EP 278 – Paul Meinshausen – CEO at Aampe – We’re Against a Backdrop of Very Rapid Change

by | May 28, 2023

The Asia Tech Podcast reached out to Paul Meinshausen again after seeing an interesting post he made on LinkedIn.

Some of the topics that Paul discussed:

Paul is one of the most thoughtful people I know. Please read the article referenced here and then listen to this episode…

The title of the Michael Lewis book I could not remember is “Liar’s Poker”.

Paul references this blog post by Andrej Karpathy…read this as well.

It would not hurt to know a bit more about logistic regression.

I agree with Paul’s assessment that we are moving from a deterministic world to a probabilistic world.

Some other titles we considered for this episode:

  1. The Pace Is Extremely Fast
  2. The Explore-Exploit Tradeoff
  3. The World Is Dynamic and Chaotic
  4. God Does Not Play Dice
  5. We All Have the Right Degree of Change That We Can Absorb at a Given Time
  6. Everything Is New for a Three-Year-Old
Read the best-effort transcript

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

Michael Waitze 0:05
Hi, this is Michael Waitze. And welcome back to the Asia Tech Podcast. Paul Meinshausen the CEO of Aampe is back on the show. Thank you so much for coming back and doing this. I saw your post on LinkedIn. And I just had to have a conversation about it because there’s so much depth here. And I just kind of wanted to learn more. How was your trip, by the way?

Paul Meinshausen 0:26
It was great. It was a it was really fun. I have it’s actually I’ve been this year been traveling, again, kind of back on the normal schedule in Berlin and India and Indonesia. But I hadn’t I hadn’t gotten back to the US this year yet. So this was my first time. And I was just in San Francisco for a couple of days. And then over in Las Vegas for this conference.

Michael Waitze 0:47
Do you find these conferences useful for you?

Paul Meinshausen 0:50
Yes, as someone who runs a fully remote, we are globally distributed, fully remote company. I’m all in on on distance and what you can achieve through distance and through phone calls and conversations like this. Yeah. And you can grow that way. I’ve invested in many companies that work that way. But I also recognize what in person contexts and conversations can do. And I think the thing about conferences that I think is very useful is the density of conversations in a very short period of space, as well as the the kind of accidental opportunistic exposure, that you get two different things that are happening within a community. And, and I’ll say this, as well as like, I never found them. I’ve been speaking at conferences for 10 years. But as a as like a as an operator. In a consumer business. For example, let’s say a b2c business where I spent most of my career, I didn’t find conferences, so, so so, so much use from that perspective, because if you go to listen, it sort of to talks, you can really hear those talks on YouTube, right? So it’s not or anywhere else. And then your work is so contextualized, to your business to your experience, that a lot of times, you have to translate things to your context for them to be useful. And that’s a lot of work. So sometimes it’s not worth it. I think it could be great, but I wasn’t like that super bullish on it in that context. But when you’re building technology for businesses, or where you’re going to a conference where it’s your customer, I can imagine if you’re building games, you can go to gaming conference. That’s where it where the I find the value to be just extraordinary, because then you’re just sort of seeing so many people that might be relevant to your product in a, you know, very high frequency and high intensity interaction. So

Michael Waitze 2:42
I’m trying to figure out how to redo a conference where there are no speeches. Yes. Do you know what I mean? Where there are no presentations, because nobody wants to be in them. Even the people that are speaking on them are nervous, right? They’re not used to it. That’s not their job. But being there, like you said, with 1000 people, I don’t think 10,000 Works for me. But being in a room with 1000 people where you can mill around the serendipity is unbelievable. Somebody overhears you talking about something and says, I got to talk to Paul, like that works. But the speeches, nobody cares, do they?

