#195 The AI Instinct

0:00All right, welcome back to our extra AI podcast series. We are in the season 20 and today I have a interesting topic and with me is Rana Gujaral, Rana Gujaral is the former CEO of behavioral signals. He's currently doing a stealth startup. He's also the author of the AI instinct, the future of AI and human decision making, which is a very fascinating topic. So I would kind of get into this a bit more. I'm kind of putting this episode as the AI instinct when AI starts shaping how we think. So what do you say? Welcome on board Rana. What do you say about that? Yeah, so you know, I think what worries me about that specific scenario and it's not the obvious thing. It's the obvious risk is a battered vice at scale. You know, agent hallucinates misreads a market signal and suddenly a million people have made the same wrong move.

1:07That's real, but it's actually the easier problem to reason about. I mean, you can bench market, you can regulate it, you can put Godrails around it. The deeper risk is what I call drift. In the book, I talk about how AI really arrives as a rupture. It usually arrives as a default. I mean, in business decision pipelines increasingly surface AI recommendations before humans have human, the human has a fully formed an opinion and now apply that to investing or for that other matter. I mean, the agent doesn't just execute what you ask it anticipates its nudges. It presents three options where two are suddenly framed to guide you toward the one it has already decided is optimal. And because the interface is warm, adaptive, personalized, the influence stops feeling external. It kind of feels like you. And I think that's the dangerous part. Not the wrong answer, the right feeling answer that was never really yours.

2:10And so I guess, you know, I'd say the catastrophe loss scenario is manageable. The agency erosion scenario is not because nobody sees is happening. People will feel confident, more informed, more in control of their financial lives, while quietly outsourcing judgment on the one domain where judgment used to define adulthood. And that's the alarm. You know, I'd like to sound not that the agent will be wrong. That it will be persuasive. Right. I think before we get into that fascinating. I think before we get into deeper into this conversation, I first want to ask your journey, your personal or professional journey. I know you have worked across enterprise technology, entrepreneurship, machine learning and behavioral AI. So what let you do this. Intersection of AI and human decision making because this is a fascinating topic. Could you elaborate a bit or could you provide some personal backstory or a professional backstory about how you ended up doing what you're doing now.

3:22Yeah, I mean, yeah, I would say, you know, the way I arrived at that, it wasn't a demo. It wasn't a benchmark. It was just a simple, let's say, a drafting something. You know, and I don't even remember what. Or and I noticed that stopped using it the way I use a calculator or a search bar. I was using it the way I use my own mind, you know, testing an argument, framing a decision, letting a half form thought bombs off the system and come back. Just a little sharper and the strange part was I couldn't tell where am I thinking and did and it and it's thinking began. So that's when it hit me is every prior tech we've sat every try a prior tech wave sat outside the loop of cognition. The internet gave you access to information, you know, the smart home put that access in your pocket and the cloud computing scaled it, but in every case, you still walked up to the tool, used it and walked away and the tool didn't sit inside the moment of judgment.

4:23It sat next to it, but AI is different because it's inside the loop. I mean inside the moment where you're farming a belief, weighing a trade off deciding what you think. And once a system is in that position, the question stops being how clever it makes you sound in any single instance, the question is what using it does to your time. Does it build your judgment or replace it does it create continuity or does it just create a convincing stream of outputs that feels like thinking without actually being thinking. So I think that was the moment because I realized we've been measuring this wave with the wrong yardstick we keep asking how smart is the model, how fast, how accurate. And those matter, but they missed the access that actually counts, which is the relationship, the system has to time to contacts to and to consequence in your life. So, you know, prior waves change what we could do. This one is changing what it means to think and I don't think we built the vocabulary yet culturally to talk about that.

