EP 345 – Predictive AI: Does AI Know What You Will Do Before You Do It? – Zohar Bronfman – CEO of Pecan

Read the best-effort transcript below (This technology is still not as good as they say it is…):

Michael Waitze 0:00
Blissing. Okay, let’s do this thing. Hi. This is Michael waitze, and welcome back to the Asia Tech Podcast. We are joined today by Dr Zohar Bronfman, a co-Founder and the CEO of Pecan AI Zohar. Thank you so much for coming on the show. Before we get into the main part of this, what I think is going to be a really interesting conversation, let’s give the listeners a little bit of your background for some context, please

Zohar Bronfman 0:23
with pleasure. So most of my background before starting Pecan is academic. So after my army service here in Israel, everyone service going to the army, mandatory, I went to university, I did actually, quite a lot of studies. Yeah, I did Bachelor’s in Economic Sciences, and then I did a master in philosophy and another master in cognitive psychology. And then I moved on to do a PhD in Tel Aviv’s grad school. I did one PhD in philosophy and history of sciences, wow, and another PhD in computational neuroscience. This is also where I met our co founder and CTO Noam, who did the PhD studies with me.

Michael Waitze 1:16
Awesome. Can I ask you this, if you don’t mind, like, what got you interested in the in the neuroscience, the philosophy of science, and just like all these cognitive studies, was it something that, like your family discussed when you were a kid, that, like your grandmother did, like, Where does this come from?

Zohar Bronfman 1:30
Oh, my God, you know, in all honesty, yeah, I consider myself quite knowledgeable when it comes to popular science of physics and math and and biology, nothing is even remotely as interesting to me when it comes to neuroscience and the mind. And this is the thing, it’s the mind on the one hand, and neuroscience on the other. Obviously, they are very much interconnected, and in all reality, what happened is that I started my master’s in philosophy out of pure intellectual curiosity. I wanted to understand how very smart people thought about knowledge, about the world, about what’s out there. And while doing my philosophy, I realized that I’m looking only at a part of a puzzle. I realized that a big other chunk, an adjacent or a related point of view is the scientific point of view, and that’s where I decided I’m going to understand the cognitive and the new neuroscience things in parallel. And it gave me a very holistic view of the same topic, just from different angles. Yeah,

Michael Waitze 2:57
it’s really interesting. And again, I don’t want to deep dive into this so much, if you don’t mind, but I find this really interesting. To understand philosophy at some level, right? To really understand it like, it’s super interesting, no matter how you look at it, no matter what you look at it from. But then when you start studying the neuroscience of it, like, why do we have these philosophies? Where do they emanate? Why are they there? It must give you a completely different view on just like how the mind is working and what its output is, if that makes any sense,

Zohar Bronfman 3:29
absolutely. Again, depending on how deep we want to go in that direction, go for

Michael Waitze 3:35
a second. Yeah, go for it.

Zohar Bronfman 3:37
I’ll just say that we live in an era when there’s so much knowledge out there, yeah, that it’s not a question of whether you can get to that knowledge like it’s used to be the case. Yeah, it’s a question of which knowledge you decide you want to consume. Yeah, right. It’s how you build your knowledge for yourself. And my personal approach was that, because of the structure of knowledge today, which is extremely vertical, very narrow and very deep, right, you can spend 30 years understanding a sub discipline, there are huge advantages, because people become quite experts in a specific field, and they know like, 100 times more than their peers 50 years ago. Yeah. There’s also a disadvantage of you losing the bigger picture, yeah. And that’s why I wanted to kind of supplement those two rabbit holes

Michael Waitze 4:41
combined. It’s so interesting. I want to do a little more of this, if you don’t mind. When I was in college, I studied Japanese, and the teacher, the the professor, was teaching us language. First of all, they only spoke to us in Japanese in the class, so there was no translation, right? And actually, I learned this in yeshiva, which will probably not surprise you. The. Principle because we were studying the Bible in Hebrew. I don’t know why I’m talking about this, but we were and the teacher would say, don’t translate into English, because there’s no direct one to one translation. And I learned that when I was, like, nine or 10 years old, and I kind of used that for the rest of my life. But the point is, when I was studying Japanese, the professor was also getting a PhD in like, the derivation of words, etymology, right? And he would come into class like on Monday morning, and his hair would be a mess, like it looked like he hadn’t been sleeping for a while. Because when you consider like, where a word comes from in any language, right? And particularly in multiple languages, at some point, it’s got to blow your mind in a way, right? Like, why do we understand what any of these things mean that we’re saying? And I’m curious if you had a similar experience when you were doing this cognitive neuroscience study, particularly in conjunction with the philosophy, Were there times where you were just like, Wow, do you know what I mean? Where you learned something, you figured something out? You’re just like, that’s insane.

