051: Adam is My Machine Learning Model

So what video game are you buying? Well, I'm buying one that's from the Steam store. It's on sale. So I'm buying it.

Oh, it's on sale. So that's what's for dinner. I see. What is on sale in the Steam store?

Like, what are we talking about? Like, what isn't on sale in the Steam store? Have you ever been to the Steam store? Everything's on sale.

No, I haven't. So what is on sale? Are we talking like $20 or is it like $50 or like what is the price of a game in the Steam store generally? Well, the game that I just bought is $5.

Okay. It's not terrible. I mean, it gets like the price on app normally or something like that. The Steam like goes on heavy sale, like at random times, like they're just like, hey, you know what?

We're having a great sale. So yeah, this was half off. Okay. Okay.

And I got an email and it was half off in the sale ends tomorrow. So what does it do today? Jump on it. Yeah.

Nice. Okay. Well, as long as it gives you hours of phone, $5 worth it. I was, I mean, this is my naivety with video games is that I assumed that like it would be like, you know, Xbox pricing where it's like, what is the game on Xbox like 60 bucks when it's brand new?

And even if it's on sale, it's like 39 or something like that. If you're going like AAA games, then yeah, it's going to be pretty expensive, but I'm supporting the indie scene right now. So this is from an indie developer and this game is old too. It came out in 2015.

So it's not like I'm buying anything that's like cutting edge. It's not like I'm out buying Spiderman on PS4. It's I'm buying Undertale by Toby Fox and heavily discounted on Steam because it's been out for a long time. But I've seen the game been played a bunch of times and $5 was within my gaming budget for the month.

So why not? Okay. All right. That is what it is.

I'm going to just let that sit there. I'm not going to have to read you anywhere about that. So well, you can't because I always spent $5. Oh, well, you've gotten mad at me for sending in-app purchases for $2.

So I'm pretty sure I could be ready, but I'm not. Let's just say that this game is a unique game. It's not like the other games. It's not the exact same genre.

He's the exact same. And it's something my kids can play. A weather app is not something my kids can play. I mean, sure they can.

They could they could open every weather app. But I mean, the mere fact that you have a $5 a month gaming budget tells me that you're spending $60 a year on games at least. And I'm spending a whopping probably $0 a year. Maybe there's a couple dollars in there.

I just so like happened to splurge on. I mean, I did buy Altos, obviously, so it was not like $5. I think it's the only game I bought in the last 12 months. So, you know, my other $50 or $55 is gone to weather apps and every other app purchase that I needed to.

I would venture to say that I get more time and enjoyment out of my games with the budget that I have than you do out of the mini apps that you try and buy and then never use again. I think if we compared our little iTunes receipts, I'm pretty sure I would come out ahead in terms of just overall enjoyment, unless you enjoy looking at, you know, weather maps all day. I get enjoyment out of things that you don't. That's OK.

There's nothing wrong with that. And you wanted to berate me for paying $5 for your game. That was on sale. Good.

I'm good. So what's this business about you running Linux? Well, I know we've talked like a couple times where I do like to every once in a while go on like a distro craze where I'll just try flavors of Linux for, you know, like a couple days and be like, hey, you know what, I'm just going to try this out, see how this is. So I'll spin up a VM and, you know, install Linux and mess around for a little bit.

And usually it quickly just gets kind of pushed off to the side. It's like, OK, well, that's done. That was fun. That's great.

But I'm actually thinking about instead of running it on a VM running it on bare metal. So taking a computer that I have and installing a Linux partition on it, like just running it actual on actual hardware and just kind of living on that for a week. And I think I'm going to use that as my main computing machine when I'm not at work. So I think I'm going to see how that goes probably next week, just because this week I'm probably not going to have the time to get everything installed and configured.

But because this weekend, you know, I'll download the latest Ubuntu and or Ubuntu, I can't remember how we were supposed to pronounce it. I don't recall either at this point, but that was my only question was which flavor? Yeah. I think I might just go with Ubuntu just because it's pretty friendly.

They have pretty wide driver support. And so I don't have to go like hunting, compiling my own drivers, which I don't really want to do, especially since they're going to be using an affordable machine. Like I don't want to have to go through. Oh, the display doesn't show up.

Like, great. OK, well, now what I'm going to do. So I don't want to have to go through that. And Ubuntu has pretty wide support.

So I think I'm going to start there. And we'll see how it goes. So I'll have follow up in a couple of weeks on how that went. And obviously I'm not going to, you know, do any editing on that machine.

So there won't be any slip in our dates and our timelines. So in a little bit of follow up, I was kind of reflecting on our episode last week ahead at work this weekend. I sort of just kind of broached the topic with some coworkers and just said, hey, how do you guys upgrade your TV and when do you upgrade your TV and what do you look for in a TV? And I was really interested in, I think the motivation for me was sort of how often they upgrade their TV.

And I should kind of color this with like, hey, I work in the tech field as it is. So if there's people that are on faster upgrade cycles, it's probably us more than anybody. But and I heard some really wide ranging answers. I mean, I heard everything is the lowest answer I ever heard back was around two to three years.

And the highest answer I heard back was when it breaks, kind of like what you said originally. So I, I took that and I just kind of was, I was kind of surprised by it. But I also went to the web and kind of was like, has anybody ever written anything on this? And I did some digging around.