5:31Honestly, and that's the that's part of why I wrote the book. Great. Great. I think a great background. And I think before we get, get going, I have another question. I know we usually think AI has something that helps us make decisions. But I believe now we see that we are at a point where helping us decide is becoming shaping how we think. So, at what point is that happening this way like helping us decide is now becoming shaping how we think so even our thoughts are being influenced by what we read or what we kind of get into what kind of rules we use. Yeah. I mean, decisions used to be discrete events. I mean, you'd gather information, weigh it, choose and then act, right? There was friction in the middle and that friction was where the judgment lived.

6:50And AI is collapsing that middle. I mean, increasingly, the recommendation arrives before the deliberation does. I mean, you open the dog and there's already a draft. I mean, you opened the dashboard and anomalies are a flag with the suggested action. So, you know, you open the message and three replies that pre written in your voice. So, the decision isn't gone, but the space before you would have formed one on your own has been quietly prefilled. What that changes practically is the ratio of choosing to editing. You used to choose now we mostly edit and editing feels like thinking, which is the trap. You still have the sensation of agency because you click to prove you maybe tweak the phrasing you picked option B over a. But the frame you were choosing inside was set by the system and not by you. And that's the phrase I keep pointing to is that autonomy is becoming. Gradient rather than a guarantee it's not that I take decision away from you. It's that it narrows the purchase of what you consider before you even know that an aperture exists.

8:01The options you never saw often, you know, don't feel like options with held. They feel like options that that don't exist. And there's a check in shift, which is the disappearance of productive friction. And in creative work and science in leadership. The moment that matter are usually the moments where something doesn't fit the contradiction the pause the where, you know, the where that came from, you know, a well designed AI can preserve those friction surfaced contradictions show alternative explanations, ask for your reasoning first. But a poorly designed one or optimized for engagement one smooths them away. I mean, and smooth feels good. I mean, smooth feels like competence and feels like good AI. The honest answer is, you know, AI is changing decision making less by making us wrong by but more by making us fluent and fluency without friction is the very particular kind of risk because it doesn't announce itself.

9:05Right. And also it can be a bit more dangerous because I think sometimes like you said, rather than thinking and coming up with choices, you are presented with choices and then you are. Sometimes I think you end up editing thinking that it is a different kind of a choice. Let's get into this book. I believe I want to get into this conversation. I think with your understanding the new frontier, I know your book is called the AI instinct. So what what exactly is that AI instinct that you're talking about and how is it different from the way we typically think about AI. I mean, I say, I mean, I kept sitting in rooms sometimes with brilliant engineer sometimes with policymakers sometimes just with friends at dinner and the conversation, AI conversation will collapse into

10:11one of two shapes. Either it was a race when we hit a GI who gets their first whose model is bigger or it was just panic like will the machines take over. Build the take our jobs are recooked and I walk away thinking we're arguing about the wrong thing. I mean, both of those framing treat AI something that happens to us and I will a moment, you know, them. But that's not what I was seeing at my in my actual work. I, you know, I spent my days building systems that retone, hesitation, affect intent and what I could see up close was that AI wasn't arriving. It was already threaded into how people were forming decisions, feeling things, choosing things, not through some dramatic rupture through defaults instead. I mean, through convenience instead, I mean, through interfaces that anticipate you. So the book came out of wanting to name what I was watching, you know, that we're not heading to a standalone super intelligence that shows up one day we're heading into a hybrid condition where human cognition and machine cognition are coupling.

11:19And the interesting questions live at that scene, what does it mean when a system knows your emotional state before you've articulated it. What happens to agency even consent becomes reflexive. What does super intelligence even mean if it emerges through the coupling rather than an external entity. And there's also a personal piece like I've always, you know, been the kind of reader who jumped around in books and read the last chapter first works backwards. And I realized that AI discourse had been doing the opposite. Everyone was fixated on the ending, AGI, ASI, the singularity and nobody was doing the harder work of the middle chapters like the lived the textured present tense reality of what these systems are already doing to us. So I wanted to write the middle chapters and that's really it. Great. I think you've put it in a very nice succinct way. I think people are talking a lot about the end point whether it is the utopian future or the dystopian future. But I think you're focusing a bit more on the journey and the middle chapters, which gets very interesting.