Paul Meinshausen 3:14
Yeah, I mean, I’ve heard them and I’ve seen I’ve been in a talk or whatever. Once in a while that I have enjoyed, sometimes it catalyzes conversations around it, if it’s really good. It’s just you have to have a pretty high bar. And so I tend to appreciate conferences where the curation is, is very high, right? Where it’s really, they’re very focused on high quality context, I’ll tell you the conferences where they kind of gives it a bad name is where the conference lets vendors just bring in a bunch of customers. And the vendors are not doing that great curation, even though it’s in their interest. But somehow it just they don’t have the skill set or the attention to detail. But then so they just bring in people who will do case studies and talk about their thing. And that takes up, you know, 80% of the talk slots. That’s not very useful. Yeah, that’s usually good. And again, the person giving the talk may have never done it before. So they’re not they haven’t learned how to construct or architect the talk very well. Yeah. Yeah. Anyway,

Michael Waitze 4:09
I’m just selling how to re architect this. Yeah, I do think it’s useful to get people together. There’s definitely a place for it. But I haven’t figured it out yet. Anyway. So you were at this thing in Vegas called M. Au Vegas, like what was that? And what was the big takeaway?

Paul Meinshausen 4:22
Yeah, so the the conference itself is called M au Vegas. It’s is the MEU stands for mobile apps unlocked. And it is, I would say maybe the largest specific to the world of mobile apps specific to mobile app growth. Now you have, obviously like the Apple developer, iOS conference, and things like that, which are massive, but from just a community perspective of here’s a bunch of independent or any kind of apps from a Strava or Nike or a Walmart to indie meditation apps or game apps or education apps. So you really have a nice span, but it’s really folks who are really specifically interested in growing their app in terms of usage, engagement, downloads all of those things.

Michael Waitze 5:08
Right. So that’s kind of super useful for.

Paul Meinshausen 5:11
Yeah, so those are our primary customer set. But I think also, and we’ll, we’ll see how I think there are some some of the, the changes I talked about in the article are relevant for just the future of apps, because and the nature of apps are, it’s the application that are downloaded through a particular place called an app store. Right? That’s really interesting. Because if you think I mean, if you’ve been around for more than 10 years, just 20, I still remember buying software in a box store, same, right, like, so that’s that. So then you buy up software, that way, I remember starting to download software downloaded onto computer and then you have to worry about re downloading etc. And then we have browser based applications and and for the phone, which is a different, you know, it’s a different interface. It’s a different hardware than a computer, we interact with it differently. So the primary interface, the primary software layer, is the operating system is either a browser, which is often called like a native app. Sorry, it’s a browser. And then, and then an application, which is an app that sits on your homescreen. And, and you can get it in various ways you can download it through a file. But the vast majority of apps that you have on your phone are coming through either the Google Play Store, which owns Android and maintains that community or the iOS app, and there’s some some others, even in Asia and China, etc. But those are the two that are really the biggest and and there’s a lot of there’s a lot of effects for that structure, the degree to which Apple controls those apps, for example. And so these are often topics of conversation, this is an issue. So I think yeah, it’s this conference just sort of gives you a sense of how people are thinking, not necessarily in the engineering, this, these are not folks who are really thinking about how apps are engineered. They’re really thinking about how, how they grow with their users and their with their customers.

Michael Waitze 7:14
So I didn’t anticipate doing this at all. But what’s your view on the App Store structure? Right, and you don’t have to make a commentary on Apple or Google itself. But just the structure of this of like, that’s the only real window into getting something onto your iOS devices has to be in the App Store, you got to pay Apple, some percentage of the revenue that gets generated there. Is there even a better way to do this for most people, right? Because for guys like you and guys like I am, I could side load software if I wanted to, but like your mom’s not going to sideload software, right? Is there a better way to do this?

