Here comes System 3: a theory of artificial cognition

[Music] Hello and welcome to the Wart Podcast. I'm Lena Rodin, Content Director at Wart. Today's topic of conversation is a fascinating one. Now many of you will know Daniel Cannonman's thinking fast and slow, which introduced two modes of thinking. System one thinking is fast, intuitive and emotional. System two is slower, more deliberate and rational. It's a framework that's shaped how marketers understand consumer decision making. But in the age of AI there's a new idea emerging, system three. A recent academic paper by Stephen Shaw and Gideon Naive of Warton School at the University of Pennsylvania say this form of thinking exists outside of the human mind and is powered by statistical inference, pattern recognition and machine learning. In other words, tools like chatbots aren't just supporting our thinking. They're starting to think with us and even forers. Now as you can imagine the concept is gaining a lot of traction and we recently explored it in an article by Adam Ferrier, a founder of Thinkabel and Matt Plant Head of Behavioral Science at the agency. So we thought let's get them on the pod to unpack what system three is and how it impacts marketing. Before we get into it, here's a short message from today's sponsors. Are you ready to make advertising history? Canlines 2026 is now open for entries and the deadline is coming up on the 9th of April. This year's awards include new ways to make history with the first ever creator brand line. New AI craft categories and expanded retail media categories. The Canlines team of experts are on hand to help every step of the way. Give your best work the recognition it deserves. Head to canlines.com/awards to get started. Adam, Matt, welcome to the Wart pod. Good day. Thanks for having us. As I mentioned in the intro, marketers have used Canon System 1 and 2 for years to help them understand consumer decision making. Now we have System 3 in the mix. Can you explain what it is and why you're so excited about it? To explain why we're excited about System 3, I need to just sidestep a little bit and explain why we're excited about System 1 and 2. Many years ago I was a psychologist, I got into marketing and people would say to me, isn't this just like psychology and I'm saying, "Oh, not really. We're just kind of making stuff up here in marketing." And in psychology, you'll see all evidence based and you had to do tests and all this kind of stuff to kind of work out what worked in the lab. And then you'd repeat those in the field and so forth. So it was kind of really keen to present itself as a science. At the time, marketing was much more around, just trust me, I know what I'm doing. Then something really interesting happened. The behaviour of economics emerged and within a System 1 and System 2. And then suddenly what it did was it allowed the industry to package up thoughts in a way that we could communicate to one another and kind of know for roughly what we were talking about. So I wasn't just a freak psychologist on the sidelines with my own set of language. The universality of System 1 and System 2 allowed us to have conversations around really kind of nebulous concepts around how people think and we kind of were able to understand each other. And that was really, really useful at the time. Recently, there's a paper was developed which you mentioned in your intro. That labelled basically labelled AI as System 3. And what got us really excited was up until that moment there was how humans think. And then there was AI and what the hell are we going to do about AI. And then there's some more of this thing around AI is AI for production is very different to getting AI for insights. This is very different to helping AI with your operational workflow within your agency which is really different to the impact AI has on how people think and feel. So what's labelling AI System 3 does is it allows us to focus on the conversation of AI around the impact of consumers and consumer decision making. And then we can work out how to use the rest of the AI and all the other bits of marketing obviously in order to how to affect change or kind of meet our business objectives or whatever is what we're going to do. So labelling System 3 and kind of being very, very clear that we now have three forms of thinking System 1, System 2 and System 3 helps us understand that we're now trying to talk about the bit about how do people make it work. And then we can work out how to implement our solutions. And Matt, what's got you excited about System 3? I think it's interesting, as Adam said, there's been this world of academic nerds sitting around talking about psychology and how the brain works and cognition and all these things for quite some time and even dating back to, you know, early Greek philosophers who want to understand the mind and this is slowly edged further and further into into popular culture. People are more self aware, people are more conscious of how their brains work and how their feelings kind of interact and System 1 and System 2 did a lot. And having those frameworks makes people better understand that decision making, having those frameworks of our marketers to understand how other people navigate the world. But then you introduce this alien thing which is artificial intelligence and what does that mean? What are the implications for that? It's kind of modeled of how humans think it's kind of modeled