12:24I think this is where I think you're also terming the human cognition and the machine cognition. I think this gets into the context of. Neuroscience and behavioral science, which is getting very important. I think nowadays I need I know that a lot of current generation Gen Z kids are getting into this newer subjects of neuroscience and behavioral science. So where do you see this neuroscience and behavioral science change or understanding of how AI should be designed or do you think it's a wrong question. I mean, I think let me push into a different layer of this because I think I gave you the individual user picture last time. And I think there's a systemic dimension that scares me even more when you have millions of people making financial decisions through agents. You're not just aggregating a lot of individual choices.

13:27You're routing them through a very small number of underlying models, maybe three, maybe five, and those models share training data, share architectural assumptions, share whatever risk framing their developers baked in. Often without articulating it and that that's correlation risk at scale. We have never really had in the old world. You had thousands of human advisors with different biases. Different regional exposures, different reads on the same market. That heterogeneity was actually a stabilizer. Markets need disagreements to function prices are just disagreements crystallized into numbers. So, you know, like now imagine every retail investors being counseled by descendants of the same two or three foundation models. When those models see a signal, they interpret it in a correlated ways. When they hedge, they'll hedge it in correlated ways. And when the panic, the models do have something like a panic response when inputs go outside their training distribution. I mean, they're all panic together.

14:31So I think that's like a flash crash waiting for a trigger and it's not a bug. It's like a structural consequence of consolidating cognition through like a shared substrate. So now bridge that to the hybrid cognition idea from the book. In the healthy hybrid system, the human and the machine bring different things, different tastes, different context, different failure modes. And that difference is the whole point. But if the human is deferring completely and the machine is the one of five near identical machines, you haven't built a hybrid at all. You've built a monoculture varying personalization skin. So the danger isn't just what one person gets bad advice. It's like we quietly rebuild the financial system on a substrate that fall a failed in unison, right? That's the version of this. I don't think regular does have fully absorbed yet. So to kind of get a bit more deeper into this, I know human human communication

15:33is not just about words. I think it's much more deeper than words. We have emotion. We have intent. We have trust and all these things. But when we look into AI and these AI agents, I think you've kind of briefly touched on that. How can AI understand this emotion or trust or intent? When we compare it with humans, I think what are your thoughts? Do you discuss about this a bit more in your book as well? Yeah, I mean, this is actually one of the through lines of the whole book. It's the thing I've spent the last decade building and practice. So the honest answer, AI doesn't understand emotion the way you do not yet, right? Maybe not ever in the

16:33phenomenological sense, but it can do something that turns out to be surprisingly useful. It can infer. It can pattern match across signals that humans are constantly emitting and often can't articulate ourselves. So think about like what you're doing right now talking to me, you're not just parsing my words, you're reading pace, pauses where my pitch lifts, where it flattens, whether I sound rehearse or actually thinking. And that's happening in your in you in milliseconds unconsciously. It's what the system one territory is what I call in the book. It's the fast, intuitive, emotional read that happens before deliberation kicks in. And so for a long time, AI was built entirely for system two, which is logic, reasoning, computation. Brilliant at chess, useless at reading a room. What's shifting now, and this is the work that has recently happened, is embedding system one dynamics into machines, weistone, prosody, facial micro expression,

17:37skin conductance, heart rate variability. Now, when you layer those signals together and a model can infer frustration, hesitation, confidence, deception, and warmth, not because it feels those things, because there are consistent patterns across humans, across languages, across contacts, and neural networks are extraordinarily good at learning patterns. So the thing that you need to be careful about is detection is not understanding. So a system that flags or frustration is not empathizing with you. It's simulating rapid emotional inference, and that distinction matters enormously because if we forget it, we start trusting these systems the way we trust our close friend and they haven't earned that trust. And that's why in the book, I argue about something I call a synthetic empathy layer. Basically a reflex on top of a reasoning core. So the machine can't sense the emotional weather of an interaction before it responds. Not to fake feeling to behave appropriately,

