eye trackers to study something called negation, which is a language learning ability where children basically understand the nonexistence of something, which has huge ramifications for consciousness. So right up my alley, super fascinating. And that’s really how I discovered I loved research, but at the same time, I also discovered I didn’t like academia. Not because I don’t like it more broadly, but more just it didn’t quite move fast enough, and it didn’t incorporate enough of my other skill sets. So I was really lucky when I found UX research.

It really just was this sort of synthesis of everything I cared about, from building to technology. Even things like writing is such a huge part of our career. So it really brought all the pieces together for me. And then how it merged with AI: I went to Stanford for undergraduate, and Stanford is just so steeped in Silicon Valley culture. AI is something that everyone was talking about. You couldn’t even take a creative writing class without somehow ending up talking about AI. So it was just so relevant and top of mind.

I ended up studying it quite a bit and ended up at Stanford making my own major. That actually allowed me to study artificial intelligence from the humanistic perspective, which is not something that existed as an off-the-shelf major at that point. And then I went on to extend that program into a graduate program at the Harvard School of Design, where I actually deepened that even further. I would say it’s really the intersection of consciousness, what it means to be human, and the cutting edge of technology that interests me the most.

And I’ve been super lucky to be able to actually just follow my passions, and passion for learning has led me to this place.

Are AI models conscious?

Henrik Mattsson · 03:31
Oh, cool. Okay. So that’s so interesting. So, consciousness. I have to ask: are these models close to something that resembles consciousness?

Savina Hawkins · 03:42
Yeah. What’s funny is I was at a literary salon on intellect last week called Breakthroughs in Consciousness. And I think the main takeaway is that consciousness doesn’t really have a good description. If you ask thousands of people, everyone’s going to tell you a slightly different answer to what consciousness is. So if consciousness is free will, like being able to decide what you’re going to do next, we now have things called AI agents, which do exercise a form of free will, where they think for themselves and they decide how they’re going to act given certain context and information that they have at their disposal.

So all of this cutting edge of technology is really problematizing what it means to be human, what consciousness is, what’s distinctly human. That boundary between the artificial and the intelligence that nature designed is very blurry and nebulous. I’ve loved poking into what that boundary means.

What surprised her about AI: tool use and GPT-3

Henrik Mattsson · 04:44
Awesome. Yeah. I felt a little bit like I was becoming an AI the other day. I was watching a Christmas movie with my kid, and I’d never seen the Christmas movie before, but because they’re all the same, I could kind of predict it. And she’s like, “Stop it, how are you doing that?” I’m like, “I don’t know. I’m just doing next word prediction based on all this stuff I’ve consumed.” So I don’t know, all of this stuff is so interesting. Obviously you’re summarizing a deep academic background in just a few minutes here.

But as someone who’s been working with these questions for a long time and gotten pretty deep into it, what have been the biggest surprises, and what have you learned from this technology compared to how we thought the models used to be?

Savina Hawkins · 05:36
Yeah, absolutely. So I’ve worked in different AI labs, and that’s given me a really wonderful vantage point where I’ve been able to work directly with folks who are working on the breakthroughs, so those AI research scientists and AI engineers. And then also, I come from such a humanistic background, I’ve always contemplated: what does this mean? What’s the big picture? How does this impact humanity? So it’s been really fun, being able to bridge the gap between those two things. I would say I had several major “oh my gosh, AI is even better than I expected it to be, and it’s happening even sooner than expected” moments.

The first thing that absolutely blew my mind is Meta was working on this paper, and this is all open source, so nothing private here. They were working on developing AI that could use other tools by writing their own API queries. For folks who don’t know what an API query is, it’s basically how two computers talk to each other and send information back and forth. And the query itself that is sent through the API is a message: hey, can I have this information? Or, hey, can you do this thing for me?