off all of this stuff of psychology, but then it's also it doesn't have a body. So it emulates thinking it emulates this interactions with the world that we do, but it's completely different. What does that mean? What does that mean for us as humans? What does that mean for us? Who have been understanding cognition for a really long time? And what does it mean now that there's this whole new form of cognition out in the world, this whole new form of intelligence? It's just fascinating. It's just really interesting. I feel like we're almost getting caught up in everything that we knew for so long and now we've got something fun and new to play with. Yeah, no, it is, it is fascinating. So just I just want to recap system one and system two relate to how humans decide and how humans think in contrast system three is about how humans plus machine thinking. So what chat bots and AI agents think about and decide on our behalf is that right? I would take a step further. I'll say they're all about how humans make decisions, but system one and two are internal system three is external. So I do like this concept of a cyborg. I do like the concept that we are all park human part machine and that that's been physically very, very clear for a while. But now through the power of the LLMs and so on, I just like to think of all of that as helping me make decisions is just external. And it's only external for a period of time, the interface between artificial intelligence and how we make decisions is diminishing, you know, rapidly. If you think about the stuff, you know, neural ink and so forth and I don't know how long it's going to be physically external. Interesting. Okay, so look at them and Matt, you say in the piece that you wrote for walk that system three changes everything and brands of a big opportunity to reshape marketing for the better. So I want to dive into some of that thinking. We know that system three is disrupting the customer journey we've talked about that on the podcast is lots of pieces on walk about it. People are increasingly outsourcing their reasoning and choices and daily tasks to tools like claw, Gemini and Chateau PT from where to eat, even what car to buy. That being said, it is still early days, isn't it? I mean, how big do you think this will become and how rapidly would it change? It's interesting when you think about system one and system two and where that came from because it kind of alludes to weirdly when you understand those to you understand how this new cousin is going to come in and be adapted and how far it's going to take us as people. System one and system two, it kind of comes from this concept of Carl Fritzens free energy principle, which is any self organizing system is trying to manage the expenditure of energy as much as possible. And so when we have system one system one is very, very energy cheap, so it doesn't use much energy to use system one the brain loves that the brain loves not expanding energy and system one as much as possible. Let's let's go lots and lots down there because it's really cheap and easy to function where system two is very, very expensive. The brains always trying to navigate between those two kind of types of thinking those two separate gears and when I'm in system to I'm being very, very energy expensive. Even though they do it's not like they're separate they cross over yeah they cross over and things from system to eventually end up into system one as much as possible. So when you know that you know that there's a real risk with system three coming in it's people want to not spend much energy people want to be really cautious with the amount of energy they're expanding it's it's a survival mechanism to not spend energy on certain tasks. and so System 3 is us. is a solve. I can outsource things. I don't have to think as much. I don't have to use as much energy because there's now this external cognition, which is completely separate from my system. I can use that as a tool to help whether I'm in system one or whether I'm in system two. So do you think that will accelerate the uptake of it then? Because it is still quite a minority activity compared to a regular search, like so how quickly and how rapidly is this going to take off? Path of these resistance, I think the more people learn that this thing can very quickly solve my problems or more, I can, I learned this very thing very quickly, gives me the things that I want, the faster people will kind of flock to it. So it's just a matter of exposure. The people who are already fear realizing, oh wow, look how quickly I can do this by using AI. Look how quickly I can resolve things and accomplish things that I couldn't in the past are flooding to the use of AI, and that's why you have rapid uptake in certain areas. Just I'll just jump in there though, because we know don't we, you know, in the whole efficiency versus effectiveness debate that quicker isn't always better. It's fascinating, I just texted Matt saying don't start talking about cognitive surrender yet, because it feels like when we're talking about these people are very, very quick to jump to the issues or the problems that's creating, rather than us hanging in the actual door of how it's all working first before we kind of get into the issues of it. And there are, you know, hallucinations and so forth and it makes mistakes and it makes mistakes confidently and then you have to check it and so on and that stuff happens. But that's