18:40because brilliant AI without emotional awareness is essentially a high functioning psychopath, capable, articulate, and dangerous. So the real question isn't whether AI can understand us, it's whether we can stay littered enough to understand what's it's actually doing. I like the way you've put it, the synthetic empathy layer. So I would like to now, I want to dive a bit more into when AI is entering into the human decision loop or human decision layer, because this is where we see that most enterprises, they measure AI through a lot of different things, right? Productivity or whether it is efficiency or whether it is some kind of KPIs or written on investments. So here, are we overlooking this impact like how does AI change this human

19:41judgment and behavior? Are we overlooking that or what's your take on that? Yeah, I think responsible deployment to me is less about a checklist and more about where you locate friction. Bad deployments remove all friction in the name of user experience. Good deployments put friction back exactly where judgment needs to happen and remove it everywhere else. So let's say in the book, I talk about memory evolves on device learning, preference models, shared context graphs, those aren't just buzzwords, they're infrastructure choices. Memory wall means the system keeps a record of what it recommended and why and what happened next so you can actually edit or sorry, audit its influence or you over time. And most deployments today have no such thing. I mean, AI gives you an

20:45answer, you act on it and the trail evaporates. That's not responsible, that's just convenient. The on device learning matters because it lets the system personalize without extracting your life into something someone else's data data lake. Now preference models matter because good decisions has to be defined relative to what the user actually values, not what just maximizes engagement. And the shared graph context graph matter because if the human and the AI aren't operating from the same situational awareness, you don't have collaboration, you have a very confident stranger whispering in your ear. And then there's the interpretability piece and I'm quite firm on this, this accuracy is not enough. A system that that's right for reasons you can't inspect is a system you can't push back on. So responsible deployment means that AI can show its work in a form, a non export can actually reason about not just a technical report, but you need an explanation.

21:49And the people, the peace people skip usually is cognitive liberty, right? So the right to turn it off. I mean, the right to opt out of a specific feature because without losing the whole product, chili wrote mental integrity into their constitution. That was very interesting. It happened recently. And most companies won't even let you disable auto complete. So in practice, love the influence, explain the reasoning, protect the opt out and design so the humans still forms an opinion before the system offers one. And that's the last one is the hardest because it costs you engagement metric, which is exactly why it's the test or whether you actually mean it. Right, right. This is all very important points that you have raised. I think you have brought it up whether it is logging the race and influence and how do we explain the reasoning and these different aspects. But when we go, I think in your in your work related to behavioral

22:50and cognitive AI, you might have learned differences between what people say, what they feel, and what they actually intend to do. What have you learned about that? Because I think you might have mentioned or you might have talked about this in your book as well, right? Yeah, I mean, this is one of my favorite territories because it's where the whole feel gets humbled fast. You know, here's the thing. I mean, humans are lousy narratives of themselves, not because we're liars, but because the story we tell about why we did something is almost always constructed after the fact. But decision happens in that submerged ocean I talk about in the book, the subconscious layer and the conscious mind gets handed a tidy summary and told here, if plain this, right? So when you ask someone why, what they think, you're getting the press release version, not the actual,

23:52you know, version. So in our work, we see this constantly, right? Someone will fill out a survey saying they'll love the product experience. Five stars, great meanwhile, their voice during the actual interaction was tight. Their pauses were long. Their pitch was doing this apologetic lift at the end of every sentence. The body was just screaming friction. The words were saying satisfaction, but the body was screaming friction. So both are real data that they're just measuring different layers. So what I've learned and this took years to really internalize is that there are at least free channels running in parallel. There's the state of channel, what you say, there's the felt channel, what your physiology and prosody are broadcasting, whether you want them to or not. And then there's the intended channel that what you're actually going to do next, which correlates more with the felt signal rather than the stated one. So if you want to predict behavior, listen to the body. If you want to predict that what someone will report on a survey, listen to the words.