You can think of it as almost talking to a waiter in a restaurant and ordering something, and then asking them to use something, like ketchup from the kitchen. So that’s what an API is. And what the research scientists were doing was taking the ability to write API queries, so make demands of other tools, and extend it to AI. So AI actually could write its own queries and use other tools. And as someone who’s super nerdy and deep in the academic stuff: a lot of humanity and what makes us who we are has been built off of this idea that we are the tool users.

That is who we are. It’s literally called homo faber. There’s a full-on word for this in philosophy. And what’s just fascinating about it is, if the tools we are making are themselves able to autonomously use tools, what does that mean about who we are? What is our utility to the world, and how are we going to fit into this broader landscape, in which so many of the things that we base our career around are going to be done much faster and much more accurately by AI?

That was definitely a big moment. And I think another wow moment was GPT-3. Of course, everyone knows GPT-4 and all of those big large language models. But GPT-3, especially 3.5, was this huge step change in the output quality that you really could see. In fact, OpenAI also did reinforcement learning on top of some of these other models. I believe it was called InstructGPT, which they folded into the prior models. But that made it so steerable and so accurate, and just so precise with how it was mobilizing language.

At a certain point I was like, wow, I grew up loving writing, and this is going to be a way better writer than I possibly could ever be.

Henrik Mattsson · 08:42
Oh, that’s interesting. Do you think we’re going to have autonomous authors, AI authors that just produce books, not out of their own will, perhaps, but because that’s what they do? How should we think about that?

Savina Hawkins · 08:58
Unquestionably. And I think what changes is what it means to author something. For example, think about academia. All of academia is researching and reviewing existing materials, coming up with interesting hypotheses, and then going and producing new knowledge that answers questions we don’t have the answers to. And that is something that absolutely could be done by an AI, and actually probably significantly faster. It could explore multiple hypotheses at the same time and gather data against them across the whole corpora of materials in academia, which, of course, if a human were going to try and read that much information, would take them years.

So we’re absolutely going to see a lot of autonomous agents emerging, multi-agent systems that all interact with each other and are able to produce outputs that are maybe on par with what a team of humans would do. That’s what a multi-agent system is. But when it comes to things like art, I’m a big believer that real good art comes from that specifically human place. It’s very connected to consciousness and all of this stuff. And so, yes, AI can make art. It can write works of fiction, it can do all these different things.

But if it doesn’t have that sort of spark that humans have, that makes us who we are, then there’s a lot there, I think.

Henrik Mattsson · 10:27
So not poetry, but…

Savina Hawkins · 10:30
Well, I mean, AI definitely can write poetry. It’s a better poet than me, for sure.

What AI means for UX research

Henrik Mattsson · 10:34
Yeah, super interesting. Okay, cool. So this is all very interesting. And what I’ve learned doing these interviews for the podcast is that, I mean, I kind of knew this intellectually, that there was a field of machine learning, and this didn’t just start with the ChatGPT moment. But I’ve realized how incredibly rich it was, compared to how little I think the outside world knew about everything that was going on before this moment. Now everyone knows that there’s some AI going on, right? For most of the research community, I think most people have started thinking intensely about this stuff in just the last six months or something like that.

Because you’ve had a head start: now, in these six months where people are trying to grasp what this is, what have you learned about what it means to UX research, to our field, to tooling and to all the different things that go into being a researcher? So, big question.

Savina Hawkins · 11:41
No, no, I mean, I’ve thought a lot about it. I think that probably every user researcher, at some moment in time over the next five years, will transition into helping existing products evolve how the workflows work and how the experience functions to better leverage AI, so that every moment of the experience is actually enhanced by this superior technology that’s suddenly available. I think a good metaphorical comparison is the internet. You can imagine there were plenty of businesses that existed before the internet, and they had existing systems set up for how they were operating without it.

And then, of course, the internet came. It was this massive revolution where small businesses, big businesses, everyone had to reimagine what they were doing, their operations model, the product that they were offering, given this new world. And I think we’re again in one of those transition points, and UXR is exactly what’s needed, because we have the skills to basically triage between what the technology can do, what’s possible, and what people want, what’s a meaningful improvement. That’s not just, “Oh, here’s a few ideas that ChatGPT brainstormed.”