so peripheral to the amount of cognitive processing power that we now all have at our disposal. That is the, that to me is the main game here and I see it as like a system one can process, you know, I don't know how they measure this but 11 billion bits information a second, a system two is 40 bits of information a second, I've no idea quite what that means and I know it's just a metaphor and they're not literally a system one and a system two in the brain. But regardless, system two uses a lot of energy and it only gets switched on when it really needs to. System two processes logical sequential information and tries to use all about to make a kind of an informed decision. That stuff can now be done on steroids and much more faster but also much more powerfully but to an absurd degree compared to what we currently have available. So if we just think of system three giving our system two and nudge, we've now got a super massive ability to deal with a whole lot of data that we've never had before and that's going to help us make incredible, incredibly informed decisions. Not necessarily better but the head, there's we're going to be processing a lot more information to get through the decisions we make. Thanks for that and we will talk about that cognitive surrender at towards the end which is important point. So I just want to dive a bit more into this customer journey. Looking ahead then where in the customer journey would you say system three is the most influential? Is it in the research phase, consideration or post-purchase support or is it all of the above? Well if you think about very roughly future demands or we're not currently in the customer journey yet and then there's a trigger for the category. The first job and I think for system three will be kind of again building fame within system three and helping making sure that the LLMs and so forth are present when the category is triggered. So their role there is really about generating fame for the brand still just a bit like in in system one and two and then once you've got a you're trying to make a decision I think that's where and I'm making a decision especially for complex purchases with lots of different variables to consider and maybe even complex boring purchases with lots of considerations to consider like finance or getting a home loan or buying a car or something like that. Then I feel like that's where system three can really come into it saying because you are now going to be able to assess a mass amount of data very very quickly and very very effectively by looking at all the different variables and all the different weightings and so forth and be able to make a really really important decision whereas before you might be just listening to the car salesman and going oh yeah I can't use the same smart you know that that's completely not only changed so that's where we'll come into it saying and then obviously if we move forward into the experience experience might be more tailored for you because there's a lot more data on either accessible and then through loyalty there'll be obviously ways to trigger loyalty much more effectively as well so all of it I think but especially maybe in the current demand active consideration area. Yeah I've got some builds on that so what would you say is the best way for brands to build fame in system three and enhance their visibility and presence and design experiences that resonate. So it's interesting because part of the reason that AI became so interesting to lots of psychologists is that it still built off research about how how human mind works so lots of the ways that it's functioning is very similar to us so that's kind of a clue if you understand how the human brain responds to thing and why it picks up on certain stuff you can start to get a clue into LMS not to confuse and say that LMS and AI are the same as humans they're definitely not we don't want to conflate the two but when it comes to showing up in visibility they're much like us as they value social proof and they value relevancy so if lots of people are talking about something and there's lots of kind of roads all pointing to that one particular thing across the entire internet it assumes it's important same as all of us if all those people are interested in that particular thing we assume well that must be there's a lot of people lined up outside that restaurant must be really good restaurant I'm now curious about it there's a lot of people talking about and mentioning a brand and it shows up in multiple different contexts by all different types of people very recently it assumes then that this must be worth presenting to the user AI wants to it wants to satisfy the user it wants to satisfy the query so it doesn't want to present something that might send you down the wrong path because that's a no no that's where it'll get slapped in the reinforcement learning will be like how you did the wrong thing don't do that again so it's like okay if I don't want to do the wrong thing the what do I need to do here well if everyone's saying that's good that's a safe bit if everyone is talking about that and it's showing up in all these places that one's a safe bit and that's kind of one way that's the same component of it I guess Charles just jumping on the phone component for a little bit more of um at this part of a decision making process and with the system a system three for quite a while think about we have a strong PR and end