And so I threw a link in the, in the show outline here on them that it pretty well kind of summarizes, I think exactly the results that I kind of queried at work and, and found that basically, it seems like very little people, like there's like single digit percentages of greater TV under a couple of years old. People either are between the five to 10 year area and there's like 25% of people that wait until breaks and there's 25% of people that pretty much are like somewhere between years three and five. So I thought this is kind of interesting because I upgraded my TV right around nine years, eight nine years, and that puts me pretty squarely in the big group. But, but it was just interesting to me, the distribution of how people upgrade.

So I think we, we, we, although we are in different camps at him, we're not too far off from the general public. Well, the other thing here is like, once you actually start looking into the, like the, the comments on this forum that you sent me, a lot of it is like, yeah, you know, I'll wait, you know, this many years, but if there's some brand new technology that comes out or four minutes change or something that's big enough to get me to buy a TV that will break the cycle. So I think, yeah, most people, like you said, an overwhelming majority are over three years and the only thing that would break that is if like, Oh, tomorrow eight K comes out and it's, you know, standards adopted and there's tons of eight K content out there and you magically have eyes that can see eight K. So yeah, then everyone goes and buys it and you and I know that doesn't happen like what every 10 or 15 years, something like major comes along like that and it just takes over like fire.

But for the most part, it's pretty rare that that stuff happens. Well, unless you're the TV company and you want to try to get them, you know, to buy TVs and you say like, Hey, 3D TV, that's the next big thing 3D yep, working with your bottom line there. I like it. Yeah, it's the TV cycle man.

It's vicious. Anyway, all right. So this week, I want to do kind of dive into machine learning, which I guess I should start off by saying, are you a little happier with this week's topic? Well, it's not antiquated technology and it's not app reviews.

So yeah, it's a little better than the previous topics we've had. Okay, cool. One step in the right direction. So with that being said, yeah, I wanted to just jump into this a little bit and obviously we're going to try to keep this really, really, really brief because of course we don't dive kind of crazy into topics.

But I guess start out by saying, you know, sort of wanted to find just loosely what is machine learning for anybody who doesn't know because I feel like it's just kind of a phrase that's been like thrown around a lot lately. So every hot tech Silicon Valley company is saying they're doing machine learning and it's been around for a long time. But do you want to kind of summarize that for us? I don't want to do it.

You can do it. Okay. Well, you're the big stats nerd. So this is very much more like up your alley than it is mine.

So my limited familiarity with machine learning is kind of going to be trumped by your number crunching 18 weather app knowledge of models and how this stuff all works. So go ahead and give me your 50,000 foot view of machine learning. Oh, well, I was going to say really abstract in this, but I mean, that's why I was going to keep a lot of the stats out of it. But I mean, my sort of trivial understanding of all this is that essentially we are sort of training computers to do what humans can kind of do.

And that can be, you know, a million different things. And so, you know, basically, usually you have to feed a computer some set of data and outcomes the ability for the computer to hopefully then recognize patterns in the data. And that can be, you know, from everything from as Adam says, my weather apps to, you know, it could be something like health related, it could be crime, it could be a million different things. It could be finding a dog and a photo, you know, that's kind of like the simple task we hear about a lot today, I think.

But I say simple in air quotes there because it's not trivial, but it is something that is less complex than some of the higher end uses. And does that, is that fair to say, Adam, right, from a 50,000th view? Yeah, so basically it's just taking, you know, a set of data and then training something or using this machine learning to train an AI to basically identify certain things or identify patterns or trends in data is kind of what I gather. Yeah, I think that's a good working definition to start with.

And then, you know, I mean, without going too in depth, I kind of already alluded just a real basic amount of kind of where it's being used. But I mean, it's being used in everything from, you know, data centers all the way up to, you know, really complex levels to solve, you know, really big problems in health and so on and so forth. I think IBM's using it in their Watson, you know, platform for health and then all the way down to just like your cell phone is using it now. And so, and you can imagine every computer and, you know, piece of hardware in between, you know, the two is, of course, using some flavor of this I feel like for the most part and it's growing and growing and growing.

So this is probably a really hot area that's only going to get more hot over the next, you know, couple of years. Well, I think, you know, you mentioned all these different fields where it can be used. So pretty much anything. And I said that you would like this because you're kind of the statistics nerd, like you love the numbers and you love kind of looking at large amounts of data.

And the reason it's kind of catching on is like, well, we have all this data, right? Like, let's say, you know, your favorite search engine has been harvesting this data for years and years and years. What are they going to do with that? Like, what's next?

Like, okay, we have this data. What do we do with it? Right? It's going to take a massive army of humans to kind of analyze all this data and a bunch of, what's the correct word for statistics, people, is it statisticians or statisticians?

Yeah, there you go, that's the one statisticians. So it's going to take a large army of those who are very expensive, you know, to kind of get anything useful out of this data or we can use kind of computing because computing is much cheaper than it used to be and we can use, you know, these artificial intelligence or this machine learning to analyze this data for us, right? We can train it and say, hey, I want to see all of this or I want you to do X, right? Here's a bunch of data, learn from the data and identify the trends and the computer's like, okay, and it just does it.