Because if you look across the product landscape right now, a lot of the functionality that people are getting is not super imaginative yet. We’re really in those early phases of people just trying to figure out how to infuse large language models and predictive analytics into their products. But I think we’re going to get a ton of creativity, and I believe a lot of it will come from designers and UXRs, in how we can leverage these technologies in more unique ways.

Henrik Mattsson · 13:24
Yeah, because you have to put it in a product, in a workflow that actually enhances something, for it to really… I’m an economist by training, and I often think about supply, like what does it do to stuff. And it’s so interesting when you get a new resource. It’s like, oh, it’s so cool, right? But very quickly it just becomes abundant. And then you realize that what really matters is, out of all of this abundant resource mass, what is it you want? What is the goal, right?

And it’s going to be exciting to see how quickly we get there, because there’s definitely the hype, if you know what I mean, about all of these things that we can do. But understanding what of all of this humans actually find useful, I think, is a great field for research. Anyway, a little bit of a ramble.

Savina Hawkins · 14:15
I love that. Yeah. What it makes me think of also is I was chatting with some really excellent research practitioners who work in semiotics for the most part, and brand marketing. And we had this conversation about whether UX researchers should use AI in their practice. What are the downsides of doing that? This was way back, quite a few months ago, so I was still really early in my thinking about this. But I mentioned to them, well, the way I think about it is: think about washing machines.

Do you really want to wash your clothes by hand? It’s like, no, right? And what he said back to me was, well, hand washing is a meditative state, and there’s actually a beauty that emerges from doing something by hand. So UX research has both a science piece and an art piece. I think we should definitely find a way to make the things that are in the science camp faster, so you can get to something that’s substantiated, rigorous and accurate faster than you would be able to with human capabilities, how fast we can read, how fast we can think.

And pair that with the idea that there is a lot of art in how you position things to stakeholders. What insights do you pull out, and why are they important? How do you phrase things so that they really impact the organization? I hope in the future we find a way to negotiate that, so that the art is brought right to the forefront.

Hands-on analysis, critical thinking and AI tools

Henrik Mattsson · 15:48
Yeah, that makes me think of something, and I’d love to have your perspective on this, because it might just be me getting older, right? When I started out in research, academic research way back when, I did my analysis in spreadsheets and stuff like that, before we had any of those fancy ways to just calculate this stuff. It wasn’t punch cards, but it was early, right? And there’s something in the hands-on work, in the details of working in the spreadsheet, that taught me a lot about analytics and all that stuff.

So it’s not just meditative. It’s also that you’re in the details of the work. And you can see the same thing with developers: some of the best developers I know are the ones that were so early that they had to work all the way from machine language up to the interface, and they truly understand the whole stack, right? And I wonder if there’s something like that in UX research too, where there’s a risk that we become too distant from the methods, that we lose connection with the art somehow.

I don’t know if that’s just me, you know, “it was better in my day” or something. What do you think about that?

Savina Hawkins · 17:05
No, I think so. First, where my brain is going is that every time a new technology comes out, that’s usually the reaction that the community has. When the TV came out, people were worried that we wouldn’t be able to develop our imaginations if the TV was doing the work that normally your brain is doing when you read a book. We hit that type of thinking every time there’s a new technology, because it forces us to rethink and adapt how we’re doing anything with this new state-of-the-art equipment.

Then the other thing that it really makes me think about, and I’m trying to think of how to say this: in my master’s program, I studied a lot of Heidegger, who I have mixed feelings about, but the main takeaway there is that tools are either things you can think with and think through, or things that do the thinking for you and report back to you, and you are then the master of the information. And I think those two dynamics are really interesting, because when you’re doing something by hand, you are doing every piece of the thinking.