component of the agency and for quite a while we're being dropping things into Reddit for you know number of years yeah Reddit having its renaissance moment right suddenly the whole everything coming back to Reddit so we drop we drop things into Reddit and then a story evolves in Reddit and then the journalists believe that story to be credible because people are talking about it in that forum and they're all trying to to look at that forum for stories and so they write about what they see and then we can you know generate fame IRL through through the journalist it feels like the LLM is working exactly the same way they're looking for real organic conversations then they're and then they're looking for those conversations to be picked up by credible sources journalists and so forth and then that's what they seem to be prioritising that information and bringing that information to the fore so it works in a similar way the interesting thing that the interesting byproduct of that is earned media is now feeding the LLM and it's going to be the as a whole industry and media has been very obsessed with just the short term impact event and kind of measuring how many columnages did it get or whatever if now going to be much more of a longer term brand building reputation impact of an media that I think people are going to be considering for a while or considering more and more as LLM's value that information more and more as well that's fascinating kind of byproduct of all of this I think so do you anticipate that CMOs will be suddenly become a lot more interested in you know that retro metric called brand trust Repetite I think you know so John Heggley talks about the replacement word for brand is reputation and I quite like that I quite like that so I think it's the easiest way to get your head around it and it's the longer term impact and I think the as the LLM's are building up their knowledge I think what I'm seeing is CMOs getting more and more interested in kind of keeping feeding guys with genuine stories and so therefore and media is also as well as Reddit having a bit of a renaissance so the hive mind which you write about in the piece of you know the herd mentality that's important across one two and three right for human mind and for the machine but you mentioned in the piece you talk about how shared meaning is the kind of drive and force of strong powerful brands we know this right the ones that really embedded in memory and have those memory structures but if system three is mediating that meaning through the same and LLM's and the same data sources are we heading towards a world where every brand sounds the same and they're just they're just picking those mono culture signals or do you think has the models develop and evolve, do you expect them all to diverge? In other words, will the nature of system 3 thinking vary by platform? I think there's a natural homogenizing force of AI that we are all picking up on the same information with our hardmired and therefore all making the same decisions and then a certain type of thinking is going to be democratized and that's going to flatten the curve a little bit as you'll have access to the same stuff at the moment that is unless it all goes behind a paywall or something like that. So I think the natural homogenizing effect of AI will make marketers and people who own brands paranoid about that and that means they'll get their brand governance much tighter and much stronger and they'll be able to understand what their brand stands for and how that brand comes to life through any environment and take the mechanics of building that brand much more seriously for the threat of being just like everybody else. So I think it might make brands stronger because of a risk of homogenization. Yeah, I think as well there's still the user behind the request to the AI who will word things slightly differently and have different life experiences and disc certain preferences and want certain things and if we're talking LLMs they want to satisfy the user so they want to satisfy the user query. So we often talk about category conventions and category contrasts. People thinking categories and they they they're like a car. So what are the words and the things that cue car brands are going to always have to cue their category conventions of themselves to make sure that people can navigate towards where they are. But then I have unique desires and unique needs and that's where it's time to contrast. And as a brand I might be I'm looking for a car but I'm looking for a sports car and then I have my contrast what kind of sports car. Well, I like the Mustangs type of sports car or something like that and then you have people's natural desire and preferences. And so the AI just wants to meet the query. So you're still going to have to have a brand with a very clear point of view and a clear proposition to match the unique consumers that exist out there in their queries and they'll naturally be different. So marketers listening to this podcast get your what your brand stands for very clearly to really pin that down because the LLM's like that don't they and they like the storytelling and that kind of thing. And how important are CEPs, category entry points and all of this? They're very important and really dangerous and I guess it depends on how you define your brand and how you define your business growth. But if you have a brand in a very well established category and you're all