So I think you're seeing it spin up because the more data we collect, you know, and I'm just not saying, you know, search engines, but also, you know, medicine, like now that everything's digitized and just kind of how we feed all this data into these systems, now we're going to get out of it. That's where all this machine learning comes into place, right? And I have some interesting fields, I think, that are like my favorites in terms of how they're using machine learning, but I think that's why, you know, from my perspective, why things are heating up. Okay, cool.

Well, let's, before we jump down sort of the, to like the nitty-gritty, let's open the hood and look under it and figure out, you know, a little bit more of what's causing that all the work. What are your kind of fields then that you kind of like, and I have like two that come to mind for me, but you may have already kind of heard them. Yeah, the health one, I definitely heard you drop that one and that's the one where I'm like, yes, please, yes, because you hear about all these different advances in just screening, right? Screening for cancer, screening for preventable diseases, screening for everything where you'll take a data set, you know, that's been kind of curated over a long list of people and say, like, hey, let's take skin cancer, for example, which is a pretty common scenario, a common example for how machine learning can benefit.

And you have this weird looking mole and you say, hey, doc, I have this weird looking mole. I'm like, hmm, yeah, that is weird. Let me take a look at it. And the doctor usually will diagnose the dermatologist will take a look at it and identify whether they think it may be cancerous or not.

And with the power of machine learning, they can basically compare your mole to samples of, you know, thousands and thousands and thousands of other moles like it and identify, like, okay, yours has these same characteristics, the same characteristics and identify with a higher accuracy than a doctor will if it is cancerous or not, or if it is potentially cancerous or not. So I think that's like a huge plus like, hey, if we can now have better diagnosis and better identification of diseases and have early detection and screenings, then that's just going to increase all of our lives, or I guess the increase the quality of all of our lives because, hey, we'll live a little longer because, you know, the computer will tell us get out of the sun. Yeah. I think the health is a huge area for me.

I'm really, really excited about what it's going to be able to do there. I know you dropped, you know, the skin cancer thing. I think there's tons of other things that are out there. I personally know somebody who's working on some machine learning stuff to predict when adults that are, you know, later in life are likely to fall.

And so it can predict, you know, like, hey, this person is susceptible to a fall, you know, in the next six months with like, you know, 75% chance, you know, and it is feeding into the model, everything that they know about like, hey, all these people, you know, usually beforehand go through these sort of symptoms or have this issue. And so it's trying to, it's best to kind of predict that stuff, which, you know, if you know anything, of course, once you're older, if you fall, that can be a huge issue. So, you know, that is something from an insurance perspective as well as a, you know, just care in terms of like, you know, best outcome for the patient, knowing that data before you go in is really, really helpful. The other one that I find really fascinating is being able to predict crime.

And if you live from anywhere outside the United States, you may not know, but there's a city in the United States, Chicago, that is kind of, you know, in the last couple of years, well well-known for having really, really high crime. And they have, I know their police department for the last couple of years has been working on machine learning to predict where crime hotspots are going to start popping up. So they can try to predict where the crime is going next. And so they feed tons of data about, you know, suspects they rest, where they arrested them, you know, where murders have been, you know, committed and so on and so forth.

And they actually try to predict and then place officers in the right areas so that crime will go down, which I thought was really, really innovative. Just wait till it starts trending in the stock market. And then, you know, which we already talked about, I think a while back, right, where you have these robot advisors, right, that are constantly learning and trying to figure out when the best time to sell is or when the best time to hold on to stocks is and kind of identifying trends and say, okay, you know, this, this stock has been writing up high, you know, like looking at news articles and picking out keywords, right, like pretty soon you also have, you know, these models trained on just investment portfolios, right, where you say, hey, I want to get the most I can in six months and it will, you know, where you used to put together a big risk portfolio. This is going to be more dynamic, right, where it can kind of learn on the fly and kind of train as it goes.

But anything where you have tons of data, you know, you're going to see these things pop up. And I think, like you said, you're also interested in the ones that are going to kind of benefit humanity, right? So the ones that's like, hey, how do we decrease crime? How do we, you know, increase longevity?

How do we identify, you know, better ways to sustain like energy or the planet or whatever, right? That's the exciting part for me. But it also comes with some, some other not so exciting things, like I said, like gaming the stock market as well, right? Because you can also flip that and say, like, hey, you know what, I actually want to do what this model would do, but kind of sabotage it or feed it incorrect data or see if I can, you know, throw it off.

Yeah. So what's the phrase with great power comes great responsibility, right? So that's a spiderman quote. You just quoted Spiderman.

Oh, excited. You right now, big boy. Well, did you know what you were quoting? Yes, I did.

Okay. I mean, you said, you know, I think that's the quote. It's like, wow, look at you. Yeah.

I mean, I only saw what was the one that came out probably like 10 years ago. So yeah, I think I saw that one. So if it's Spiderman, I mean, if it's the original Spiderman and Spiderman, you have the same Ravi one with Toby Maguire. Sure.

I'm just going to nod my head and say yes. Well, because they've redone Spiderman like eight times, which is why I said like the one that came out 10 years ago, because I feel like there's probably like five versions of the movie. So yeah, it's the one that came out, you know, 10, maybe 15 years ago max now. So that's definitely the Toby Maguire one.