Paul Meinshausen 7:46
It’s not, it’s not my area of expertise. So I don’t know how useful I would say something there. But I would what I do know is that the challenge is with full decentralization, let’s say a completely free market, right? In technology and the software senses that it becomes you have to rely on social and kind of open collaborative mechanisms to guardrail and sort of protect the overall experience base. Yeah, right. And, and that’s, that’s an interesting, as you know, I kind of sort of, I think there’s some attractive things about that. It is a big belief to take on to be able to do that. And also just like a practical thing. And so having really large entities like Apple, and Google, which can devote a lot of attention to trying to curate those. And this is a class that this is not just true for apps, right? This is true for Twitter and social media as well, these are and why people as they move away in and out of Twitter over the last six or 12 months since that actually, you know, ownership thing happened, you know, people often would go to more decentralized versions of Twitter, because there was a top down rule. And then you start to realize, whoa, wait a second, you know, there are millions of people around the world that are, you know, loading things like pornography, or just black hacking all kinds of pernicious things that you wouldn’t want in your space. And you realize, oh, okay, now, when we run it, we have to keep it clean. And yeah, that’s a pretty overwhelming task to take on. So I think, but I think the question is not, you can a lot of times when I hear these kinds of conversations, or you know, the way we’re talking about it now is really, from an organization perspective. And also from a normative perspective, should it be this way, could it be better? And here’s the thing is the thing that’s different about the last three to six months, is that the conversation about whether there should be an app store not right now is not normative. It’s its capability. derivate if there’s a fundamental break, and change in the way we build technology that will drive change and appstore the way no political kind of normative philosophical arguments could ever do, you know, like, the government needs to step in and change this, or we need to change new, you know, the new monopolistic practices, laws, all of that stuff, is the way most people normally talk about these things. But when a fundamental change, and technology comes and breaks all of this, it won’t matter, it will change because it has to, because it’s so obvious that an alternative way is better. And that’s kind of like the interest the conversation I’m the the framework or perspective, I’m sorry, I’m more interested in just having. Yeah, political debates just go round and round and round and round. They don’t interest me as much. But this is this, I think, is different. It’s actually

Michael Waitze 10:58
again, if you listen to a lot of the stuff that we do here, I’m not interested in political debates, either, because you can just be on one side of the fence or the other side of the fence. And we can just have these almost ad hominem arguments around like why this is right or wrong. But the real thing is, yeah, what’s going to come along and break it? Because that is way more interesting to me. Way more? Yeah. Yeah, that’s right. I think. And in a way, that’s kind of what you were writing about? No?

Paul Meinshausen 11:21
Yeah. So yeah, so what I was, I mean, I kind of wrote this post and around 15 or 20 minutes sitting at the Las Vegas airport, about to fly back to Singapore, where I am now, where I were, I was my homepage, home base. And I think I was just what the conference had been driving in me was just an awareness of actually now stepping back from the technology for a minute and thinking about the organizational and business implications of it. So I was just right before the conference, I was two days and NSF. And I was meeting with really kind of people building in knee deep and elbow deep in, in building AI applications. Really cool companies. There’s a company called relevance that AI and assuming, but Jackie CO, one of the founders of that company, and they’re, they’re enabling people to work on top of large language models in this really cool way. This sort of concept of a length chain, which is just sort of enabling natural language programming, to achieve different objectives to kind of create what are called agents to do these kinds of things. And there, the awareness is high, the opportunity is high, the interest is high, and the pace is extremely fast. And that’s the, that’s the world I normally tend to live in, if I’m not talking specifically with customers about how amp works. And then I was at this conference for two days. And what I saw there was kind of what I’ve seen over the last several years, it’s the same templates, the same rule books, the same technology vendors with their booths, and, and a lot of action, in terms of energy, a lot of people talking about what’s worked for them, and what doesn’t work for them. And just very low awareness of how quickly the underlying toolset and infrastructure is going to change. And how much of an opportunity that is if you can, if you can move on it quickly, and how much of a risk it is if you can, and it’s not a risk for everybody, because there’ll be legal and other issues, or maybe, let’s say operational issues that will slow the pace of change if your business is somehow molded from technological change, because there’s like real world hardware type things or other things that we’re kind of AI is not imminently changing yet, because you know, there’s just like practical barriers to it, then you’re gonna be a little bit more, okay. But a lot of the companies are in applications that, that were at this particular conference are not protected from that. Think gaming apps, think education apps, think, any kind of experience experiential software, that’s going to change and it’s just, it’s kind of amazing how fast it’s going to change.

