How Smart Can Artificial Intelligence Be?

Welcome everyone to another episode of the... I should turn off my notifications. All right, notifications are off, so welcome again to another episode of the Product Bakery. Hi Christian. Hi Alex, thanks for turning the notifications off. It's a problem we all have, right? Yeah, the computer could be smarter in terms of understanding when I... I mean, honestly, it shouldn't be that hard, right? The computer knows whenever my microphone is on. Why would it potentially play notifications? It would be also cool to have an app where I can say, okay, every Thursday or Monday or whenever we record, mute everything. Yeah, no, but that's the pain of setting something like that up. Yeah, true. Would it understand by itself, it would be even cooler. Of course. Shouldn't computers take over the world? Shouldn't they guide us through our lives? They anyway know more than we do. All those things that can be done manually can be also outsourced to machine usually, it's true. So I have the feeling we should think about a product connected with an AI that does all this stuff for us. Yes, it could potentially do that. But at the same time, I'm always a little bit afraid of where AI could go. Because let's be honest, we talk about AI and AI is probably one of the keywords. I don't know how many people I hear talking about it or news or whatever. Open products hunt, right? You have tools that write your blog articles in AI. Oh, yeah, true. What was the name again? I just checked it out recently. I don't remember. And we should anyway, but I'm not going to advertise. So the domain is there, conversion.ai or javis.ai. That's the one. But that's the thing, right? Like literally, you can do pretty much everything now with AI. And it is impressive. I tried it, especially this writing thing. I tried it with the company that I'm currently working for. And it's not that our business model is the easiest one, or it's not it's very specific, it's very niche. And it's enough for me to write in company name, private equity, and so on. I press the button, and they generate everything that you would need from landing page copy to social media copy, and so on. And if you read it, it actually makes sense. Yeah, that's impressive. And what's even crazier, it gets better and more accurate every day. Yes, but, and this is where everything comes in. Artificial intelligence isn't so intelligent, right? It's simply learning from what humans did in the past. So that article that's written well, is simply learning from all the articles that someone wrote in the past. It's like taking the pieces together, and it's learning from masses of data that we collected over the last couple of years. And I think that's actually also introducing a little bit like, of a danger. And it always depends on where you use it. I think like one of the most common examples, and you've probably also read about those things is like, for example, when Apple launched their credit card, right? And suddenly, someone with the same bank account, the same entries, and so on, than his wife, got a much higher credit score than the wife. Why is that? Because the data that the machine got to learn from was like historically biased, right? There are old biases in place. And they can be, to some extent, racist, they can be to some extent, sexist and everything. I think the point is that AI just simply cannot be neutral. Yeah, I agree. And I used to be very skeptical as well when it comes to AI. But I recently read an interesting book, which is called Competing in the Age of AI. And I like the way this whole topic of artificial intelligence was broken down, because the author distinguishes between strong and weak AI. A strong AI has more the notion of trying to replace a human being with all its logic and decision making. And you have also the weak AI, which focuses more on simple things that make our lives easier. For example, auto suggestions on Amazon, right? Everything is based on artificial intelligence. If you look at your Netflix account, everything is based on an algorithm that learns the stuff that you like. So I think especially that part is something that is quite cool. And for sure, it can go into the wrong direction if you feed it with wrong historical bad data. But on the other hand, I see also a lot of benefits because these days, it's almost impossible to think about living without an AI. True. You see it more and more popping up. And even as an example, let's say you are a bakery, we are a bakery. And if you would just use a simple AI that makes a forecast of our sales based on the season, based on the weather, it would help us already to not produce too much waste when we're going to bake our breads and make sure we sell the optimal amount. So therefore, I see a trend to that direction. But as you said, also a lot of danger because the human work gets more and more replaced. Yeah. And I think we can't ignore that it's going to be a very important thing for the future, right? It will be everywhere. There is massive positive benefits that you can get out of it. I think it's just more, and this is where it becomes also relevant for us. It's about how we design, train those solutions. It's about the team that works on these things. At the end, we have the power to tell the machines what's right or wrong. The machine will not get there and will not be able to come up with that on their own. It's always like human trains, human programs. So it's our responsibility to do it the right way. It's our responsibility to make sure that we avoid certain things. I think that's a very good point. And if we take back your example from this financial score that was happening when the Apple Card was launched, I think it's very important as a product team these days to make sure that you make use of the AI and its benefits, but also not losing the human aspect. And a very simple example that I'm regularly hearing about is people trying to get their consumer credit. So they want to have a loan, they need some money, and they're just filling out some forms. And then an AI decides whether you are worthy or not. But the problem is there are some people and I