Okay. Cool. Yeah. So I saw that one.

I remember it. Yes. I definitely dropped that quote. Here's stuff now.

Cool. So we've dropped in like hinted at that essentially with machine learning, you need a model and let's dive into that a little bit. So a model is just for starters, like what is that is basically how you get a computer to get trained to be able to pattern recognize whatever you're trying to teach it. And so usually you basically need to start with some set of data and you need to feed that data to computers that then or like computers running an AI, as you said, and that data could be anything.

I mean, it could be, you know, pieces of text, it could be images, it could be videos, it could be tons and tons and tons of stuff. And whatever, you know, you end up kind of feeding it is your model. And I think, you know, depending on what you're trying to do, there are different ways to accomplish that. And it's, it's all kind of, you know, at that point from there on, it's a bunch of computer science, which I'm not even going to dive into.

But, you know, the more data you give it, the more accurate, hopefully it's able to predict the next, you know, sequence in the pattern, if you will, is that a good sort of summary item? I mean, I have a very trivial, trivial understanding kind of from here on out as to what's actually happening. Yeah, this is why I put the note in there saying like, Hey, let's just cover basic overviews because I don't want to get into the integrity of something neither of us understand, and this is definitely kind of kind of dive into that territory, which is like, okay, we talked about the basics of what it is, kind of what a model is, but I think kind of sticking up in our 50,000th of you and letting, you know, all the, the actual like, hardworking, smarter than us people handle all the definitions, because I mean, I don't want people to come to us and be like, Hey, Adam, tell me all about machine, like, um, machines, they, they learn, they go to school, they get smart, they do things, they got their PhD and now they're learning. Yeah.

But I think, like you said, it's, it's all about pattern recognition. So it's, your model is basically what you're trying to match in the pattern or what you're trying to identify as a pattern or how you're going to basically feed all of these, these variables into this quote unquote model to identify your pattern. So that's, that's just what I look at it. I'm done, man.

I don't know. I don't know. Nothing about this stuff. No, no, you're totally fine.

I mean, I think I have a pretty similar understanding to you. And, um, you know, I mean, like what that's like, I mean, let's, let's take pictures to a very trivial, trivial example. It's like, I know my phone is able to recognize dogs, like I can search dogs and it somehow knows what a dog looks like. And so, you know, when you go to train a phone or train anything to do that or computer, you feed it to my understanding, like thousands and thousands of pictures of a dog.

And over time, it starts to realize that like, Hey, you know, a four legged, furry looking thing standing there with a tail on the end of it, you know, it's eyes look roughly like this and so on and so forth. It starts to pattern recognize those traits over time. And it's then able to, once you've fed it thousands or hundreds of thousands or millions or however many pictures of dogs, it now gets better and better and better at detecting that. But you got to sprinkle in some cats though, you know, make sure that it knows that those aren't dogs.

Yes. Yes. This is important. I think you don't only want to train your model in the way that you kind of, I mean, as you refine accuracy, you got to be able to kind of give it a curve ball every once in a while.

You throw in a giraffe and it's like, Oh, it's a four legged thing, but it's also 15 foot tall. So that's probably not a dog, you know, these sort of things, but, you know, the model itself is only as good as the amount of data I think you put into it as it is kind of like the trivial way of saying it. And so, you know, if you give it only 10 dogs and one cat, it's probably not going to be the best at it because let's face it, there's like, have you seen the dog show? You know, every year, there's like what, thousands of breeds?

So it's like, yeah, it's never going to be able to tell that. But, but, you know, the more and more you give it, the better it's likely to accurately, you know, pull out the photos of the dogs. So can we, can we go back to the dog show? I mean, yeah, you said thousands of breeds, right?

Have you ever watched like the Westminster dog show? Yeah, yeah. I've watched. I mean, I loosely say that I watch it.

I don't know that I've like devoted my entire attention for the like three and a half hours it is, I feel like, or something. But yes, I've watched it, especially at the end. This is important because I'm training my model for movie recommendations for next year. So I need to gather some information about the data, you know, things that you like and don't like.

So that's why I'm asking about the dog thing right now. Just it'll come to fruition in a year. Okay. But, so what's your favorite group?

Because you know, they have all the groups of dogs, right? Like there's the toy and the pointers and the hunting dogs and the other dogs working dogs and I mean, isn't the, isn't the terriers considered a group or is that like considered just a breed of dogs? I don't remember. I think the terriers.

I think they might be part of a group. Okay. I think there's so many terriers that they're like, Oh, there's so many freaking terriers. Yeah.

It's definitely between the terriers and the like all the Labrador group, like all because there's like a gajillion laboratories too. So I kind of send a favor like medium to larger dogs. And so I'm not a fan of the little like terriers or get like tiny though, you can get the little rat terrier things though. And yeah, the most of them, like, I mean, my actual dog, which I guess now at this point I need to include a photo of is like, you know, 12 pounds, but, um, and he's, he's all like working dog.

And that in the sense that he is like, like, uh, to the best of our knowledge, like what breed we think he is. They, they actually like chase bunnies or branded like chase down bunnies. And I can tell you this much. He's seen a bunny like 25 times in his life and every time he's seen a bunny and I've let him off his leash, he is bolting at a hundred miles an hour towards the bunny.