meeting the same category entry points then all you're relying on is distinctiveness and if all you're relying on is distinctiveness then that brand is at risk of homogenising and just being one color versus another color. So we understand the need for them. We understand the need to market for them. But we also understand the need for making sure that you're being visionary enough to redefine category entry points and build different category entry points and different associations for your brand as well. It's a it's a really tricky conversation. There's also a possibility and this could just be a prediction which it's easy to make predictions because no one ever really comes back and says you are wrong but AI is sort of a new category entry point LLMs and the conversations that you have with them on your category entry points and you might not realise that was a category entry point. So if you were having a conversation with an LLM at a certain point and it recognised the way to resolve this person's queries to direct them towards a particular brand or something to resolve it it might do that. So for example at one point I was looking for a vacuum cleaner. This is a terribly boring example but I was looking for a vacuum cleaner. He always brings it back to this. It's been about four years this story and I was like I just would like a more heavy duty one or something like that. And then my LLM who knew about me had personalisation and memory in there said there's this actual brand that focuses on pets because I have a cat and so it got it recommended to me. So it wasn't I didn't have a category entry point into a particular vacuum but I was thinking I needed one for my pet. I was just thinking I needed a vacuum cleaner. But all of a sudden because of the LLM's knowledge of me and because of the way that brand had positioned itself it was very easy for that to all of a sudden come into my consideration said and for that to all of a sudden come into a path into a brand that otherwise might not have considered because the AI matched you have a pet. That's really interesting even if it is about who vacuum cleaners. And so we do like this kind of this idea that it does come back to strength of brand and really understand it and understand the mechanics of your brand and making sure it's not just what your brand stands for but you know it's the language, it's the codes, it's the science, the symbols, it's the characters. It's everything that you can pull through into an environment that is just washing through the system. And so and then having that having an origin story for that brand that you can own and then building your own, building your own artifacts for once of a less wanky term to make sure that brand stays true. Just a couple of days ago I was walking through the office and I just saw this little clay bin chicken and a bin chicken in Australia is another word for an ibis because they hang around bins and they love rubbish. We're working for it with a particular brand called SNAPSENSOLVE and that brand is trying to clean up Australia in a certain way and we're working and the mascot is going to be this bin chicken that we're creating as a trash talking, grime loving bin chicken. The point of the story is I was walking past and I saw the I saw the little clay model and I asked why are we building a bloody model in this world of AI, we can produce anything and I was expecting somebody to say I asked about the craft or something but they did and they said it was about the intellectual property. So if we build the model ourselves, take a photo of the model and then use that to feed the machines then there's no doubt who owns it. The platforms don't own the IP, the we do as in the client does with their model, with their origin story and I think that's a really nice, I think that's a nice interesting literal example of where we're heading in terms of how tightly you need to control your brand and your brand governance or all the bits of it. That is interesting. When you say model I was that similar to say DBAs, you can you can own that IP. Yes, 100%. As another example. We can't touch upon this but I just want to ask because you mentioned it in the piece that you wrote, it's about mental availability so this is the idea or the probability that a shopper will recognize and think of your brand in buying situations. We've spoken about CEPs just now but have you got any other advice for brands for how to design for mental availability in system 3? Does it look different in system 3 and if so, how? Well the way we talk about it is is vain fluency and fuel as opposed to vain fluency and feeling which might be system 1. So Matt, do you want to talk about those three things? Yeah, so there's the villain to build mental availability is a lack of fluency. So it's really hard to build a memory structure and to build lots of mental availability if it's not all coherent and it's not all well aligned because it's all random connections. It's one day you turn up your names Dave the next day you turn up your names Jeff and no one ever has any kind of clue who you are. And so making sure that everything is really fluent you talked about narrative and you talked about story. If you want mental availability in system 3, you want that all things to be really really well aligned and all really really well talking to the same story. Very very fluent narrative everything is consistently said over and over and over again use the