Tech Talk: "Reinventing AI" mit Philipp Wissgott von danube.ai

Tech Talk: "Reinventing AI" mit Philipp Wissgott von danube.ai

Hello, my name is Phillip Wisgott from danube.ai. Markus, thank you very much for the invitation to this TechTalk! It's really a pleasure to be here. danube.ai is a decision engine which symbolizes for us the next logical step after search engines like Google. For our decision engine we had to create a completely new form of artificial intelligence. A form that has nothing to do with machine learning. In this talk I want to tell you why we did it and how we did that, but first let's start off with a little history.

The younger ones of you might not know anymore but there was a time where Google was not the go-to search engine. On the internet there were two major companies called Yahoo! and Altavista considering the internet was very small at the end of the 90s. So when you tried to look something up, you just went on their pages and typed in the search field what you're looking for. The technology was pretty simple so it just searched through all the internet pages for the stuff that you entered in the search field. And they were pretty good, they were big, they were famous and then there was a tiny company coming to them called Google and they said: "Hey, look, we have a new technology, we've changed the way searching works, with a hierarchical algorithm that can search the internet faster and in a better way to personalize the search results in a unique way, so let's do something together, let's partner up, let's work together. You have the audience and we have the technology."

Yahoo! said: "Well, actually everything works well the way it is, we don't have any performance issues with our technology. Actually we don't like to collaborate." And everyone of us knows what came in the next 20 years. Nowadays Google is the biggest company on the planet and the younger ones of you probably don't even know about Yahoo! and Altavista anymore. However after 20 years of search engines and doing many many things with Google, some of you might ask yourself: Is there something that goes beyond search engines, that is the next step after Google? That's also the question that we asked ourselves so we analyzed Google because they're the most successful when it comes to search engines.

The first thing we did is look for the reason why Google is that successful. So Google is the biggest company and when you want to search something, you type it in the search bar and for example I just typed in: "I want to buy a car" or "I want to get a new car". The first thing I noticed was that Google gets better with every search, so every time someone enters something in the search bar, the next searches after that will automatically become better. The services of Google become better, that's actually quite a cool thing right?

Moreover, it customizes, it personalizes on a very high level, we have to admit. Regarding the location, when you want to buy a car for example it shows you a car store, a car shop close by but it doesn't tell you anything about what kind of cars you can buy there. Even though that's actually something that you would like to know or respectfully a kind of personalization you would like to have. Nonetheless, still it's quite cool when you are on the street and you're looking for a restaurant: "I'm very hungry, show me the next restaurant". A search engine is just the coolest thing to be.

When you try to see the bigger picture, it's the many microservices that made our lives a lot better and through that we all became a lot faster in doing things. I just went here from Vienna to Linz and I just used Google maps for routing and many many other things like calendar and the search function. Overall, Google is improving a lot of people's lives little by little. When you add up all the people that's the point where they are really making a difference. This goes along with the power, the business model and the technology of Google. That's on the good side.

On the other hand there are some flaws with Google's technology and if you look closely the next time you search something maybe you will notice them. The first thing is: The search field and the results are just a big filter which makes it into more of a data analytic instrument. That's how they can actually cope with this huge amount of data considering the internet is a lot bigger compared to 20 years ago. Working with that amount of data was the main technological task for Google even though it's not so much an Artificial Intelligence thing, it's actually rather a filter to some extent. Everyone of us knows that when something happens to come on the second page of search results, you would just never see it because you simply don't have the time to click on the second or the third page of the search result. Meaning, what Google brings on the first page is very decisive. The downside is that it's not really personalized in order to help you out.

For example when you're trying to look into buying a new car Google suggests you five tips for buying a car the smart way. So it's not an opportunity to buy a certain car, it's a way to become an expert in car buying and this can be referred to all consumer related questions. According to Google you have to become an expert in TVs when you buy a TV, you have to become an expert in laptops when you buy a laptop and we just don't have this time anymore. We make so many consumer decisions every day and every week. If we look on the second, third page of Google or try to become an expert in those things, then we just don't have any more time left to do anything else. That somehow reassembles the technological gap that opens up for future technologies to go the next step. How we translate it: Google is a search engine, so what it does is to finds something for you, so when you're looking for something very particular, it can find it very quickly and very accurately. But on the other hand it doesn't help you decide. However many things in life actually are about making the right decision and that's why we think that decision engines are actually the next big thing after search engines.