know- It's the human behind the AI that decides if the AI tells you if you're- Yes and no. And I want to tell you why. So if we just look at the blank data, because I know a guy who was paying his mortgage for 15 years and he fucked up one payment. And based on that, his score went down and he wasn't able to get any credit because all the AIs and algorithms said no. But the thing is, he paid 15 years every month and it was just like a payment that went wrong. It wasn't like that he hasn't had the money to pay it. And the thing is, the AIs were saying no, but there was no further step for him to reach out to a human being and ask for a manual review on his data. And I think here it comes in place that it's very important to not only build the machine that makes those decisions, but also giving the human being that is using or the consumer of it, the chance to still interact with human beings and not only the machine. It's an interesting way to look about it. So you're talking about the support systems that we put in place to fix this. First reaction that came to my mind, especially on this case, I would assume as a company, if I take this process in place, I cannot skip all the positive processes, all the people that are automatically accepted. But I think everyone would fight back on a decision where I don't get any mortgage. So in this case, where the AI tries to actually minimize the operational effort on the humans, you're reintroducing it. And you probably need to get a full team because everyone who gets a no will ask them. I hope that, but I also know that some systems simply reject you. So you get a rejection. It tells you your score is bad and some people just taking it. But the problem is that these people are excluded and maybe 80%. Who would take it if they have a way to contact? If you have to wait, for sure. But the question is if you go to, I don't know, the E in Germany. So you have the ability to fill out some forms, but the contact form is usually hidden. So yeah, for sure you have to wait, but the question is who would really fight it or who would really then go that path or rather just walk down to the next bank and trying to do it physically over there. I don't know. But at least you have someone in front of you, you can blame. What I'm referring to is you are absolutely right. If people have to wait, I hope they will go that way. What I want to emphasize is that you make sure when you build those products, that you don't lose this human aspect and leave everything up to machines. Yeah, I agree. But I think it comes down to do I want to give people the option, right? Because I introduced the AI in the first place because I don't want to have to manual work. So do I want everyone who gets rejected by the software that I wrote to get back in contact and do a manual check? Yes, I agree. You should have this in place. And here comes the ethical discussion, actually. So it's not easy to decide. It depends on many factors. But the question is, let's take the worst case. Let's take the worst case and most of the companies don't allow this anymore. What will be the price that we have to pay as consumers? And I believe it can be, and then we're going back to the discrimination. If you're not getting hurt, if you're not getting any chance to get a loan, for example. So we see already a big shift within society and people who are getting excluded from certain things they maybe weren't before. I think what you're describing is a little bit of workaround to trying to make sure to use the right data, to train the tools. And I think you can talk a lot about different ways of doing it. The easiest one, probably, to start off right, is to start off right with a diverse team. If you only have white rich guys in one room, they will probably not understand what it means to actually not having paid the mortgage for one month, for whatever reason. I agree. If you work on a face recognition software, and again, you have a bunch of white guys developing the software and they are training it with their own faces, guess what's the outcome, what the outcome will be? And we had scandals over and over again of Google recognizing black people as monkeys. That's fucking ridiculous. And that's so easy to be solved by simply having the people in the room that represent the whole population. Agreed. You need to start with good data, obviously. But even then, an algorithm has a likelihood to be right, but it's still a likelihood. And the question is, how much can we leave up to the machines? And how likely will they be right? Because there's always a chance that they won't, even though we give them the good data. And the moment they are wrong, we have a problem. And I think you also need to invest a lot in stress testing, revisiting, refining, and yeah, you need to have a way to identify maybe the false positives or false negatives, so that then the software can also start learning again. If you pull your friend back into the pool as someone who's actually eligible for a new mortgage, guess what? The AI, ideally, would also learn that specific case. And then the good thing is that obviously, if you do it right, if you correct your failures, then the AI would automatically learn. And then you could solve a couple of these issues. Exactly. And it is not an MVP, and it's nothing that you only build once. It's definitely something you have to iterate upon on a regular basis. The definition of an MVP is that, I mean, in theory, by definition, you should iterate on top of it. I think it's a misperception of many business and product people that you build a MVP and then you don't look at it for the next couple of years. Yes. Let's say you have introduced an AI that makes your life easier, and your boss is happy with that and pushes towards the next feature that needs to be built. It happens quite fast that you leave it at the stage where it was. But good. I think it's a good reminder. And I think the whole topic of the development of the AI is just super interesting. And I'm curious to see in which direction it's heading. Absolutely. And with that, let's keep an eye on it. Let's take it up again in the future. Everyone have a beautiful day.