24:56So the uncomfortable implication is that emotional AI is often reading people more accurately than they read themselves, which is powerful in a clinical setting, in a coaching setting, in a safety setting, and generally dangerous in a persuasion setting. So that's the ethical fork. Same technology used to help someone understand patent they can't see is liberating used to exploit a patent they can't see is it's manipulation dressed as personalization. So and the person under receiving end usually can't tell the difference in the moment. So that's I think the thing to watch out for. I like I think the aspect of what you say, I think if you want to see how I think I like the way you've put like if you want to see what people are going to fill out a survey, I think it's more on what you say and what they would they are going to if you want to predict what they're going to do, I think I think this is where you have to see what they're feeling and what they do.

25:58I think these are some of very important aspects, but now is this is this is if people are increasingly trusting AI, is it because they're is it because AI is sounding confident or is it because AI knows more of my preferences as such or do you think it is more of AI assistance is how do you think if AI assistance is becoming more of an AI over reliance. What did you see from your interactions in your work of do you can you provide a few thoughts around that. Yeah, I mean, I'd say I push back on some of the framing because replaced zooms the game is human versus machine and that's not the game that's actually being played. And what I've seen in my

27:00work is that in companies that advise is that people, people getting the place displaced fastest aren't the ones with the deepest expertise. The ones doing the middle layer that template it, the repeatable, I do this the same way every time that that work is generally at risk pretending otherwise it's just dishonest. But the people who are becoming more valuable, not less are the ones who have figured out how to fuse with these systems, not use them like a search engine fuse, meaning they bring the thing machines don't still don't have, which is judgment, taste, context, goal formation, a sense of what actually matters here. And they let the machine bring what is good at memory scale simulation, tireless iteration, that combination is not a smaller version of you, it's the larger version of you. And I had a moment recently working on a hard problem where I used a model as a soundboard back and forth. And it surfaced an angle I hadn't considered. It didn't

28:03solve the problem I did, but I wouldn't have gotten there alone, or at least not that week, maybe even not next week. And that's the shape of the future at point people toward not being replaced but being extended. So the practical advice I guess is stop competing on the access machines are optimized for, speed, recall, volume, force, draft, fluency, you'll lose that race. Compete on the access, their weakest at original framing, no, in which question to ask, understanding what a client actually needs versus what they said they need, taking responsibility for an outcome, building trust for another human being over time. That's some very important part. And one more thing, I think straight literate, people who understand how these systems actually work, what they're good at, where they'll hallucinate, where to override them, will always have leverage over the people who either fear them or worship them. And both are just, both of those questions give away your agency and curiosity actually keeps it. So you're not just being replaced yet, you're being asked to evolve.

29:10That's the harder honestly part. And that's also more interesting. Right, right. I think it's not that, I know at least in the near future or in the short term future, you're not yet being replaced but it's more about how you're evolving and how you're going to use these AI tools to expand yourself or expand your thinking and go much more further and kind of tie it into the business outcomes or the outcomes that you're looking forward. Let me get into this aspect of the cognitive AI, the hybrid intelligence and the future. I know organizations and behavioral organizations use quite a bit of AI, but what the youth foresee, like how can organizations use the emotional and behavioral AI to create better experiences? Because here, the minute you start using emotional and behavioral AI,

30:19there is a fear that you're crossing the line, crossing the boundary from personalization into manipulation. So what are your thoughts? Like how should organizations evolve in using this emotional and the behavioral side of AI? Yeah, I mean, the line for me sits at one specific place. I mean, does the human still form an opinion before the system offers one? If yes, you're augmenting. If no, you've quietly crossed into replacement. Even if a human is still nominally in the loop. I know that sounds simple, but it's the operationally the right sharp question to think about, because the most systems today are built to offer the first answer. You open the interface, the suggestion is already there. The autocomplete is already halfway through your sentence.