You’re figuring out the concepts, you’re thinking through things logically, you’re manipulating the data by hand, so you really can see what patterns match to other things. And I think there’s a huge value in training critical thinking skills from doing that. Our schools right now are completely getting rid of essay prompts, because people can just give it to ChatGPT and say, “Hey, write my essay about Beowulf” or something. So I’m hoping that we still get the critical thinking skills developed, because I think a lot of these skills and abilities are really crucial, for sure.

From world building to world designing

Henrik Mattsson · 18:55
Yeah, yeah. And that kind of leads me… I didn’t plan for this, but it’s actually a pretty good segue. So you’ve given a few talks where you’ve been talking about what, for me, is kind of world building. I know that you have a better term for it now; that thinking has evolved a little bit. But this goes back to what our responsibility and role is in the choices we make and how we interact with these tools. And I know that you’ve thought about this in structured ways, and I’d love to hear your latest version of this thinking.

Savina Hawkins · 19:24
Yeah. So world building is something that has weirdly been following me all throughout my academic journey. It’s been top of mind, and it’s also probably why I work in UX research, out of any field: UX research and tech. Tech literally shapes humanity’s interactions with each other and with tooling more broadly, and to some degree it actually structures our social, interpersonal reality. And because it’s that prevalent and that entrenched in the human experience, whenever a UX researcher studies that interaction and then articulates to a company what you should do, they’re basically making an argument about the world that should exist because of this information.

And that is exactly why, going back to my academic days, I loved research, but in academia you don’t get to say: this is what should happen, this is how the world should be. Academia is purely about discovering or articulating a truth that already, theoretically, exists. With UX research, you get to have a stance on what should happen. And that’s a really powerful thing, to be able to influence the world and human experience at that kind of level. It’s a beautiful thing. So, what you were saying earlier about the terms: the terminology is so up for grabs at this point.

But I think world building loses some of the thought and critical thinking that goes into the research work and into design work. So I’m leaning more towards world designing, because it really is about thinking about our collective milieu, the places in which we exist, the technologies that structure our lives. Even simple things like the clock on the iPhone that goes off every morning waking you up, all of this stuff becomes part of people’s lived experience of reality. And thinking through how that should be is an immense responsibility.

Working with engineers and AI researchers

Henrik Mattsson · 21:34
Yeah, I love that development, because I’ve been thinking a lot about how UX research fits into the workflow with people that actually build things, like code or hardware, which is not what UX research does, right? It’s more design, I think. What was it you said? It’s about thinking about how it should be: studying it, forming hypotheses about how it should be, and promoting that to the builders, in a sense. Right? So perhaps it becomes even more important to think in detail about those distinctions.

Savina Hawkins · 22:16
Yeah, I love that. I guess the one pushback I have is that you’d be surprised at how UX researchers act when they get outside of their comfort zone. Our comfort zone is usually working with design and product. Those are the people we’re used to talking with. We know the language they speak and what’s important to them. And then all of the people actually doing the building, engineers, data scientists, are just a little outside of UX research’s comfort zone. So we don’t really tend to engage with them as much as we should.

I’m an interdisciplinary mixed methods researcher, so I can do things like analyze data in R and write a SQL query. I call it data science 101 skills. I’m not a data scientist, but I at least know enough to get by, to piece things together and continue the research process. But basically, if you can break down the barriers between the people doing the thinking about what should happen and the people doing the building, that’s really where the magic happens. For example, I’ve worked on teams where AI researchers were building this really expensive model, investing a lot of time and attention into training it to do this one specific thing.

And then a UX researcher interacts with it and just sees: oh, that’s interesting, why are the inputs and outputs structured this way? Wouldn’t it actually be better if the output was more like this? And those little tweaks, where we can actually participate in the literal design of the AI algorithms, are super meaningful. That’s the most central place you can make an impact as a researcher, because everything else flows from whatever is built in the back end and the capabilities that the AI models have. So I definitely think UX researchers will work more in that going forward, for sure.