Michael Waitze 14:08
It surprises me though, that at a tech conference, at any level, that people wouldn’t be sitting around talking about this. I mean, I was at a conference in Singapore called I can’t remember anymore in the middle of April. And it was a marketing or a PR conference right? And they were talking about how artificial intelligence was potentially going to put all of them out of jobs right this is wasn’t a gigantic conference like three 400 people. But every presentation in every conversation that was there either onstage or offstage was around how all this stuff was going to impact them. Did it surprise you that that was more

Paul Meinshausen 14:41
or well let me let me calibrate and, and be a bit more specific about what I mean by not not realizing or acting on it. It’s not that people weren’t saying the words AI or that AI is gonna put us out of business where AI is gonna change thing. I heard plenty. I had plenty of that. But Think about it this way, it was sort of bookended to the front and the end of the talk. It was kind of in the air and people were worried about it. There were not clear instances of operational actionable, like, let’s we’re trying this, we’re trying this, this is what we tried. And it’s working or not working. It was, it was separated, it was more at the level of fear or worry, or progress, like prediction, not tangible, actionable. And that’s what this conference was really meant to be. It’s meant to be an actionable operator, kind of let’s do things. And that’s what I mean by so yeah, there’s lots of talk, it’s just that that talk doesn’t really go much beyond Oh, AI is coming in AI is going to change things. And, and I tried to get it and you know, get a bit more specific in my post where it’s kind of like this is how it’s specifically going to change and disrupt these processes.

Michael Waitze 15:50
I want to get to that exactly in one second. But first, I want to do a little bit of equivalency to see if it makes sense. Just to put it into a different context. You know, when I used to sit on a portfolio trading desk or on a stock trading desk, we spent a lot of time hiring. You know, most of the traders that were there were just like guys that knew how the market worked. Right? That was it. And what happened over time is those guys that knew how the market works, started hiring people in the IT departments back then who had PhDs in computer science and understand how to build algorithmic trading. Yeah. And then what we just said was, screw it, you know what, those are the people that are now just going to sit on the trading desk, because the traders themselves were talking about the tech, but they didn’t know how to really employ it and use it. And then we just removed that layer. And I think that’s kind of what you’re suggesting is like that’s going to happen in this business, too, is that you’re going to move from people talking about AI and implementing it as a tool to just becoming the basis for like how all of this stuff works. I feel like it’s kind of the same in a way. No,

Paul Meinshausen 16:49
that’s a great analogy. Actually, I we use it to describe some of how amp works actually, many times that it’s more like algorithmic trading. But I think trading is a great example of it. And look, we’re talking against the backdrop of very rapid change. Yeah, just at a different level, like in other words, most, we’re just talking about App apps. Apps are barely over 10 years old, 13 years old, they the Apple App Store was launched in 2008, July of 2008. Okay, so like, and then and then that world has liked the way we build those apps has changed the way we build software in general, the languages that people are writing code in is different than it was 10 or 15, or 20 years ago, and definitely 30 and 40 years ago. So we’re talking about a general pace of change that is historically different. So that’s true. And we have instances where we’ve also seen this Yeah, so algorithmic trading, I think the financial markets is a great example of this. I mean, financial markets are massive, they make the world go round. And I think I saw a statistic It was sort of like in 1999, or just in one or something. The amount of machine driven trades, algorithmic trades was, I don’t know, like, maybe 5% or something. Now, I think it’s 7075 80% is algorithmic is machine driven. It’s run by models. It’s not, and it’s high frequency, and it’s, it’s co located so and it’s completely changed. You watch some movie like Wall Street, which is in the 80s. It’s quiet, or even or what’s the famous guy’s book? The man he writes like this sensation, I kind of kind of being a trader and what that was likeeven Wolf of Wall Street. Yeah. Is

Michael Waitze 18:38
quaint. quaint. You’re talking about Bonfire of the Vanities?

Paul Meinshausen 18:43
Now Bonfire of the Vanities, though, the actual. Is it Michael Lewis, the one who wrote Michael Lewis up, Michael Lewis. Sorry, what was it? He wrote a book about his own experiences sort of something about being a trader? I think it’s

Michael Waitze 18:56
older brothers. Yeah. I can’t remember the name of the book. Yeah, I look it up right now. They made a movie about it.