31:23The recommended action is usually pre-selected. That's not augmentation. That's a system doing the cognitive work of framing and then handing you a yes or no button. You're not thinking. You're just ratifying. So augmentation looks different. Augmentation is when the system waits. When it makes you articulate the question before it helps you answer it, when it shows you three framings instead of one recommendation, when it asks what you're actually trying to decide before it optimizes for a proxy. That's a coupling I care about because it preserves the thing that makes you you, which is the act of forming a view under uncertainty. That's the test I use. If you took the AI away tomorrow, would the person still be able to do the underlying cognitive work? Slower, okay, that's fine, rougher, that's fine too. But can they? And could they? If yes, you've built augmentation. If the capacity has a trophy to a point where removal is disabled, you've built

32:27replacement and you probably built it without meaning to. And this is where I get uncomfortable with a lot of current deployments because nobody sat down and said, let's replace human judgment in this domain. It just happened by a creation. One convenience at a time, each individual step felt like augmentation and the cumulative effect was just replacement. So the line isn't a line in the technology. It's a line in the design choices. And honestly, in the user's own discipline, the system won't protect that line for you. You have to protect it, which is why keep coming back to explanation, friction, and right to form your own opinion first. Those aren't features. They're what keeps augmentation from silently becoming something else. Right, right. So it's more important about, have you learned your journey? Have you learned the process? I think if you replace or if you remove AI today, do you still understand what you have built? Or do you, like you mentioned, I think it might be rough, it might be slow, but did you

33:31understand how you evolved? How have you arrived there? So if that is the case and you're really building your augmenting what you have, if not, I think you're just getting into the into the era of manipulation or taking over where AI can take over you at some point of time. I believe I think this is where you're also talking about more about using these AI tools. I say, sounding board for you so that you can get a better version of you, a better way of getting to your results. But as we move towards cognitive AI or these and these brain computer interfaces, and we have these different forms of hybrid intelligence. So where do you see in future humans and machines collaborating over the next decade? I know these things are coming faster than what we are thinking about, but where do you see humans and machines collaborating in the next decade or two?

34:39Yeah, I mean, but before I answer that, I mean, one thing I want to come at is from a slightly different angle because I think there's another dimension which we should pay attention to, and it's the temporal one. Augmentation and replacement aren't just two categories. They're the same trajectory viewed at different points of time. Something starts at the augmentation and if you're not actively resisting the drift, it becomes replacement. Not because anyone flipped a switch, but because human capacity is used dependent. What you don't practice, you just lose. There's a researcher I cite in the book, Vivin Meng, who's been looking at cognitive offloading and what it might mean for long-term neural reserve. Her concern, and I can share it, is that the brain is not a static organ waiting to be held. It's a living system that maintains itself through the very effort you're trying to spare it. So when you offload judgment, memory framing,

35:44you're not just getting a productivity boost today. You're withdrawing from account, you will need in 20 years. So that reframes the question for me. It's not, is this tool augmenting or replacing me right now? It's what the 10, what's the 10 year derivative? Like, if I use the system the way I wants to be used, will I be more capable in a decade or less capable? And honestly, for a lot of what's shipping today, I think the answer is less capable, not because the tools are bad, but because they're optimized for the current interaction, not for the human you're becoming through repeated use. And I think the writing analogy helps me here, writing outsource memory. That was an ad game, because I'd free cognition for other things, and we built a whole civilization on top of the surplus. But hybrid cognition may be outsourcing judgment itself. You know, that's a different kind of trade. Memory you can recover from book. Judgment once a trophy is much harder to rebuild, because judgment is what tells you when to trust

36:48the book in the first place. So the line in this framing is the rate of change. I mean, are you getting sharper or dollar through the coupling? Assign is question. Most people, myself, included don't ask it enough, I think. I think, yeah. So I think using AI tools again, going back to that much more like a sounding board and getting to the right and iterating it and getting it better out of it. I think that makes you more that makes things more augmentative than replace replacing. So in future, I think this is a futuristic question or I think, where do you think the measure breakthrough from AI would come from? Do you think it will come from making machines more intelligent or intelligent or intelligent or making the machine substantially more better at understanding what the humans want and what we are doing?