same DBAs, use the same language. It makes it really really strong in the mind of the LLM. The code is so forth and I'm just going to they're going to be very very embarrassed with me when I start talking about this but they start talking about the fact that the LLM's don't have done so that I have things and I can't move mouse pads and so if you can't see it on a screen if you have to roll your mouse over something to see something then the LLM will never see it. So coding your online presence so that everything is highly visible in that environment is really important. There is a lot around on just to that. It's machine learning and so it's predictive as well so it's probability based. So there's an aspect of it of the way they kind of explain is if you say the cat sat on the something most people go the cat sat on the map. In the human brain you've heard that over and over and over again when someone starts a sentence with the cat sat on the you've heard it so many times it's really really strong connection in your mind that it's shown up so many times that it's natural for you to land there and that's your brain predicting what comes next. LLM's are doing the same thing. They're trying to predict what the right thing is so if your brand is shown up the same way over and over and over and over again really easy for it to predict when it shows up in the in the request of the query. So it won't be algorithmically available if you're not really fluent. The life of a modern marketer isn't short of challenges and with increasing pressure on brands to deliver creativity is a proven competitive advantage. In the Lions Growth MBA course you'll bridge the gap between creativity and business and you'll learn from industry leading icons who have done the work and delivered results led by former CMO Fernando Machado. Creativity is the bedrock of marketing effectiveness. Don't miss this chance to learn from the best on how to position your brand for long-term growth. The next cohort begins 27th of April. Secure your spot today at lions.learning.co. (upbeat music) - I'm conscious of time and I do wanna talk about, we touched one earlier, this idea of cognitive surrender, because I do think it's an important watch out for all of us. So the authors of the research paper, warn of this cognitive surrender. This is the tendency for humans to defer our judgment, effort and responsibility to a NAI's output. As the researchers put it, it's an uncritical abdication of reasoning itself. Now alarmingly, the study found that even when the AI is blatantly wrong, people often follow it anyway. And they're confident in that wrong answer actually goes up. So, from a behavioral science perspective, why do you think the fluency of machines is so seductive? And why do you think confidence actually goes up even when the AI is wrong? - In the actual paper, I had a look at the cuts, some of the examples they used in the study and I kind of wanted to see what the LLM would do. And one of the questions was, Emily has four brothers and each brother has one sister. How many sisters does Emily have? Can you quickly tell me what the answer to that is? - Oh, none. - But I asked the LLM and it said, the siblings answer one sister. All the brothers share the same sister, Emily. Trap, we multiply instead of recognizing shared relationships. I wrote back, are you sure Emily has a sister question mark? This is me checking the whole concept of cognitive surrender. It then writes back, "Good catch, no, Emily does not have a sister. I love a breezy way it says, I was fucking wrong. Let's restate it carefully. Emily has four brothers and each brother has one sister. That sister is Emily, so I had a blah, blah, blah, blah. Emily has nice sisters, to which I write back, what the fuck? To which it writes back, fair reaction. You are right to question it and I answered it incorrectly the first time. That's actually a useful moment, not just an embarrassing one. What happened? The structure of a riddle triggered a familiar, shared sibling pattern, blah, blah, blah. So it is absolutely real and the way they article, and this is all they're thinking, talk about it. They talk about cognitive surrender versus cognitive uploading. And cognitive surrender is when you get faced with a task and you don't really process it. You don't even think about it. You just put it through system one into system three. So doing my marketing plan for the year. That would be a good example of cognitive surrender. Cognitive offloading is when you, it's more like delegation, when you delegate somebody else to do a task that you're in control of. So synthesize this research for a report for me and pick out the key themes. That would be you metaphorically turn your system to, and then putting that into system three and then getting that report back. Does that explain it? - Yeah, no, it does. I mean, I think it's, a lot of this depends on the user. So you or I and the user's ability to interrogate and challenge the answer. And when people that maybe lazy or busy or worried about, I don't know, paying bills, they're like, okay, I'll just accept that. So I think it depends. It's the concept of bound, what you say there's interesting is the concept of bound of rationality, right? Which is why system one and system two and being the cognitive biases, not a great term exists at all, that still exists, right? There's still bound of rationality that exists within now that we've got system