Paul Meinshausen 19:05
Yeah, they made a movie about they’ve made, you know, movies about the move to high frequency and how all of a sudden it mattered. You know where you were? Because whether you’re close to the Chicago exchange, right, or the New York exchange? So yeah, I think that’s a good example. And in the same way, as speaking, specific to marketing, which was the industry and problem space that I was talking about that post, just so we can get very concrete, the way people do marketing looks like the way they used to trade. Yeah, it’s all you will do is you make a bunch of this as you look at a bunch of patterns and in some charts and things like that. And then you set up a rule, right, like trade here or trade. They’re the kind of old stuff where you look at a chart and you’d be like, here’s an elbow or I don’t even know the terms, but, you know, like, it’s very manual. It’s non systematic, if you don’t really They track whether it’s really effective, what the counterfactual would be with the very simplistic stuff, right. And, and that’s how we do marketing in that old way. And the way and so I mean, I don’t know if now’s a good time, but you mentioned this quote, you know, the quote from Karpati that I had an a quote, which it basically in the post, I’m using his his post, which is titled software 2.0, which he wrote back in 2017, I’m using that to just kind of try to give a very, because the post is not that long, to try to give some very clear, and I hope concrete examples, because you know, I don’t want to just wave hands and say, is changing everything, I want to be very specific about what we’re talking about. And to me enough, marketers understand the way marketing is done, to be able to recognize that that way of setting up a bunch of rules now even include some visuals in the post of a typical interface. In your Standard Toolset. Any marketer who works in marketing will look at this graph and will know exactly what I’m talking about. That’s how they do their job, right? That is broken, that’s not going to work. You cannot, you don’t set up a simple like, do this and do that then do that.

Michael Waitze 21:13
But this is why I want to have the conversation with you. Sorry to interrupt you, because now I think I have a much better understanding of what you were saying. And that’s why I love doing these calls. You’re right, this is very much like trading, right? In other words, if you look at programmatic advertising, it’s not different at all, actually, from programmatic trading. Yes. And yet programmatic trading a long time ago said, we have all this data coming in. And I think that’s why you made this quote, where he says Carpathia says it turns out that a large portion of real world problems have the property that it’s significantly easier to collect the data exactly what we were talking about right, then to explicitly write the program. But the key here, I think, and tell me where I’m wrong is that you’re still gathering all that data. And it’s way more data than ever before. But you’re still running it through the same rules that you’ve always used. So you’re not used, you’re not taking the benefit of having all this new and really information in important data come in. But then changing the way you use that data to make better decisions. Is that fair?