Zohar Bronfman 5:58
Yes, yes, absolutely. It’s a you almost bring me the goosebumps when you talk about that, because I think in general, whenever you, like I mentioned earlier, go very, very down the rabbit hole with a specific discipline of knowledge, you’ll find that you are at some point, staring in the eyes of God in a way. And it relates it relates to etymology, it relates to religion, it relates to every scientific and non scientific discipline that we have today. And yeah, I can definitely recall one or two areas for me. I’ll give you just one example, which I think is mind blowing. I don’t know how many people actually think about it in the day to day. But science has proven, historically beyond any reasonable doubt, that human behavior is predictable to a very large degree, in the sense that individuals, individuals come to the awareness of a decision they made after specific events in the brain already took place and they were not aware of them. So in other words, the science community can, today, in theory, build a device that can tell you, Michael, in 200 milliseconds, you’re going to grab the water, but you’re not even aware that you’re going to do it. And it goes for bigger and bigger decisions as well. So realizing that so much is going on in our brain and eventually affects our decisions, and that these things can be deciphered and predicted on has been a moment of our complete humbleness when I first encountered it. Yeah, today’s the rest

Michael Waitze 7:58
of today is going to be amazing, because I’m not going to be able to be able to stop thinking about it. But this is a really good segue into the other stuff that I want to talk about. I want to talk about predictive AI, right? Because this is something that you guys have been working really hard on, and again, I like to presume I understand some of it, but if I don’t, I want to break it down into components, right? So maybe you can separate this out for me into generative AI, something that most people feel like they have an understanding of and predictive analysis and then explain for like a layperson, how the power of combining these two things together actually work in a business context. Is that fair?

Zohar Bronfman 8:32
Yeah, absolutely. So obviously, two big topics both have a very long and deep legacy. Yeah, so it’s important to say generative AI is not something new. No, we’ve been doing the scientific community in the AI community. We’ve been doing a generative AI for quite some time now, more than a decade for sure and generative. AI has a very self accelerator name. From that perspective, it helps. It uses artificial intelligence models, specifically neural networks, which is one subclass of AI models. It uses specific types of neural networks to create something that wasn’t part of their initial input. So if you think about it, neural networks take input, for example, think of it as hundreds of 1000s of images of cats and dogs. It’s the first incarnation of neural networks. Neural networks were very good in separating patterns, okay, recognizing and separating patterns between cats and dogs. As an example, they would be very good in saying, this is a dog because of this specific set of features, and this is a cat, right? We do it also with our brain. We look at the picture of. Cat or a dog, and instantly we are able to say whether it’s belongs to one class or another. Generative models took it one step further, and they were able to say, not only that, I now understand, quote, unquote, what is the class of dogs and the class of cats. I can now take some kind of input and I can create a new kind of a cat that has very similar features to the class of cats, but is not part of the samples that I consumed as a machine learning model, right? That’s the generative AI. It’s the ability of a model to generate another sample, basically, of what it’s learned from predictive modeling, which is many times also called machine learning, is a cousin, right? Both are AI technologies, but it does something very, very different. Again, different and similar, depending on your philosophical view, machine learning, predictive models consume data that had some kind of historical pattern, some kind of a historical pattern, and they presume to say something about a future event in the sense of what is the likelihood that this specific event is going to happen in the future. Again, the big difference, the big difference between the two, is the type of inputs that they are dealing with, Okay, why neural networks and generative models are very much focused on what is sometimes called unstructured data. It can be images, it can be audio, it can be text. These are the areas that the neural networks focus on. The classic machine learning prediction models focus on numerical data, transactional data, tabular data, structured data, these types of data are very much abundant in the business realm. Think of any business you can imagine that business gathers data like, when did the user log into the website? What was their action? When did it took place? When did they make competitors? When How big was the purchase, the amount? All of those questions, these are numerical questions, and they have their own very unique structure based on the business. The interesting thing, and I’m coming to an end, in fusing those two. The interesting thing is that predictive modeling, machine learning, predictive modeling has a very, very long historical business application. Think about it, if you’re predicting the future from an event, from a business event perspective, you are making a dramatic change to your business. The first areas when it started was using it was during the 50s and the 60s when financial algorithm companies, like Renaissance, if you’re familiar, started using these models to predict how derivatives and stocks are going to perform. Right? That’s how long back these AI methods go. The generative aspects are still looking for their business application, because the generative AI capabilities are very good in interacting with the world and with creating, generating specific texts, videos, audio and so forth, as we see from open AI and others. The interesting thing is that they were never merged. It was never a case where you used those land large language models, those generative AI models in the business context, to provide predictions. That has been something that technology hasn’t allowed yet, and that is now becoming something that can be transformative organizations. So you can think of a generative interface, like a chat, like experience, like you’re talking to the machine, but being able to ask questions about the future business outcome, which is a completely ml machine learning based AI. So can