So I have to assume it may be true. Okay. Um, the movie I'm picking up is best in show. I don't know if you've ever seen it, but it's a movie about, yeah, okay, well, it's one of the list next year.

It's a comedy movie about the dog show and all these characters who bring their dogs to the dog show. So all right. All right. It's on the list.

It's on the list. Yeah. My machine learning model has been updated knowing that you like dogs and knowing that you like Terriers. You'll definitely like this movie.

I, uh, I'm looking forward to that. That actually sounds good. Comedy movie dogs, like, you're already kind of, you know, training the model in a good way, you know, I might like it. So see who needs machine learning?

You just got to feed out of, you know, like two things, things of data and he knows with a 99% accuracy, I can recommend a movie that Lou may or may not like. Um, I'm just going to say from here on out, Adam is my machine learning model. So, you know, uh, I will just use a human machine learning model and I'll feed you data. We'll be good.

No, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, train on the data. Um, so with all this, though, like, there, there is definitely concerns, I mean, like, we always kind of play that card on this show of just a counterpoint of, um, you know, like, hey, what concerns do you have? So I'm kind of curious. Do you have any, um, you know, big concerns right away?

I know you kind of alluded to some, but um, what, what are your kind of eyebrow raising concerns with machine learning? Well, Uncle Ben, you dropped the quote with great power comes great responsibilities. You know, I mean, like, what if I'm not responsible with the power, like, if I started using that power, what can I do? Lots of bad things.

So I think with anything, the more data you have the, the more, the more you know, is a rainbow, the more you know, the more you know, the better you can influence to the positive or the negative, any outcome, right? So let's, let's take something that's a little, a little more volatile. Like, let's say, I don't really want to do the stock market one again, but like housing markets, let's say. So let's say that there's someone who has gathered a whole bunch of open source data of when houses are selling, um, what they sell for, what the trends are, you know, and they can identify kind of like hot markets that have, that have popped up and then a developer who has spent, you know, millions and millions of dollars goes out and buys up everything and then they know with 100% certainty that this market is going to explode in two years.

So they buy up all the lots for cheap and then they rip everyone off. Like, there you go. There's greedy, you know, capitalism, you know, taking this machine learning and using it to fill their own pockets, which you can argue is not evil per se, but it's kind of shitty for, you know, everyone else who wants to buy a house. I thought of another example as we were going through this and now I've totally forgot.

I don't remember if it was something with drugs or something with a prostitution was one of the two. Or dark, dark always dead down, but, uh, all right. So I'll, I'll let you think about that for a second. I'll throw in a couple that come to my mind, but, um, so one that came to my mind was sort of an abstraction layer above the actual misuse of machine learning, but it's just the concerns about the fact that if you're feeding computers, a lot of data and they have to crunch on that data, you know, there is a degree of energy impact to the earth.

And so, um, you know, I mean, I know data centers are there anyway. So they're, they're there built anyway, but if they're being used in lots of data is being fed through them, then, you know, there is a adverse effect to energy as a whole. So, you know, I think machine learning in general, you're feeding it tons and tons of data and it's really crunching, you know, CPU processors on and GPUs and all sorts of stuff on a lot of computers. So you very well are actually, you know, causing some offset.

And so, you know, we have to always, I think be concerned about, are we doing this for the right reasons? Are we getting, you know, more out of this than we put in? So as long as we can see that, I mean, of course, there's a whole other argument down the path about, you know, well, you can offset it with X, Y or, you know, different ways to make the, um, the energy clean, but if we're doing that, cool, that's great, but that is a really good thing. And the other thing that I think is important to point out is I brought up the crime statistics one and it's not exactly what I would say would be easy to do is, you know, that I would say just like a lay person or say the criminal could, you know, go out and train a model, but it's very unlikely that they would do that.

But reversing that sort of stuff. And I mean, I don't know if this is exactly where you're going with drugs, Adam, but, um. No, I remember my drug example, but you're talking about like playing the reverse card and you know, right, where it's like you're going, you're going, boom, you get that reverse and it's right back on them. Yeah, exactly.

So you're like, you train the model and you're like, okay, well, where are the police going? Well, if you know where the police are going, let's commit a crime somewhere else, or let's, you know, let's do the drugs and prostitution over here. You know what I mean? Like if you're smart enough about that stuff and you have access to the right and you can get it like you, I mean, with things like drugs and stuff like that, like those industries are well funded.

So, you know, that being said, you know, you could make an argument that, hey, there's a way to, you know, get this done or like, hey, you're trying to move drugs across borders. You know, there's tons of stuff there. So, you know, reversing the approach. Yes, I cat mouse game, like meow, meow, meow, like just back and forth like cue the Benny Hill music, right?

In the same place. The cat mouse game of machine learning, I love it. And you're like, your cats definitely sound like they're like cats with lasers. I'm just going to point that out.