Paul Meinshausen 22:08
That’s, that’s well said that. So here’s the thing. We’ve, we’ve made it easier and easier from an engineering perspective to track and collect and store data. And let’s be, again, specific, when we’re talking about mobile apps. Every time you’re in a mobile app, there’s a data set that’s basically being generated, created. And it’s variously called like a clickstream. Usually clickstream refers more to web based, or an event stream. And just if we get really just, for anyone who’s not familiar with the technical side of this at all, let’s be very specific, any action, any thing you do in an app, let’s say you swipe down, you open the app, you swipe down, you click something, you open up a new screen, everything that you do is creating a lie in a in like, imagine just a spreadsheet, like a log, and it has its the log, right, I’m trying to be even more concrete than that are specific. So the spreadsheet will have a different columns, it will have a timestamp column. So that means it takes the exact down to the millisecond. When you did that thing. It’ll have a user column. So it’ll have your user ID, because it’s doing this for everybody in the app. So there’s millions of people in the app, have your user ID, and then it’ll have the event name that you did. And a lot of apps will have hundreds or 1000s of events. Because there’s such micro kind of details Did you swipe on one page versus on another? Did you click the Share on Facebook page or the share on Twitter page, you know, button, each of those will have a different name. And you end up with this event stream. So all of this data is being tracked, etc. It’s all there. It’s all possible to understand how are people using your app, but But still, the marketers making very concrete like simple rule decisions, and there’ll be setting up five or 10 or 2020 campaigns that they’re running. Or maybe they get to 100. But I’m talking about 100 with a million people, or really 10 million people, right how, and constantly changing and you’ve got 100 basic rules like that, that just doesn’t that doesn’t work. And the old model was and still is, and this has gotten where it’s gotten, you know, the change that has happened is you will ask a data scientist, someone who is specialized that looking at interpreting data, building statistical models on top of it, and then they’ll ask them and then that person will tell them okay, yes, this rule is good or not, and then they’ll put an but they’re still ultimately writing the tool in a separate you know, platform and a separate tool and, and programming that role. And what Kibana is talking about is that we’re easily generating that data. And what you could be doing is just reading that into very large models allows you to statistically separate, lots of different kinds of people write down almost to the user level to predict like, this is the next best thing for them, or this is the next best thing for them, and then have the machine respond to them dynamically to their specific user history and to the history of folks that look like them and behave like to benefit from that. Yeah, to benefit from that more collective view, that larger level statistical view. And that can, that can be what it does. And I’ll give you an example of this. It that also illustrates that it’s already here, recommender system, just think about any app that it has any kind of large scale inventory. Any app today has large scale inventory, a movement like Netflix, lots more movies than you can ever watch. An audible many more books that you can ever read in Amazon, many more things to buy than you can ever buy. And that goes on and on and on and on. Doesn’t matter what kind of app it is. It’s that has that that nature. If that so the question is, if you’ve got a small smartphone screen, you can show three things. And maybe the average person will swipe two times, right? 99 things, which nine things out of a billion things should you show to that particular user, right? That’s called the recommender system. Yeah, recommender system tells you which nine things to show on that user’s screen. Now imagine if we did that, based on a roll, like, that would be so dumb, I would just

Michael Waitze 26:34
so dumb, it’s like, if you bought this book, you would also like this, but really, how do you know that? Because I was just thinking how to humanize this in a way, right? Like, if you’re my friend, and I went to college with you, and then I was at your wedding or whatever, and I read a book, it may not even trigger me to recommend that book to you. But it may trigger a thought for me to recommend another book, because I know you so well. Right? And that’s kind of what you’re, that’s kind of what you’re suggesting. Right? Because

Paul Meinshausen 27:00
and imagine this, like you have that, you know, you know, that friend, and you just read this book about, you know, my inner strength or, or quietness of mind, or mindfulness or something like that. And this person is always partying, and you’re not going to recommend that book. But then you catch up at the wedding, and they just gone through something. Yeah. And they told you, they just gave you a couple of hints that they’ve gone through that, all of a sudden, contextually now you’re gonna say, this is the right book for me to mention to that person at this moment. Exactly. And that and that is what we do as humans, and why we’re able to successfully interact with each other. And, and, and where we break down interaction tends to be where we’re not able to take that context into, into picture. So imagine social media. Now, social media has increased the rate at which we interact with each other without any context. So now you just see a tweet, or you just see a message, you don’t know where that person was, when they wrote it, or what they really meant. They just wrote some words, and you just sort of say, Ah, no, you’re wrong, or very fast interaction, right? That’s a context of human interaction. And that illustrates that humans can be just as inefficient and ineffective at their interaction as machines can be when they don’t take context into into, into, into the picture. And that’s, you know, same as a dumb machine which recommends you to buy, you know, to buy a TV when you just bought a TV, like, why would you buy a TV, you just bought it right? That’s, that’s a, that’s a terrible recommender system. But the point is, is that, in general, nobody launches an app today without running a recommender system. The problem is, is that there are many places that marketing is working, which is still not using a recommender system, it’s still running manual rules. It’s still programming, just super basics and a message now send it three days later, send it one day later, say this generic thing. So that generic thing, and that’s, yeah, why a big part of the technology is not very effective today, and why it’s going to be so exposed to change and disruption.