Michael Waitze 14:43
I ask you this? Because one of the things you said earlier, when we were talking about cognitive neuroscience, you said, it’s a given in the scientific community that, given the right information, a scientist can predict when I’m going to drink a bottle of water, right? And in your mind, while it’s amazing. Thing, the fact that we know this now makes it kind of unremarkable, right? But then how does that translate into business and this combination of predictive analytics and generative AI? And then think we can talk about Renaissance, right? Because I come out of a program trading background we ran, and one of the questions I had was like, algorithmic trading, is the history of this is long? We’ve been doing this for 20 years, and now almost 30 years, on portfolio trading desks and on equity trading desks, right? Is this combination of predictive AI and generative AI going to make that trading so much more sophisticated and so much better? It’s just one use case of it because of the combination of those two things, all we were doing in the past was saying, here’s a whole bunch of data that’s verified and clean given certain circumstances, which we can then map onto this data that we have from the past right back testing. How do we then predict what’s going to happen right now or in a moment from now, based on all the information that we know is happening at the moment as well, and this is what Renaissance did at scale, right? And really, really quickly, with huge throughput, and then make trading decisions. And then the other question is, how does it apply? Right? Because you talked about this like, what’s the business application for a company like an E commerce company that’s just doing sales, right, and that wants to sell dog clothing or cat clothing? Where do all these things come together? Yes,

Zohar Bronfman 16:22
yes. So I, I’ll start with your permission with the second question. I think that if you’re let’s imagine you’re an ecom company, right? You’re not a Amazon, because they are obviously huge and they have their own predictive and machine learning capabilities. By the way, it’s a huge part of what made them Amazon, right if you think about in general, if you think about what made the large, huge, successful companies, large, huge and successful, it usually boils down to sophisticated machine learning in the right place in the right time. So if you’re an E commerce company, and you have your decent volume of activity, and you have your business processes, right, you have your promotions, you have your email marketing campaigns, you have your next best offer recommendation, your retention efforts, you know, eventually there are classic pillars of customer acquisition and engagement and retention efforts that you are engaged with. Those usually in a standard e com company. Those usually are not informed by any type of a prediction. You’re basically using your basic data analysis and your gut feeling and saying, We should give a promotion of whatever, 10% to every customer who hasn’t made the purchase in the last two weeks, right? That’s, that’s a great business logic. It’s a great business rule, but it’s like saying Michael is going to drink in this in this episode, is going to drink water in the next 15 minutes whenever he reaches his arm. It’s a fair prediction. We as humans, we make predictions all the time, all the time, but imagine I knew far in advance that you’re going to drink at a specific point in time, or that I know that a specific customer is going to churn, or what’s their lifetime value, or what is the promotion I should give them. If I knew it by making predictions the KPIs, the business optimization would have looked completely different,

Michael Waitze 18:33
yeah, and that’s what, sorry, go ahead. Yeah,

Zohar Bronfman 18:37
no, I’m just saying. That’s what Gen AI fused with machine learning, provides those classic e com and other companies, it provides them with the ability to operate an interface that is Gen AI like so it’s like talking to a virtual agent, but that virtual agent is not just creating text or Code. That virtual agent is actually your virtual predictive agent. It tells you what’s going to happen and helps you run the business and optimize your main processes. So can I