Henrik Mattsson · 24:16
Yeah. And look, I totally agree. This is why I’ve been thinking about this. You should have been on last season, too, when we were talking about how to get your team along on the journey that is research, right? And by implication, how to get researchers early into building too. I’ve been interviewing a lot of engineers lately for the stuff that we’re building, because we kind of figured out the thing: how do we make non-researchers feel welcome in a research session and interact? And one of them, my CTO actually, who I was talking to about this, felt like: yeah, I love being there and I love seeing all the stuff, but I always feel like an afterthought, because I’m always invited to something that’s already been decided.

I’m like, oh, you should get in earlier. So how do we make that happen, right? I totally agree, and I think these are super interesting problems. And I think AI is going to make it much easier to create some of these connections at the right places, without forcing people out of their workflows. These LLMs are really good at picking up, like, hey, there was a conversation over here, this person should see this thing. So yeah, I think there’s a lot of super interesting tooling coming there.

Design fiction: start from the ideal and work backwards

Savina Hawkins · 25:29
I completely agree. And one tip or technique that’s worked really well for me in the past: when talking with technical people, they own the domain of what’s possible. And you, as a user researcher or someone who’s thinking in the design world, really own what should be. But what can be lives at the intersection of those two things. So you have to do some negotiation. And sometimes I do this by using a technique called design fiction. It’s not as common a user research practice as it should be.

But it allows you to skip quite a few years into the future and figure out what the long-term solution looks like. What is the absolute ideal? If we solved exactly what the customer wants, what does that experience look like? And articulate that. And then often you can actually work backwards from there to what’s possible today. What would be the incremental steps towards that? I often use the metaphor of evolution, where you start with single-celled life forms and then it gets more and more complex over time, right?

Designing these moments of evolutionary step change to get you towards the big vision is another thing that UX researchers should be participating in, because it really is influential for product roadmaps.

Henrik Mattsson · 26:52
Yeah, I totally agree. I love this stuff. And again, it’s an area where LLMs in particular can really help, because a lot of this is about transparency, right? These discussions do take place. I don’t know if it’s specifically that, but this kind of thinking takes place, and the right people might not always be in the room. And some of these things are not very planned either. We’re just talking. I had no idea we were going to start talking about this. Perhaps you knew, right? But I had no idea.

Let’s say that we were working on a product now instead of recording a podcast. This would be a moment where we would probably say, oh, this person should have been here for this, right? If we had known, we would have invited them, and this would have been even better. But we didn’t know. And this is where, let’s say an LLM is listening to this and picking up on some of these things, you can do a lot of stuff to bring people in across the whole process, because the process, as you said, isn’t linear at all.

It’s just this craziness. I don’t know what to even call it. So yeah, I’m super excited. Am I too excited about LLMs?

Savina Hawkins · 28:05
It’s OK. I am always too excited about LLMs. So it’s exciting for me. It’s like me, to be excited, which is good.

AI’s limitations and hallucinations

Henrik Mattsson · 28:15
But to reformulate that: you talked a little bit about some moments you had when you were like, “Oh my God, this is even better than we ever thought.” What are some of the limitations that you see, and how hard will those be to overcome, in your mind? I’ll give you one example: hallucinating models and stuff like that. That’s something that people are like, “Oh, they’re hallucinating, we can’t use it for this.” And I’m like, “Oh, they’re going to fix that.” I don’t know, am I right?

Savina Hawkins · 28:46
They are on track to fix that. I would say the AI field is moving so fast that if you aren’t following it week over week, you’re going to be way behind. It’s the equivalent of being years behind in the discourse, because of how fast it’s moving. But definitely hallucinations are something the field is working on fixing. And really any limitation that AI has right now, except for its ability to interface with what’s real in the world, which we’re going to have to build sensors for.