Well, I mean, if it's any cat on the internet, it's going to be cats with lasers, of course. But like you said, if you can identify a model, then you can either, if you know your model is being identified, you can gamify that model or try to throw off that model, which the more data you give it, it's much harder to do, right? Like, if you fed, you know, years and years and years of crime statistics into this model, it's going to be hard for someone to kind of like overthrow that in like a year, right? Or like, to your point, like, hey, if they also know all of this stuff or all of this data that they have as well, then great.

I think the thing that naturally comes up with that and other concern, and we'll get to my drug example, because it definitely was drugs that I wanted to talk about, but not the kind you're thinking of. And the other thing I wanted to say was like data privacy, right? Where this is kind of a little more abstract because I think, you know, the data that we're dealing with is very, very high level. It's not very finite, but you can, you can get very finite and start, you know, really identifying trends with people in certain areas.

And like I said, the more data someone knows about you, the more they can train the machine learning. I want to say something really quick here is actually pretty relevant that you're talking about machine learning right now, because I was just listening to a podcast the other night by RadioLab, and it's called Post Note Evil. And it's basically all about Facebook's reporting, like when you report something as abuse or as offensive on Facebook, basically how that team kind of grew from a group of people basically going through and manually auditing all these flag posts to now a machine learning model, which will automatically identify and categorize and only escalate things that it's unsure of, right? So you have all these people kind of feeding this model to identify what is considered abuse on their platform.

And then you get into the whole thing of, is that invasion of privacy? Is that censoring free speech? Is that all sorts of things? I'll give you a link so you can throw that in there, but I think, you know, as with everything in technology, we have to walk the privacy line.

And I think, you know, it's worth mentioning that, hey, privacy is a thing, be concerned about it. And if you're not paying anyone anything, then you are the product. Just remember that. Yeah.

We've hammered that home. I feel like a number of times. So what was your drug example? Okay.

So my drug example is insurance. So I've walked this dark path of insurance, like I think insurance companies, I don't think they're evil. I just think, you know, like anything, they're managing risk, right? Which is their whole job.

Right? But you mentioned, you know, hey, we have basically models that can identify when someone is likely to fall or when someone is likely to get sick, let's say, insurance companies who make their living off of data and risk analysis would love themselves some machine learning, right? Like, hey, show me all these trends or show me these trends about people in this demographic in this area and starting to give you a hyper accurate, hyper personalized rates based on your current situation, right? So I think that's a way that can be a positive, right?

Because if you are taking care of yourself and, you know, you have a decent, you know, track record and they can identify that you're low risk, hey, you'll probably get a lower rate. But if they identify like, hey, you might think you're fine, but you know, your lifestyle choice is all points of this. And it's like, boom, now your rate's much higher. You can argue like, hey, well, that's just, you know, good risk analysis.

But that was the thing I was getting to. So, so drugs being like, hey, drug companies, you know, working in machine learning, saying, hey, I can identify that, you know, this year or this amount of time that all of these people are going to come down with whatever I can jack the price on whatever, which they do every year anyway. So I don't know why this is any different. But I think just the more data, you know, the more that they'll be able to kind of game those systems, which can be detrimental to people who need care.

Yeah. I can agree more with that. I mean, it's, there are some of these industries that I think it's very questionable whether they should maybe be allowed to use some of this and, you know, hopefully regulations kind of come in there and sort of mitigate some of that if there are concerns there in the future. Oh, yeah.

Government regulation. I was going to, we'll just wait, you know, 20 years for that to happen. Yeah. I can't wait till we're trying to explain to, you know, some like 75 year old Senator what machine learning is.

Yeah. When we have the all hands. So how do you make money? We feed the AI and the AI spits out ads.

This AI shits ads. It's like the golden goose that shits ads like what do you want from me? Yeah. Every time you like a puppy, it's feeding this thing and then we can spit out puppy ads.

So thank you, Senator. Yeah. It's not, it's comical to watch them try to understand anything tech related. So anyway, yeah, they're getting better, right?

Like we can't say like, Hey, it's all bad, right? Because at least they're actually kind of holding these hearings and these committees, right? Because now they're starting to learn, right? It's a whole thing of, you can be ignorant to the and just say, Hey, I don't know what that is.

I'm going to pretend I don't know what that is. And I'm going to stick to what I know. Or you can kind of try to learn based off these experiences and like, Hey, no question is a dumb question, right? Because it makes you smarter because now you know something you didn't know before, right?

So I'm going to say all these senators right after the first one, which I think most people watch them. Most people find pretty comical when, you know, Mark Zuckerberg is getting grilled about, you know, his platform and what they actually do and how they make money and if they view themselves as like a censoring free speech. And I think they learn a lot from just how the internet kind of works and how people are using things. So like I have to say, like we may find us under this, but props to them for actually holding these committees.

So good job guys. Keep it up. Guys, guys and gals. Yeah.

I mean, I'm all for them learning more. I don't want to send a message that I'm not. So it is a, I'm all for the support. I just know it's a, it's an uphill battle and it's one that's kind of flow at times, but we're moving.

So that's good. So the sort of final thing I was when I kind of touched on is where are we going? Where is ML, you know, machine learning, you know, where's it headed? And so, you know, I think we are really at the very, very earliest ages of this.