Michael Waitze 19:12
ask you this one of the things that you wrote, and I did a little bit of reading on some of the stuff that you’ve written. So forgive me if this sounds like I’m just reading back to you something that you know for sure. I I’m going to quote it right, even when maximizing decision accuracy requires weighting all the evidence equally, decision makers often give stronger weight to evidence occurring early or late in the evidence stream, right? Can you tell me, just from the research that you’ve done, why that’s the case? And then maybe talk about business examples where that was the case, and how AI kind of solves that problem. I think it’s a persistence thing, but I’m curious how you, how you view this? Yeah,

Zohar Bronfman 19:48
it’s, it’s such an amazing anecdote, because it’s, it’s a, it’s a window into the amount of biases that we as humans have. Have ingrained in us, right? So it’s not that we make a conscious decision to underweight all the information we have during the regular process of our deliberation, right? It’s just our brain is built with a way that whenever you have something at the beginning, like, if you consider like a stream of information that you are considering as a decision maker in your business, you would pay a ton of attention to the beginning, right? There’s a very strong primacy effect. You will remember very well what happened at the beginning. And there’s a very strong recency effect. So think of it a little bit as a U shaped weighting function. You weight the primary and the reasons of the evidence the most. Now there are good reasons. If you think about the nature, if you think about evolution, if you think about different considerations, there are good reasons to do it. It’s very hard to be completely unbiased and take everything into consideration the same way. But obviously, in the business realm, you can’t afford underweighting all the good stuff that comes in between in the middle. You want to understand what’s happening in your business throughout the journey of your deliberation, if, if you are going to consider the data, you want to consider it in a completely unbiased manner. Now, it’s extremely hard, and this is one thing that I really wanted to underline. It’s extremely hard for the human mind to make a pattern recognition on numerical data, really on business data, unlike images and texts and again, these are there are evolutionary reasons for that. Tell me. Tell me. When you see an image, you understand the image immediately. When you hear something, you understand what you hear immediately. Our senses, our cognition and the entire sensory apparatus are all tuned for text, for images, for audio. We don’t handle business data, number data. We don’t handle it well, and that’s where the machine learning actually comes in play. It compensates for our innate inability to analyze numerical data, and by relying on machine learning, you can correct some of the things you just don’t see. You’re blind to, and that will surface. That will surface, it will never make a decision for you, you know, what’s best for your business, like any other business owner, but it will just surface considerations that were not there before.

Michael Waitze 22:54
Is there a specific thing that you studied when you were doing all your research and getting your PhD that underlined this idea that that stuff that happens in the middle, like, is there a physical or cognitive or neuroscientific reason why? And you said, maybe an evolutionary reason why humans ignore that stuff. Like, why are we, what is the build inside of us that says, recognize, I look at a dog. I know it’s a dog. I don’t have to think about it kind of thing, right? And when I hear music, I know, like, that’s Billy Joel, I don’t I’m not thinking, Yeah, I’m not analyzing it, just, I just know it. But if I look at a bunch of numbers, it’s just a bunch of lines and circles. Do you know what I mean, except for very few people. So what is it in the evolutionary process, or in the neuroscientific process that makes that the case, like, why weren’t we built that way?

Zohar Bronfman 23:35
Yeah. So the obviously, you know, you put two scientists in a room, there are four opinions, so it’s still debated. Fair enough, I will, I will tell you my personal interpretation and how I think today the scientific community thinks about it. Okay. I think it’s actually not that you are underestimating the quote, unquote, middle I think that we have an overestimation of the initial and the last pieces of evidence. Got it okay. And why is that? I think that, for different reasons, the initial pieces of evidence, they have something with it’s called abruptness, right? The first time you see something, it leaves a greater mark. That’s why, many times people say you don’t get a chance to make to leave a second impression. Yeah, you

Michael Waitze 24:38
only get one time to make a first impression, right? Like it’s that’s exactly what I was thinking. Sorry, go ahead,

Zohar Bronfman 24:42
so that’s why people overweight the beginning of a narrative about evidence now the end, the recent elements have a different consideration, and that’s, I think, effect that comes from when. You run a business and you have to reach a point when you need to make a decision, you would probably assume it might not be the case, but you will probably assume, and it makes sense to assume it in some way, that the things that happened most recently are the things that are most likely to persist, because you don’t really know how things change or whether they change, because, again, you don’t know how to analyze numbers well enough, because we are humans, we’re not machines, yep. So you basically tell to yourself, if I seen something happening yesterday, I will assume it will it will continue pretty much the same today, yeah, whereas maybe there’s seasonality or some kind of a cycle happening and something that happened three weeks ago is actually far more important. You won’t be able to say, right? You need a machine.