Michael Waitze 28:59
So you made this point in this in this post that you wrote, and I’ll put a link to the post in in the show notes for this. But you’re suggesting then that that these recommendation engines are gonna have to change they’re gonna have to move from being ruled based to

Paul Meinshausen 29:12
what? Okay, well, recommenders are going to have to move from being pretty simple statistical models like logistic regression, that don’t condition very well that don’t, okay, let me deviate for just a second. So, in machine learning in the area of machine learning called reinforcement learning, there is what is called a concept of the Explore, exploit trade off. Okay, explore, exploit trade off. And that is a concept that is relevant to machines learning, but it’s also relevant to humans learning. What it means is just that, let’s say you move to a new city, okay? And you like to eat out, okay? And you go to a restaurant, and you enjoy the food. You explored you didn’t know you would enjoy the food, you went to the restaurant, you went to the restaurant. Now, imagine now you, if the next you know, the next week you want to go out to eat again, if you want to be guarantee that you will enjoy the food, you can go back to that same restaurant. Yep. Right. But there might be there’s a whole world a whole city of restaurants out there. And there’s probably many that you will enjoy way more than that one. So you will now explore, you’ll probably you’ll go to another new restaurant. Now at some point, you’ve been to so many new restaurants, the risks and this is especially true, for example, for higher cost exploration. Think about vacations, yeah. But then he got like new places, right? But you only get to go vacation once every every while. So the risk of going to a place that’s horrible, it means he’s blue, right? It’s hot back there, you go back to the place you like to explore, exploit trade, babies do it as we learn, they try new things. And then they go back to the thing that they like. And then they every once in awhile, try new things. Because also there’s like a treatment effect, which is you go to the thing you like you liked it because it was new, you’ll like it the second time, you’ll like it the third time, but the fourth time you start to like it a little less, right? There’s some curve we could draw, which would just say you like this less and less and less, the more you have it go conditional on some frequency or some time length, etc. So the machine is the same way now. So the problem is, yeah, so most recommender systems don’t navigate that explore, exploit trade off very well. What they do is look at what you’ve ordered before, or what somebody like you has ordered before, and that they recommend that thing. They don’t, they don’t have a good way of trying new things. And that’s what a different form of machine learning can help you with. So basically, yes, recommender system is gonna have to get a lot smarter. But there’s a whole lot of places in the world where we’re not even using basic recommender systems. So mobile app marketing is overwhelmingly not even using a basic recommender system. It is literally a human Bayes rule. And actually, there’s another space. So my lat the last company I founded in India called Pay sense, was a FinTech company. And it we gave loans. And the vast majority of loans not too long ago, in a market, were driven by rules, it was basically a set of instructions, do you have this much income? Do you have a score of this much? Do you have this? And this? If you do you get a loan? If you don’t, you don’t get a loan? No. And this, I think, is part of the broader change is that we are moving away from deterministic fixed things, a deterministic world where input output is consistent all the time, it’s exactly the same towards a more probabilistic world, where the room for error is acknowledged within the system. In fact, the system embraces air as a way to get progressively better over time. And to realize that the world is just it is itself dynamic and chaotic. It is not, it is not fixed. I mean, that’s what allows us to get better is the fact that it’s not fixed. So I think if you look on a longer time horizon, we’re just on that Lauren, Lauren, longer track of change in history, is, do you know, this quote from Einstein, which was, he was sort of rejecting a particular kind of science when he said, God does not play dice. And he was, he was rejecting a form of physics, despite how brilliant he was, because he just too uncomfortable in human history, we’ve always viewed that things should be deterministic. They should if this happens, and then that happened, and then this happens, and then that happens fixed and more and more as, as a species, we’re able to recognize that that’s actually not how the world works at all. Actually, it’s very probabilistic. Yeah. And so our software will have to have to get that way as well. And it’s a good thing.