And I think, you know, we don't even kind of fully understand everything about it and where it is going at this point in time. But I think the hot button fields we've kind of touched on. And I think the biggest thing that I can see right now is healthcare, like over the horizon, I think health is sort of where this goes is the immediate, you know, future. But I think down the road, and this is a little bit of a hint of to where I think it might be going is I think it might have the ability to literally solve other problems like traffic.

And I think that is like one of the first things that came to mind for me is that alleviating sort of more real world, everyday problems of, you know, traffic being an issue, if we're able to feed a model with, you know, the traffic congestion problems we're having in major cities, we can easily move people around and solve other problems as well that are just kind of the everyday pain points. But of course, the well-funded stuff right now is kind of healthcare and crime being, you know, one of those. And I think the only other thing that I think came to mind was for me at least was like military because this is, of course, like, you know, one of the well-funded branches of government that, you know, being able to use this stuff on the battlefield to sort of tactically figure out where to strategically put people or to, you know, understand how we're going to attack something, you know, might be incredibly useful. So I think those are some maybe hot button areas, but I'm kind of curious, where do you see it going, Adam?

Oh, my gosh. I can't believe that we forgot all about self-driving cars. Like, I can't believe these glazed over that one because, yeah, you mentioned traffic, right? Like, fuck traffic.

Like, let's just not have to drive, right? You train a model on how to drive and identify what's a car and how you, you know, proceed and how you handle certain situations, right? Like, there can be tons of calculations that go on there. And the other thing I have to say is like, I'm going to echo all your sentiments, but I'm going to put them in a much more like flavorful way is like follow the money, right?

With all of these fields, like follow the money and follow the data. Where do we have a lot of money and a lot of data? Healthcare is definitely number one. And that's the one that everyone's going to spend because, hey, that has only positive outcomes, right?

Like, hey, we're curing cancer. Hey, we're identifying diseases earlier. Better than doctors at identifying stuff. But doctors are still there to treat you.

Like, those are all good things, right? No one wants to talk about like, hey, we're finding out how to best, you know, assault this, you know, region and, you know, prevent deaths of however many people by killing this many people like ahead of time. I'm not saying that's where it's going, but the way I see machine learning, it goes hand in hand with AI, right? Like, we're training these networks or these models and like this artificial intelligence to identify certain things.

So with that, we're talking about very specific things, right? Hey, you're an expert in your field, you're a jack of this trade and a master of none, right? Or you're a master in this trade and a jack of all trades, right? Isn't that how it goes?

I totally butchered the saying, I'm sorry, a jack of all trades, master of none. This one is very, you know, single focused. But so where it's going in the future, I'll just say, like, if you want to look down the bleak future, just look at movies like Minority Report, The Matrix Terminator, where, you know, AI takes over and identifies that humans are stupid and can kill all of us, which, hey, good job AI. And then the other positive ones is like, look ahead, movies like her and what are some other positive ones?

Holy crap. They're all negative. They're all negative. I know.

A lot of them are overwhelmingly, you know, the more I feed this model is like, yeah, it's pretty negative. It's a pretty bleak outlook. It's a flowery, you know, movie about the advancement of the human race. There's got to be conflict.

You know, it's the whole point of a movie. And I love how I named like five movies that you've probably never seen. I only have seen, was it, you listed Minority Report? I'm pretty sure it's that one, or maybe you mentioned some other one.

There's one that's a scene that I think I triggered that it came back to me. I only remember one scene where he walks into like, I'm like a lobby or something and there's an intense amount of gunfire that happens. Like, I don't know, 3,000 bullet shells go off and I just remember the scene really well because it's like, it's just, I just remember seeing the massive amount of gunfire that happens and I'm like, wow, okay. I don't think I've ever seen a movie where there's that much gunfire in one room, so.

So many people are screaming into the air right now, just screaming. Do you know what it is? Yeah, so Matrix. Oh, come on.

Do I know what it is? Like, what are you talking about? Of course. Yeah, the lobby scene is pretty intense.

Minority Report is here. Okay, spoiler alert from Minority Report. Like, sound of spoiler horn right now. Minority Report is all about predicting crime, right?

And they actually get to the point where they can identify with precognition. It's not really machine learning. It's more like, you know, sci-fi, like, someone can see the future, but they basically identify these murders before they happen. So there's a division of the police force called pre-crime and where they basically go and arrest people before they commit a murder and, you know, charge them with being murderers by saying, like, hey, you were going to commit a murder on this date at this time.

You know, it's pre-meditated, so you're under arrest. So it's by Seaman Spielberg. It's really good. It's got Tom Cruise in it and it's a good action movie kind of sci-fi with kind of those undertones of like, is your destiny predetermined or do you have choice, like, are you in charge of your own destiny?

Are you a Tom Cruise fan? Oh, totally. Yeah. He's my celebrity doppelganger.

I feel like he's a very like, I know obviously I know a little about movies, but I feel like he's one of the actors that's like, most people either love him or hate him. Well, you have to love Tom Cruise as an actor. I mean, you have to separate the actor from the person, right? Because Tom Cruise as a person, well, let's just say he's a person.

Tom Cruise as an actor. Wow, he's a great actor. He's like, he's putting some pretty crappy movies sometimes, but he can make some action movies and he can pull off his own stunts pretty well. So you got to give him props for props or do, but Tom Cruise as a person just pretend that he doesn't exist.