Michael Waitze 25:57
This is true for almost all walks of life. Can I make, like, a completely unrelated analogy, when, when Patrick mahomes was playing college football, people looked at him and he was like, throws the ball too much, throws too many interceptions, never gonna win anything, because that’s all they saw. And there were exogenous impacts, like all these other things happening, and now he’s like, the greatest quarterback of all time. And I’m not saying he has arisen, but I’m just saying, like, that’s how the opinion changes, and you’re changes. And you’re right, it’s because of the abruptness, the recency of the data that they have. That’s, that’s, um, that’s changed. I want to ask you this too, though, right? Because if humans don’t know what they don’t know, and they’re not sure, right? This is a sales problem for you, or a sales issue. So when you go in to sell pecan and you say, Okay, we’re going to pay attention to all this stuff. Like, how would you characterize the teams to which you’re pitching, the products that you’re building, in the sense that you may understand as we’ve been talking about, like, the nitty gritty details of all this stuff way better than they do. There must be a learning curve for the institutions that you’re selling to. But also, what drives those aha moments that you’re talking about, that abruptness, where they go, Oh, I get it. We’ve been ignoring that stuff, and now that we know that, and we can see the predictive benefits of having this, then it changes our business. What are those moments? If you can run through some of them from here, at least one of

Zohar Bronfman 27:14
them, yeah. So very early in our journey, we had a ha moment of our own. Okay? We realized that there’s a an audience or a subset of individuals within companies that are perfectly, perfectly built towards starting to use machine learning the way we think companies should start using machine learning. And I’m referring to a role which is usually called analytics, or business analytics, or data analytics, or bi and I want to explain the role a bit, and I want to explain why they are such an amazing audience for using machine learning. So the business intelligence, business analytics, data analytics individuals, are individuals who are working with the business shoulder to shoulder and supporting them in those business processes I mentioned by running human analytics on the data, so they can ask, for example, how many customers visited the website last week, of which, how many converted to our new offering? That’s a human analytics it’s a human individual asking the data a human intelligence question, right? It’s amazing. It was actually a revolution in our business realm, 1520, years ago, not like, you know, a century ago, right? Those individuals know everything you need to know to do machine learning. They know the business they know the data, and they understand the business processes that rely on the human analytics. They know that as humans, you can’t do a sophisticated analysis that relies on more than two or three data points. It’s just beyond our cognitive capabilities. We can analyze two or three points after that, we are lost in the woods. So those individuals are quote, unquote, dying to put their hands on something that can augment their human analytic intelligence with AI analysis, right? So they can then take the data and ask a question that doesn’t inject their own intuition into it, but rather lets the machine provide the predictions that they believe the business requires. Now the other nice thing about this community is that although it’s not as experienced and skilled in. Specifically machine learning and AI, like the data scientists, which are the other group of data individuals in other areas of the business, they are completely relevant for the world of AI. They only need to have the tools and the capability built into their organization. So that’s the personal revolution we are taking.

Michael Waitze 30:23
Yeah? So when you walk into these companies and you find these people like you said, that are tangential to the business, right? They’re not doing sales, but they’re providing the analytics to the sales teams. And you find them and you say, how about this? If you use this, that’s their aha moment. They’re like, Uh oh, this could, like, really change the way we do this other thing that I’m closely related to, but I’m not doing myself. Yeah. Does that make sense

Zohar Bronfman 30:45
100%

Michael Waitze 30:47
but what are the risks, right of getting to those people, giving them the tools and like, what if they fall short? Do you know what I mean? And then what is the response? Because this is the problem, right? Because there’s all this magic around artificial intelligence. And maybe those guys and gals that are doing these biz dev roles, or doing the data analytics roles or business analytics roles, right? They’re thinking, if I had these tools, I could be way better at this. And then there’s this human inclination to kind of go, wait until you see this new report that I have kind of thing, you know what I’m saying, though, right? And if it doesn’t generate, this gets all the way back to the beginning conversation we’re having. Right if it doesn’t generate like new business or live up to these new KPIs, what are the risks there? And then people just going, like, we’re not going to do that kind of thing. Do you know what I mean? And how do you disintermediate that, or risk mitigate that? Yes,

Zohar Bronfman 31:38
the flip side on having this great power of AI predictions under your fingertips is that you lose a bit of control because the clarity of why exactly the predictive model made the prediction it made is murkier than our own intelligence. So again, this comes back to our cognitive structure as humans. When we say, give 10% promotion to every user who didn’t visit the website for two weeks, we understand exactly why that user was given a promotion. Yeah,

Michael Waitze 32:19
and there’s an expected result too. You didn’t do this for two months or whatever. Here’s a 10% discount on something you’ve bought before. You’re probably going to buy it again, kind of things. It’s very clear, yeah, sorry, go ahead.