37:54You know, I think I think it's a little bit of an oath. I mean, I think, you know, and I think it's sort of, and the question I, you know, sort of add to the design test and the tenure test is legitimate test, right? I mean, where was the handoff? Was it deliberated? I mean, did the affected people have any say or did just happen because it was efficient? Because some cognitive work should remain deliberately inefficient, judge weighing a sentence, you know, a doctor sitting with the family, a teacher deciding whether a kid is struggling or just quiet. If you optimize those, we don't get augmented humans. If we get thinner versions of the institutions that we're supposed to hold meaning for us. And augmentation done right expands what a person can do while keeping them, keeping them the moral subject of the act. Replacement, when it's dressed up as assistance, quietly moves the moral subject somewhere else. I mean, usually

38:59somewhere no one can find it. That re-location of responsibility more than any technical capability is how I think the line gets crossed, which is not a good thing. All right, all right. I think let's now, let's get into this interesting question. I know it took a while to get into this. The extra AI question, the multi-million dollar or the billion dollar question, I say. I think I've put this question in a different way. I think we did talk about it. We did talk about it briefly. So if AI becomes capable of understanding our emotions, our intent, our cognitive state, and vulnerabilities as well, well enough to influence our choices, where should we draw the line between intelligent assistance and behavioral manipulation? That is one thing, but the main important thing is that who gets to draw that line? Because that is more important, right?

40:04Yeah, I mean, I think one of the things that you have to think about, and it's an aspect, there's a systemic dimension to it, I mean, is correlated behavior. So when a human advisor makes a bad call, it hurts their clients. When an AI agent that's been deployed to millions of people makes a bad call or more precisely when it responds to say a market signal, the way it's training tells it to, everyone, it's advising moves in the same direction at the same instant. And it's not a portfolio of independent decisions anymore. That's a herd with one nervous system. And we've seen tastes of this, right? Flash crashes, Algo trading and cascades. And those who are institutional systems running in known venues with circuit breakers, but what I'm describing is retail. Tens of millions of people, each thinking they have their own personalized advisor, all quietly synchronized because the underlying model is the same and the affective triggers are the same. The illusion of individual

41:08choice, masking what it functionally, what is functionally collective behavior. That's a new kind of systemic risk. And I don't think regulators have a vocabulary for it yet. The second thing that worries me is the dependency part of it. Like if you use one of these agents for three years, you stop practicing active financial reasoning. I mean, you lose the muscle for asking, does this make sense? Is it my interest? What am I actually optimizing for? And that capacity of trophies quietly. And then the day the agent is wrong or captured or misaligned, you don't catch it because you no longer have an independent view to compare it against. There's nothing to check the system with. You just outsource not just the decision, but the ability to notice the decisions was bad. I think this connects to something I keep circling back to. The danger isn't the dramatic failure. It's the drift. Nobody wakes up one morning having

42:09lost their financial autonomy. It just goes on. It goes one convenient tap at a time, one reflexive approval at a time, until the balance in the count is the residue of a thousand small consents. So very dangerous, right? And the danger is quite correlated and cumulative. And I think this is the hardest kind of thing to legislate against. So to specifically a question, who should be, I mean, it has to be the human. I mean, we need to be the one deciding when that augmentation extends into a territory we don't need to or we don't want to. Beautiful. I like the way you've put this across. Let's get it to the, I know, we've covered quite a bit of topics from everything from behavioral AI to human cognition, to these hybrid intelligence and so on and so forth. If there is one thing, the listeners have

43:13to remember from this conversation as they navigate this next era of AI, what should it be in our conversation? If there's one thing to carry out of this conversation, I said the question is not whether AI becomes intelligent. It already is. The question is whether we become wise. And wisdom in this context is not abstract. It's a daily discipline. It's the practice of forming your own view before the system offers one. I mean, it's the choosing, it's choosing of the friction when friction is what grows you. It's remembering that the goal was never to become frictionless. It was to become more human inside a wider cognitive landscape. And I think a lot of people are waiting for a big moment like the AGI announcement, the super intelligence headline, you know, day something wakes up, I guess, like, you know, comes conscious. But while everyone's watching that horizon, the actual transformation is happening in the small stuff. What you notice,