There are always limitations. Humans are embodied in a world; they can sense their environment. That’s how nature built us to function. Versus AI, which is very abstract and disembodied. There’s no place where it exists. There’s no central point where all the information is coming together, like it does with the human subjective experience. We’re actually building that. We have to build that for the AI, in fact, which again gets into the world designing piece. But yeah, it’s crazy to think about.

Henrik Mattsson · 29:50
Okay. So you’re also an optimist here, I see. So either we’re both right or we’re both crazy, you know. Who knows?

Savina Hawkins · 29:56
Yeah. I don’t know if you’ve seen that meme about the different dispositions you can have politically towards AI, the quadrants. There’s the techno optimists; I think the joke in the meme is “let’s build the AI god.” And there are people who are very fearful of AI and want to slow down. I think that’s all very fascinating. And it just reminds me of how complex this dance is, where we user researchers are such an important part of this conversation, because we provide the pacing required to move slowly and think things through thoroughly, even if the emphasis is that you must still move fast.

I think that’s what we’re faced with as a field moving forward. And I think we’ll have to adapt to this new context, because it is a new context. Things are just going to get a lot faster. Prototypes can be created on the fly by AI, or you could be in a conversation with a customer and they’re describing their ideal solution, and AI is making wireframes in real time to show the customer what that looks like. All of that is possible, and I would expect it within the next year and a half.

So the change is coming, and it’s effectively inevitable at this point. We really do need to figure out how we’ll adapt our practice and how we’ll also stay true to our underlying values.

In Lookback Show a prototype while you talk: in a moderated Lookback session, push a link that opens in the participant's browser. Learn more

Keeping up with AI while thinking critically

Henrik Mattsson · 31:20
Yeah. Right. So, to get super practical on that: that sounds like a bit of a paradox to me. On the one hand, it moves so fast that we just have to spend our time keeping up to date with what’s happening, and nothing that we think are limitations now will actually be limitations in, whatever, three months. And on the other hand, we’re supposed to play this active role in this, us as world builders, world designers. I have nothing against a good paradox like that. That sounds very exciting to me.

But how do you manage that as a researcher trying to take both of those seriously? How do you do that in your day-to-day practice?

Savina Hawkins · 32:08
You know, what it reminds me of is a paradox we’re already managing pretty well, I think: the complexity of stakeholder management, where everyone has something completely different that they care about. Their world has their information in it, and they’re pretty isolated and separated. So if you’re making a presentation about, I don’t know, lost sales calls and the patterns that you came up with, you’re going to present something to the sales team, something pretty different to the marketing team and something pretty different to the product team, because it’s about what information fits into what they already know about the world that will actually be useful for them.

So even that is a paradox, because research hopefully creates something, and I put it in air quotes, because what really is true, but it’s aiming to create some sort of truth or substantiated knowledge. And the paradox we deal with is that other people don’t want to spend the time that we spend engaging with what should be, thinking through all the nuances. They just want to consume what’s most important to them. So we’ve already figured out how to navigate that pretty well, in our environment and in how our role is set up.

So I would expect that the same skills that got us to solve that problem can also be applied to solving this problem. Which, again, maybe is a little bit unfair, that we have to take this burden on our backs. But I think it’s about being the critical thinker at the table, taking the time to do the extra thinking and think through: what are the potential downfalls of this? What are the consequences that would be bad? Are there ways that it could be better? Are there assumptions baked into our thinking about the problem that are just left over from Web 2.

0 and don’t even apply anymore? Right? So if we have the ability to think through all of those details, and then do the same thing we normally do, research our stakeholders, what they care about and what they need, you can still position what is going to be best for humans and best for the organization to the people who will need to act on it, at the right time and in the right way. I really believe that we’ll be able to do that going forward.

In Lookback Record what each stakeholder needs to know in a stakeholder interview, and Lookback turns it into a research goal. Learn more

AI stakeholders and research repositories

Henrik Mattsson · 34:20
Okay. Super cool. So now we’ve reached that point in the podcast where we could go on for hours. That’s how I feel about it anyway. But we have to start moving towards the end. So I have one final question, or it might be more final questions, but we’ll see. Do you think that we’ll have AI stakeholders shortly, as researchers?