Just pretend he's only Tom Cruise in the movies. Whatever his character is. That's who Tom Cruise is. Okay.

All right. Cool. Well, is there anything else you wanted to touch on with a machine learning before we wrap it up? Nope.

Okay. And I think, I think I'll spend the positive on this one. Let's end on the positive note. It's like, Hey, there are some great advancements.

It's going to, you know, help us out a lot just in terms of our daily lives, anything from driving into work every day or just unlocking your phone in the morning. So machine learning is helping you. It's on your side. Yeah, not to get like too meta, but I'm pretty confident there's machine learning computers or algorithms that are, you know, smarter than us about machine learning already.

So they've been feeding them models for years and they know a lot more about machine learning than we even know. Hey, this guy call was probably also brought to you by machine learning as well. Oh, yes. Thanks Microsoft.

Cool. So in a little bit of wrap up, I know I mentioned that I wanted to give yoga a try and so I'm kind of like one foot in on yoga and I say I'm one foot in because I have to give a little like clarity here. So I bought a yoga mat. Yes, it's a little lemon.

I'm just going to like curve that right now, but I'm kind of returning yoga mat. Wait, wait, you're returning the yoga mat? Yes. Yes, yes.

So you bought a yoga mat and you're returning the yoga mat? Yes, I'm returning the yoga mat because someone I know has a discount at the lemon because I guess a little lemon does discounts if you're like a yoga instructor. And so someone I know who's the yoga instructor gets like 15 or 20% off or something like that. I'm like, dude, hell yeah, I'm going to take advantage of that and save my money.

Just like Adam would do. Adam taught me well that way. So I'm just going to turn this is unused so far I got it, but I like it hasn't been enrolled yet. So I didn't even like, you know, take it apart and actually like, you know, use it yet, but I'm going to like take it back to them.

Be like, hi guys, cool. Here's the one back. And yes, I'd like to order another one and on that person to see out right there. So let's say I probably have to wait another week to get it because I don't think they'll do that one.

But you said you're one fit in on yoga. And I'm like, oh, okay, cool. So you've actually done a practice. But no, you're going out and buying a mat.

I think you said you had access to like a map that you could use just for one practice. I do. I do. I definitely do.

I could totally use one, but it's like, I don't know. It's definitely too small. So I mean, I guess I mean, I could suffer through it and like be okay. Like it's not the end of the world.

Yeah. You're just finding out if you like it. I mean this. Okay.

You said like, oh, Adam taught me well this way. No, no, no, you're not listening to Adam right now. Yes. You might have found something on sale.

That doesn't mean return something and, you know, buy something that you're potentially not going to use. Man, free trials, free trials. You have a free trial. You have a free trial yoga mat right down and you're not taking advantage of it.

Adam's free trial is like, go grab a bath towel, put it down. Oh, seriously. I mean, like do it on carpet. Like you don't have to have a yoga mat to do it.

Go find like an exercise mat. Right? Like there's so many things you could do just to find out. Yeah.

I don't get me wrong. I am a little nervous about like, I already like open the app and like looked at the, um, um, sound like the walkthrough roses. Okay. Like even in the beginning, I'm like, okay.

There's like, there's like, you know, downward dog. And like a couple of those that I'm like, okay. Like I've seen those before. I don't even know the names of the other ones where I've like seen people do some of these and I'm like, okay.

Okay. Like those don't look too bad. There's a couple like in the transitional phases. I was like, oh, I don't know how you transition from that to that.

And it doesn't just get weird real quick. So we'll see. There's going to be some learning for sure. I just finished my practice, you know, before the recording too.

So my little $20 costume yoga mat has been working out just fine. Yes. Yes. So I will, it'll probably be like another two weeks, but I will get my first one in hopefully in just a couple of weeks.

So I'm one foot in, as I said, but I'm kind of, I feel like the toes are in within a whole foot. You barely, you barely did your toe on this water. That is yoga. You barely did your toe on this bond.

I want to edit the fucking pages outline, but pages like, Oh, no, we gotta fucking update. And I'm like, what the hell, man? Like you won't even let me freaking edit anything like a half. Yes.

Yes. Yes. You're buying again and you can't. I'm so happy right now.

I was like, I was trying to edit stuff like as you were talking to. I'm like, okay, like I want to change this. I can't do anything. Like, uh, you have no idea how excited this makes me because you and you're like, Oh, I didn't do my updates.

So this is a day zero update. You can't, you can't give me shit for a day zero update. I mean, like I did my yoga practice and then I came to the computer and I started recording with you. Like you can't give me crap.

Like I wasn't going to risk like downloading anything during the podcast because whoa, you know, this guy gets a little antsy whenever I start trying to use too much bandwidth. So yeah, I'm sure you're loving over there because this is just the story of my life where I can't use fucking pages to edit anything because I'm two versions behind. Yes. Yes.

What's my uptime right now? Now I have to look at my uptime. It's only 10 days. Oh, oh, I probably did do an update.

Not bad. No, it's 10 days. It doesn't actually isn't great. It's a pretty bad.

You're with the hate mind. Your uptime is like four, four hours. Uh, let's see here, uh, 22 hours and nine minutes. Not even a day.

Nope. It's less than one day.