Zohar Bronfman 32:29
It’s clear. It’s an it’s reportable, it’s understandable, and it’s very hard to tell that data analyst who made that rule? What in God’s name, have you been thinking whenever we say, okay, give a promotion only to people that the Peking predictive model said you should give a promotion to? There’s an act of almost like losing control, yeah, almost like a feel a fear of the unknown, yeah, and that’s the real thing. Organizations don’t adopt AI in many cases, because they don’t have enough understanding of what happens under the hood, and for that purpose, we’ve invested as a company and as a community a lot of calories into providing an overview and a view in general into what made the machine make Its decision. But there’s one additional thing that we have to take into consideration, and it is that, again, these are like, psychological biases, right? Yeah, go for it, because when you do marketing right, and when you pay Google and Facebook and LinkedIn a ton of money, and they bring you back specific leads. You don’t really know what happened under the hood. You just trust them blindly. What you’ve done as a marketer, if you practice in good marketing exercises, you’ve conducted AB tests because you became results driven rather than process driven? Yeah, I

Michael Waitze 34:27
was gonna say, you

Zohar Bronfman 34:28
think about the results and you say, if LinkedIn bought me these leads and I invested that amount of money, fine, fine, because they use their own AI. Now the same thing applies here, we always recommend our partners and customers, and we work with them on implementing it. If you’re going to reuse an AI, a Gen, AI, an ML, or any blend of these models into augmenting some of your business processes, you have to a embrace the fact that you won’t have. Full clarity, because you’re no longer in the realm of human intelligence, right? You’re in the realm of artificial intelligence. Embrace it. Two. Always be testing, always be testing. Right? You can always divide, you can always measure, and you can always see with your eyeball the big difference between relying on human intelligence and relying on AI, yeah, so

Michael Waitze 35:26
this brings up another topic that I want to discuss with you, and that’s something also that you’ve written about and done research on, that’s transparency, right? And this gets into this kind of murky little world of I trust what’s happening, but I want a little bit more information on what’s going on in the back end. And we used to do this on the trading desk. The trading desk too, right? Like, we had VWAP algorithms. We had best price we had all these different algos, and at some point, the buy side traders would ask us, what is the algo doing? Right? Like, why did it do that? Because it didn’t seem to make a lot of sense. And to be fair, I don’t know if you know what a VWAP algorithm is. It’s one of the most simplest algorithms. Simplest algorithms, right? Value weighted average price, I think, or volume weighted average price, yeah, so I haven’t done it in a while, but it’s volume weighted right? So if a ton of volume trades, the V wop algorithm is going to try to catch up. It’s not going to make a decision, at least in its in simplistic terms. It’s just going to say, like, I just saw a ton of volume trade, I don’t want to be behind volume, right? But over time, as you develop these algorithms, you can think maybe that’s an anomaly, maybe it’s all going to come back. Maybe that pushed the price up too high, it’s going to come back down. If I wait I get better prices, right? Which a human trader can make that decision. But Can an algo do it? But at least we understood what was happening. Do you want to talk at all about this transparency?