Michael Waitze 33:55
It’s a it’s a great thing. But it’s just such a great quote, right? Because the world is going to become way more prevalent, probabilistic, because it’s that way anyway. And we I think we’re constantly removing that discomfort. I want to share a funny story with you. And then I’ll let Joe because it reminded me first of all, I went to a new restaurant, I mean, a new city today. So that whole, that whole analogy was just so fucking perfect. But when my daughter was two or three years old, her mother would feed her corn soup, right? So why does that matter? Because every now and then my wife would ask her, and she’d feed it to her a lot because she liked it. And every now and then she’d asked her as a three year old or four year old, what do you want for dinner? And she’d say, corn soup? And my wife be like, Why do you always say the same thing? And I’m like, that’s the only thing she knows. Yeah. So her personal recommendation engine was based on the rules that you’ve already set up for her. It’s no different there than it is in the marketing world, right? If you have this fixed set of rules, and it’s very deterministic, but if it’s not, then it changes, the probabilities change. It’s way more Yeah,

Paul Meinshausen 34:55
yeah. Side note on that because I have a two year old now and is this book Bread and Jam for Frances which is like a very old school book. But the thing about that as well is the reason is because everything is new for three year olds actually, they’re massively overriding on explore. So they because everything is new. So I think a lot of times, they do things that are frustrating to us, because that’s their sense of stability. Like we all have the right degree of change that we can absorb at a given time. And a young person’s like, they’re dealing with so much change in all of their contexts. Sometimes they hold on to that story, while they want to read it over and over and over again, or that food, they want to eat it over and over and over again, because it’s the one thing that they can control that is predictable. Exactly. And I think that as a society, we’re going through something very similar, which is like, we can’t handle so much change at one time. So we are kind of clinging to certain things that help us feel even though there’s no real good reason fundamental like object like outside of ourselves for that thing. It’s like it’s just a way of anyway, but yeah, I think

Michael Waitze 36:05
Pretty interesting stuff

Paul Meinshausen 36:06
Yeah, yeah.

Michael Waitze 36:07
Okay, Paul, I’m gonna let you go i’ll put a link to and thank you so much for doing so I really appreciate particularly short notice and I will say this to you as well. Anytime you want have a conversation like this and record it, let me know I’m happy to do it at a moment’s notice.

Paul Meinshausen 36:18
I enjoyed it and thank you for offering and for doing it. And yeah, well, I’ll we’ll post it on our Website and LinkedIn and all that stuff.

Michael Waitze 36:26
I’ll give you the link. Thanks.

Paul Meinshausen 36:28
Okay. Bye


Latest Episodes:

EP 327 – Event Mesh Networks Explained: What They Are and How They Enhance Operational Efficiency – Sumeet Puri – Chief Technology Solutions Officer at Solace

EP 327 – Event Mesh Networks Explained: What They Are and How They Enhance Operational Efficiency – Sumeet Puri – Chief Technology Solutions Officer at Solace

“So an event mesh is a data movement fabric, okay, yep, so the network, but it’s the layer between the network and the applications. So if we explain it a little bit further, event broker is this technology that allows you to move data in a one to many manner, in a publish to subscribe manner. So for example, if an order is created, and we say that order event is published to the event broker, that order may have to go to the order management system. It may also go to the digital receipt system. It may also go to the loyalty system. So the same order has to go to multiple systems. And this pattern is called publish, subscribe, which is done by an event broker.” – Sumeet Puri

read more
EP 326 – Never Forget Who the Enemy Is – Gene Yu – Founder and CEO at Blackpanda

EP 326 – Never Forget Who the Enemy Is – Gene Yu – Founder and CEO at Blackpanda

“Resistance training is, you know, they can take away everything in terms of like, your comfort, your warmth, your food, you know, etc, make you disoriented, but they can’t take away your will to resist, right? And they can’t take away your mind, right? That’s the thing that you can always control in terms of resistance. And so I emphasized that to her what they taught us over and over, which was, never forget who the enemy is.” – Gene Yu

In this episode of the Asia Tech Podcast, featuring ⁠Gene Yu⁠, a Founder and CEO of ⁠Blackpanda⁠, we explored innovative approaches to cybersecurity and the vital role of crisis response. Gene’s unique background as a West Point graduate and a former Green Beret significantly shaped BlackPanda’s approach.

read more