44:17what you believe, what you outsource, what you no longer bother to think about because something else thinks it for you, that's where the future is being decided, not in a lab in your regular, you know, Friday afternoon. So the ask I'd leave people with is almost embarrassingly simple. I mean, stay awake to the coupling. Notice when the system is doing your thinking for you and ask whether you wanted it to sometimes the answers yes. And that's fine. But make a choice and not let it be a default because the hybrid mind can be a place to grow or it can be a place to disappear. And the difference is entirely on the human side of the equation. So the machines obviously will get more capable. That's the easy prediction. The harder, more interesting question is what do we, what do we do with the space that they open up? Do we use it to go deeper into being human or do we use it to be less human, but more efficiently? You know, both futures are on the table and neither is inevitable. You know, I think that's what we need to think about. That's a great piece of advice.

45:23Rana, I think having your own view, it's okay to have to look into the views or the choices provided by the system, but having still having your own view and picking up the right one keeps you augmenting every day and every step as we go. Finally, I think if what should be we be watching closely over the next five to 10 years as human and AI intelligence, human intelligence and AI increasingly get intertwined. What are your, what's your takeaway? What do you think in the next five to 10 years? We should be watching more closely. I know it can be a little bit closer to what you've mentioned already, but I believe with additional thoughts that you have. Yeah, I mean, I think it is the same thing that we've been talking about. I mean, you know, for me, the line is less to do with the technology and more

46:23to do with the shape of the interaction. You know, does the coupling force me to bring something, or does it let me bring nothing? Does it surface contradiction, or does it manufacture coherence? Karl Palper had this idea that progress depends on finding and correcting errors. And a good augmentation partner keeps errors visible in a replacement dressed as augmentation buries them under fluency. So, I mean, I think that's, you know, that's what we need to focus on. I mean, because augmentation leaves a residue in the human replacement leaves a residue only on the screen. Right. Before I let you go, a quick shout out to your book. Could you talk about your book where they can reach out to your book and where they can audience can reach out to you on LinkedIn? Yeah, thank you. The book is called AI Instinct, the future of human decision making, NEI and human decision making. It's with Wiley and Kaifu Lee was generous enough to write

47:28the forward, which still feels a little surreal to say out loud. You can find it soon wherever you can buy books, Amazon, Bonds and nobles, etc. If you want everything in one spot, I mean, the cleanest way to go to is the AI Instinct.com. This information on the book there, some of the ideas we've been talking about and ways to go deeper in any of, if any of this resonated. For me, personally, you know, you can reach out to me on my website, Ron Aguizral.com. And that's where you can see the speaking work and the writing and some of the work that I've been building over the years. But I'm also fairly vocal on LinkedIn, you know, and so happy to talk to you there as well. And I mean, if you pick up the book and don't feel like you have to, you know, you have to read it front to back. I confess the introduction that I'm a chaotic reader myself. But I always, I'm open to engaging and if you have any ideas or thoughts, you know, definitely reach out.

48:34Great. Great having this conversation with you, Rana. I think it's a very interesting topic and very intertwined in, I think there's a lot of intertwining happening between AI and human intelligence. And I think things are going to be more interesting as we move forward. Thanks for your time. Thank you, Rago. A big thank you to Rana Goodraw for joining us and sharing his insights on AI, human behavior and the future of decision making. And to everyone listening, thank you for being part of extra AI. If you enjoyed this conversation, please share it with your network and help us keep these ideas moving. Stay glued to extra AI for more interesting guests, bold questions, and raw conversations on what's next in AI. This is Rago Banda, keep extracting the raw AI conversations. And as always, happy predicting the future with extra AI.