Zohar Bronfman 36:39
Yes, I do. And, you know, it’s so funny that eventually these issues come back to my academic philosophical background. Because eventually, when you boil it down, it comes to the fact that people are also, you know, a cognitive animal. We have our own way of thinking about things and analyzing things, and it actually is a gap between how we think about things and how machines think about things. Yeah, and the thing is, you know, there is a ton of things that can be done from a transparency point of view, and I’ll get to it in a minute, but there is a fundamental thing that we have to put on the table and be honest about. We will never be able to understand the machine learning and AI algorithms the way we understand our own decisions. Because the reason is these are two different levels of explanation. When you want to explain something for human, you use something that we call in philosophy, intention. You try to explain their intention, their goal driven behavior, the consideration logic and unwell considerations they take. He reached out to the bottle because he was thirsty. But when a machine tries to explain the its own decisions, it would be something like this group of neurons fired at this rate, that group of neurons fired at that rate, therefore the arm is going to be reached the bottle of water. It’s a mechanistic explanation versus an InsurTech explanation, and the same logic applies one to one with the business understanding, if you want to give a promotion to someone that hasn’t visited the website for two weeks, it’s an intention. You have an intention to help those who forgot to log in, right? If you want to give a promotion to someone because the machine learning algorithm wants to give it to them, then it means that something from the geo that he logged in, or she logged in, from something that he made a week ago in the website, some other elements that resemble and all of those are being multiplied by a merk number that no one can really Explain. All of that combine provides you with a mechanistic explanation of the likelihood to convert given a promotion. So when you think about transparency, as an individual who is considering adoption of AI, you have to put your intentional stance aside, yeah? Because you have to flip the switch and be completely comfortable with using a mechanistic explanation, which is what the machine learning model is going to provide you with. Yeah, because at some

Michael Waitze 39:57
level, even if you listen to the explanation, even if. It’s completely transparent. You may not understand it, and I think in some cases, most people would never understand it, right? Because it’s mechanistic as opposed to intentional, which is the way you live your life normally, right? Like, if I said to you, why did you do that? You could say, I’m thirsty. I completely I don’t need any more explanation. I completely understand it. But if I asked the machine why you did it, and it gave all these other numerical explanations, I’d just be like, Fine, let’s just do some more sales, because it doesn’t make any sense to a human who’s not thinking that way, and frankly, never will think that way. Is that fair?

Zohar Bronfman 40:30
Completely fair. Okay, I

Michael Waitze 40:31
want to ask you this too. I was having a conversation, and this is more just like an opinion of yours. I was having a conversation with somebody this morning about artificial intelligence as well. We were trying to make an equivalency to something more on the persistent side, like the fact that it’s just going to be everywhere, right? As opposed to being a functionality in and of itself. It’s just going to be part of everything that we do it. And the thing that he came up with was electricity. And I was thinking about this after we hung up the call. It’s like, I have no idea how those lights work. And if I’ve been thinking about this for a long time, like 20 or 30 years, if everything that we have in the universe just somehow went away, whether it’s this table or this microphone or this computer screen, just went away, and I had to, like, rebuild it, not possible, because I don’t know how any of this stuff works. Right? There’s a camera over there. I have no idea how it autofocus is. I can guess, but I could never reproduce it. But even something as simple, which is a relative term, like electricity, I flip a switch over there, the light goes on. I don’t know how any of that works, either, and I don’t think most people do. Are we going to get to the point where the AI stuff that we use is so persistent that it’s going to be everywhere as well, and in a lot of cases, most people are not going to have any idea, one, that it’s there, or two, how it works. Yeah,

Zohar Bronfman 41:49
this is a very nice way of thinking about it, which definitely resonates. I would like immediately agree I think a as humans, it takes us time as a complete civilization, it takes such time to adopt and adapt to groundbreaking technologies. Every new technology and maybe chewed is being greeted with healthy cynicism, with healthy concern, and it’s fine. It’s completely fine. But eventually life transforming, groundbreaking technologies are always being adopted, and we adapt as a species. It’s actually one of our core evolutionary strengths, specialization, basically what you’re describing is specialization, right? I have no idea how to make a screen, and my colleague who makes screens has no idea how to make podcasts or knowledge, or whatever your profession is. The fact that we as humans were able to specialize and have each and every individual become those deep verticalization of knowledge like I mentioned earlier, is only going to increase more and more and those who master AI and can help us as a civilization to enjoy from it are obviously going to be extremely important to us as civilization, as others. It will take a little bit of time. It won’t take a ton of time because the benefit of adopting is so high. It will take a little bit of time until regulation is better, until the common education is better. And then once it’s going to be as abundant as you’re going to refer to it, it’s just going to be part of our day to day. By the way, I should say, AI has been around us for a very long time, and we are using it, even if we don’t realize that we use it exactly. Whenever we buy on Amazon, we get recommendations that are based on AI. Whenever we talk to Siri, we use full blown AI algorithms.