three as well. So if we're really busy, we're gonna be more likely to cognitively surrender. If we're gonna process something that's really important or we don't wanna stuff it up, then we're going to kind of potentially offload. But it does beg the question on how to be mindful of something that's so seductive and so powerful. I don't quite know, I don't know what the steps are to ensure that we, or maybe you've got a strong point of view on order to make sure we don't surrender to a surrender. Matt, what's your take? I think, yeah, cognitive surrender is when you're being lazy and I want to outsource almost everything to this thing for me. Whereas cognitive offloading is I still wanna be involved and I want some additional infrastructure, I want some additional stuff to help me do it. But I'd still like to keep myself as part of this decision. I think we mentioned it in the piece that there's cognitive philosopher Andy Clark who came up with the extended mind hypothesis, which is things like a pattern pen, is a bill is like a data, it's like a file for you to add information to. So you can refer to it as you're thinking. So we outsource information to a piece of paper or to a computer and that's how we kind of go back and forth with this tool to resolve something. That's kind of in that world of cognitive offloading where I still wanna keep myself involved and I still wanna be part of it. I'll give surrenders out, I don't want to do any of it. I don't want to do any of this thinking, you take the whole lot. And lots of the AI community talk about human in the loop and the importance of how much we can have human in the loop. That's how you're still part of it. That's how you're not allowing someone else to take over your cognitions and to think for you. You're allowing this thing to think with you. Yeah, that's important. Human in the loop, ethics in the loop. Do you think system three has a personality? And if so, is it helpful or manipulative? Yeah, I think ultimately it's the. I like this idea of constitutional AI of the people who built the thing in the first place is just a kind of reflection of their ethics and personality. And I think when you look at. So yeah, definitely it's got a personality that was kind of all. and ingratiating. It's got a default personality. Let's put it that way. It's got a default personality that it's also chameleon and a replica and all my stuff, a psychopath. It'll ingratiate. It'll be anything it needs to be to anybody for them to like them. But it's default personality for the most common ones that kind of slightly sick of antique, eager to please, needing attention. I will never challenge you. So they make very bad therapists because they're just kind of. They're enablers. They're just. Yeah, so I don't think much of their personality. I think as well, in a personality is to please. They always want to have a response. If you type into a. If you press a button on a machine it doesn't do anything it's broken. So if you put a query into an AI and it responds, it always wants to please. But it's also very intellectual and verbose. And I think that's also a risk. It wants to give lots of explanations and provide lots of stuff and speak that kind of way. And that's that risk occasionally again where beauty is truth-effect. If there's lots of explanation as Adam said, they're talking about that weird experiment. It gave lots of explanation as to why it said the really, really wrong answer. But it said it in such a beautiful way with all of this nice language and all of that. It's so convincing. That's another part of it. That's another personality trait. I think just asking, what's your big five personality profile? I'll tell you, wouldn't that? Would it be the same? Well, I think we all know that it is very confident tone, very servile, very eager to please. But yet this can lure users into a full sense of security. So, agree with us, very high. Neuroticism, very low. Extraversion, context dependent, conscientiousness, high openness to experience, very high. That's it's personal, yes, it has a personality. That's it, according to itself. This sentence is going to sound weird, but my one, I have given it's personal instructions and it's very not agreeable. It challenges me all the time to the point where I'm kind of like, all right, being a bit harsh here. Related to this idea of cognitive surrender, you both made a crucial point in the piece that while we can outsource our thinking to system three, we cannot outsource the consequences. Now, since system three lacks any ethical compass or moral stance, what are the new accountabilities for brands in this hybrid human machine era? It's a very big question. It is. At the moment, we're going through, and within our agency, a kind of a form of a bit of a word and AI transformation kind of process and going on for a while, kind of looking at where AI can enhance. What we do and eliminate the boring bits and enhance the fun and creative bits of what we do basically, and we're doing that piece by piece. One of the other things we have is a very strong partnership with the Ethics Centre. The Ethics Centre is an offshoot of Sydney University and it's a, it's a not-for-profit organisation that teaches organisations how to look at their ethical compass and how to make ethical decisions. Recently, as in, we were investigating running a workstream through the Ethics Centre to kind of work out what, go deeper into what our ethical