Savina Hawkins · 34:38
Interesting. And when you say stakeholder, what type of stakeholder?

Henrik Mattsson · 34:43
How about someone that you report your research findings to? Well, how would you do that? Like prompt writing, or…?

Savina Hawkins · 34:51
Yeah, I mean, I think the number one thing is that research repositories are really hard to build correctly right now. A lot of teams have tried to cobble together solutions to make something that would work. But, doing that thing where you think well in advance about what you really want: what we really want is a centralized brain for the organization that knows who your customer is, knows what your product is and what it does, and then knows how to create the links between different research projects, both to expose the knowledge gaps that the research team can close, and also to help us resurface things from the past that are relevant, so we can be more effective in figuring out things in the future.

I think of the academic knowledge production process, which always starts with: what do we already know? How do we come to know it? What are the pitfalls with that information? I think we do a lot of that thinking too, and it’s super valuable that someone at the organization is doing it. But of course a lot of that is being done manually. It’s manual knowledge work. And I think in the future it will be at least somewhat automated, if not fully automated. And if fully automated, that begs the question of what all our roles become in the future.

Henrik Mattsson · 36:02
Yeah, that’s super interesting. We’ve done a lot of research into repositories. This is one of the cool things about my job: I get to do research on researchers, about the research, which is really cool. We did a huge project on repositories, and, no shade on any repositories out there, but the conclusion of the project was that repositories are where research goes to die. It’s a bit harsh, but that’s what my researcher told me: too long, didn’t read, repositories are where research goes to die. Right?

And again, this is a hard problem. I have nothing against any repository tools out there, but I can see how that happens, because everyone is in the moment. Everyone is coming up with the insights, and then they go somewhere. And to actually bring that back into the workflow is a really difficult barrier. But I can totally see how… I mean, we’re already starting to do some of this stuff with the other team, where it’s like, hey, it seems like you already did this, or, hey, you’ve already asked this question like 50 times.

So I like this idea of the brain that’s kind of like, “Hey Siri, have we researched this before?” Right? And another researcher I talked to was like, “I love that people are interested in research, but I hate being the receptionist for the library.” Do we know anything about this? Do we know anything about this? Ask the LLM. I’m busy researching. Anyway. Okay, cool. Well, nice to invite guests and then end with a big rant. But it was your fault, because you inspired me with all of your super cool learnings here.

So…

Savina Hawkins · 37:43
Yeah, thank you for being okay with all the pedantic “let’s talk about consciousness.” So that came up. If you’re talking to me, I can’t help it.

Henrik Mattsson · 37:52
That’s what I love so much about this podcast. I think this is what the audience loves about the podcast: we never know what we’re going to get. And I think everyone has done so much interesting work in this field and so much thinking about everything. And it’s so orienting, especially when things are moving so fast, to have someone like you come on a podcast like this and summarize, give us a map a little bit. That’s super helpful, and I know the audience is going to love it.

So thank you so much.

Savina Hawkins · 38:18
My pleasure. Yeah. Thanks for having me here.

Henrik Mattsson · 38:21
Awesome. So if people want to know more, should they follow you on social media, or are there some resources? How do people reach you after this?

Savina Hawkins · 38:28
So aiforuxr.org is where I’m publishing updates about what’s going on in AI. The last blog post was about the mixture of experts model, why that’s relevant and what that means for the future of our field. And I think the next one will probably touch on world designing and tools to think with and all stuff like that. So that’s a good resource. It’s also exciting for me to get back on social media, I guess. As someone who currently works at Meta, I have been very absent from social media, so…

Henrik Mattsson · 39:02
Okay. Good luck with that. I’ll follow you anywhere, so just tell me where you are. Okay, great. Well, thank you so much. Happy researching, everyone, and see you next time. Bye-bye.