Michael Waitze 44:10
Yeah, Gmail, so like all this stuff, yeah, it’s already there. You can’t

Zohar Bronfman 44:14
escape you can’t escape it as long as you’re part of civilization and as long as you’re part of technology, AI is just going to be more and more around us. We do have to be aware. We do have to exert consideration common sense and regulation just to make sure that the technology is being used in the best way possible, because there are always people who are trying to make you know the technology serve their own agenda and a debt agenda might be bad sometimes, in all honesty, I want

Michael Waitze 44:43
to switch gears for a little bit and just talk about something at the end of this conversation, if you don’t mind. And again, I was listening to somebody talk about something this morning, and they were talking about the Lincoln assassination. And there’s this famous quote, I can’t remember it exactly, and it goes something like, you know. Except for the thing that happened up in the thing like, how was the play mrs. Lincoln? And I kind of feel like, I kind of feel like that’s what it’s like to be an Israeli entrepreneur, particularly for today, particularly today, but today, really, in my mind, is no different than it’s been for the last, like, 75 years. And I’m curious, like, if it builds a certain kind of resilient person who is, like, fundamentally better at building things from scratch and able to put up with like, all the adversity that goes with building I was talking to a buddy of mine this morning, who’s my business partner. We signed a deal today, and it was like, literally, touch and go. And I don’t share this a lot, but we signed a deal today employ. It was really cool, and we weren’t 100% sure it was gonna happen, and if it did happen, it’s amazing. If it didn’t happen, we were just gonna have to regroup and figure out, like, Are we okay, right? But that I feel like, is the day to day life outside of building a startup of an Israeli entrepreneur. And I’m wondering if you just feel free commenting on that a little bit so people can get a better sense of what it’s like day to day, not from a political perspective per se, but just from a personal perspective. If that’s okay,

Zohar Bronfman 46:07
yeah, sure, I would say it’s a obviously I’m not happy that this is a topic I wish,

Michael Waitze 46:14
I wish, but it is right. It

Zohar Bronfman 46:17
is, it is, you can’t escape it, especially not nowadays. Yeah. I mean, look, operating as an Israeli from Israel has its challenges and advantages. I can talk about the part of the advantages first, because I do think that when we, when we, as Israelis, have to face with the stress that we are facing as as citizens, right, you are surrounded by people that explicitly state that you want they want you dead, Yeah, that’s, that’s the situation, I think for most people out there in North America, South America, Asia, Europe, it’s not something they can even imagine, right? That everyone around you explicitly states that they want you dead. I think that forges you in many ways. It makes you very resilient. It makes you understand that nothing can be taken for granted, that you have to succeed, because it’s only up to you, right? You can’t relax, you can’t be laid back, you can’t be calm, you can’t be you can be resting. Basically, we don’t let ourselves take the foot off the gas pedal, if that makes sense. So that’s one thing you know, whenever there’s a big challenge at work, it also helps, it also helps you keep things in perspective. Yeah, you know, we live again as an Israeli, I’m sure our listeners are aware it’s a tough time for us here. There are threats, real threats, that happen all the time. You know, work is important, extremely important. We give our day to day to work, but we know what’s really important. And as long as we are touch wood, healthy and together, we also it’s a very grounding, right? It makes things it puts things into perspective. Last but not least, one big other advantage that I can share as an Israeli entrepreneur is that it takes the role of leading a company and building a great company and contributing to the Israeli economy. It takes it to one additional extra level of almost a patriotic level. I want, I want to help the country, not only, quote, unquote, my company and our employees, I wanna, I wanna, I wanna be successful for the country’s sake, for the state’s sake, as well. But there are also challenges. It’s obviously very hard. It’s hard operationally, because it’s very hard to operate when there’s a war going on, as you can imagine, yeah, and it’s also hard to, you know, keep your focus on what counts from a business perspective, when there are so many distractions unfortunately going around. But as I always say we are tough, we are strong, we are just, and we have to prevail. We will prevail. And that is encouraging for myself, for the team here, who has been amazing, and for all my colleagues and peers that are running amazing companies out of the Israeli ecosystem. So. Okay,

Michael Waitze 50:00
I went in on that. That’s really, really powerful. Dr Zohar Bronfman, a co-Founder and CEO at Pecan AI, that was amazing. And again, thank you so much for sharing. At the end, it’s, I think it’s really important for people to understand and to humanize these things so they get a real sense of what’s happening. Thank you again, so much for your time.

Zohar Bronfman 50:17
Thank you, Mike.