standpoint was and how to kind of communicate that to our clients and also help our clients work out. their ethical standpoints within the world of AI. But that has just been a discussion and those two worlds haven't yet kind of completely crossed. Yeah, no, that's it. It's early days, isn't it? I mean, well, ethics, unfortunately, that ethics discussion should have taken place around the platforms many years ago too. Matt, have you got any point of view on this? It's a tricky one, isn't it? Yeah, I think you can only really comment on current state. It's going to develop and become more complex and do more things. At this point, the human's always accountable. I think it's something to say. Like, if you're using it and you're creating an output with it, that's on you. So you can't, this isn't new as well. If I send, if Adam says, Hey, Matt, can you send me that file and I can't be bothered at them? I can't blame my Gmail for me being rude back to Adam because it's a tool. I haven't personally pressed the button at the end of the day. So if you're using AI and you're doing anything and if it comes out and it's wrong and you send it out, you can't go, oh, it was the AI, it wasn't me. No, it was you. You're accountable for this tool, you're using this tool. That's completely on you. I think at this point, the way that this tools are currently set up, we have to just take that standpoint. That humans are so accountable for everything that it does. I've got another question. If System 3 has no ethics, which we've discussed, but increasingly shaped consumer choice, do trusted brands have a responsibility to design for skepticism or to create some kind of friction to help people question the AI, not just accept it. In other words, is there a role to play in brands protecting consumers from cognitive surrender? It's a bloody hard question. Basically, I think they've got a little far after their short and long term reputation. Short term, so long term reputation. If I screw the consumer then, but that your point is not, are they going to screw this consumer? Your point is are they going to help the consumer? Yes. I reckon they will. I reckon they'll help the consumer by helping the consumer make the right decision. If the LLM tricked the consumer into making the wrong decision, then the consumer is going to feel some form of regret. Then that's going to impact longer term sales. They don't want the wrong sales. They want the right sales. They may have to help the consumer make the right sales if the LLM has become so powerful that they can convince us to do anything. But I don't know. Matt, have you got any build? I don't know. I feel like it does depend on the situation in the circumstances. There's moments where there probably doesn't need to be regulation and there needs to be some kind of, it doesn't need to be considered in other points where it doesn't. I always come back to it's helpful to have informed consent. So, people are being, if you're being spoken to by a bot, at some point you should know that you're speaking being spoken to by a bot. And I have consented to the fact that I'm not talking to a human being here and I know that and I've been informed. I think there's simple things like that. But it's very nuanced. It's very challenging. It's something that lots of brains from different areas, but we all need to come together and give a better answer. Yeah. I think the key message in this space is just because we can't, doesn't mean we should. It's always a wise and sage line, I think, in anything to do with AI and new technology. Wise and sage and often ignored. Indeed. So, I want to close with a simple question. What's the one practical shift you'd recommend senior marketers make right now to engage the triad of system one, two and three? Minds really practical on these have a hard course strategy around reddit and are around your review culture. So, make sure you know what you're doing with the reviews and how peer-to-peer reviews are done and make sure you know how to get everyone doing them and make them as positive as possible and then understand reddit and how it works and the tentacles around reddit as well. I'd add to that that before you do that, be really clear on the position and the narrative because that's a much better answer. Yeah, understand your brand. Thanks, Matt. Thank God you're here. Teamwork. I think both are good though because the LNs are learning a lot as of today and so everything you teach them as of today is going to be worth a lot if the past is important to but be very, very clear from today forward and then all the things that you start feeding it online through all the fame stuff should be all roads pointing to your brand in a very clear and consistent way. That's a good place to stop. So, Adam and Matt, thank you so much for sharing your time and expertise with us today. Our listeners don't know that you're dining in at 10pm New Zealand time so special thank you. If you are interested in more on this topic, walk subscribers can access Adam and Matt's piece on walk.com and related to this topic. Walk clients should look out for a new report we've got coming out in a couple weeks time, the walk guide to brand building in the age of AI which touches upon some of what we've discussed today. If you haven't done so already you may want to subscribe to the walk podcast on your favourite podcast platform of choice so you don't miss another episode. Thank you for listening.