So how can it look like? Speaking of, what I'm talking right now is a little bit in the future, also for our company. So if you go on our website and look for the search bar, it won't be there yet. We are just doing it for our customers right now in the background but we are working to get this general service ready for everyone. Let's just imagine what it will be like when we have completed the decision engine and you can type in: "Which car should I buy" and it will automatically take your personal preferences into account, so for example: The kind of lifestyle you live, where and what kind of vacation you typically go on. This will just be correlated with the way what kind of cars you like, what kind of car brands you like and also the easiest way, just what kind of colors you prefer, it will be filtering out a lot of other colors, that you don't like, that you don't need to be recommended.

And when you type it in and click on the "tell me" button then our vision is that you get just five results of cars you can test drive in the shop around the corner in one hour and also already have an appointment assigned the moment you click on it. "I want to test drive this particular car", so it gets a lot faster and it's not kind of a full optimization in an Artificial Intelligence but rather kind of a semi optimization that's filtering out and recommending stuff.

But still you make the final decision out of a reduced selection of suggestions. Another thing is: You might say: Oh my God, the algorithm, the Artificial Intelligence knows where I'm going on holiday and all the other stuff, so the data must be everywhere. But with danube.ai it won't, you'll be able to choose where your data is, so you can say: I just like all my data with me, then it will stay in a cookie at your place or just in a local storage in your browser. On the other hand, you can of course also make an account, that's up to you, However if you choose to keep your data then we won't use it so it just stays with you. Now you might ask yourself, how we want to achieve that technically. Bear with me, I will come to that and it will then come all together at a later stage of my talk.

So that's the vision and we've come a long way. Now I want to tell you something about how we actually tackled the problem and accomplished it with the master plan of Artificial Intelligence. We've looked at the way humans do it, how the human brain is working and how we make decisions. Our brain works so fast you usually don't notice the decision-making process, you just end up with a gut feeling of the final decision. But these kind of steps happen in all of us every day, thousands of times probably.

Firstly, you collect all the memories starting from the first thing you remember and obviously that's just a lot. That's just like the internet in comparison. Next, you filter out all non relevant stuff.

For example you filter out all the stuff that has nothing to do with cars. However memories about you driving a car or a car breaking down somewhere might be important for your decision so that stays in the data pool. The first step is you create a pro and con list. A car breaking down on holiday in Italy, very far from the next place you can repair it might influence you to never buy that car brand again and I won't mention any brands right now. And that's just a very important factor on your pro and con list.

The next thing step is to weigh the pros and cons. Breaking down in Italy might be much more crucial than a handbrake not working for once. But it's not just that. Not all pros and cons have the same weight, so you have to evaluate them. At last, the final step is to sort your alternatives or your options according to your personal costs and your personal value with the emphasis on "personal". To state an example: You're a rich person and you have a car worth € 1000 or more. In this case you might not care so much about the difference between the suggested options. But if you don't have that much money, one thousand might be a huge sum influencing your decision. The same goes with personal value. If you like a good audio system in your car, those things have to be taken into account.

And so we made it our mission to automate the human process of decision-making in our decision engine. With danube.ai we've tackled the steps 3 to 5 (create a pro/con list, weight the pros and cons, sort according to cost for value). The first two steps (collect all options, filter out non-relevant options) are what Google has been doing for the last 20 years and I'm not saying it's something small. Doing these first two steps with that amount of data is actually a huge endeavor.

After solving the points 3 to 5, we also have to tackle the points 1 to 2 at some point. We're just at our stage right now. We didn't want to copy Google for good and it also wouldn't make sense to us. When we build the decision engine for our customers, the data comes usually from them so it's kind of a pre-processing step. Then we usually start with the great pro and con list. And that results in our Artificial Intelligence algorithm so we can create a very personalized decision engine that just works for our customers and for us, that's the next step after search engines.

I've already switched a little from the why to the how. Let's go into more detail about the how. I told you that danube.ai is not based on machine learning. That's a very important thing that constitutes the foundation of this algorithm and what comes afterwards in technical terms which makes danube.ai a multi-dimensional sorting algorithm that has nothing to do with machine learning.

It follows completely other philosophies and does certain steps at different times than machine learning does. In machine learning you have one central server that is just collecting data and you just ask the server: "Please give me a recommendation or please give me some pattern recognition or something". In our case it's not like this, it's the Artificial Intelligence that is actually going from the server to you, so every user will have a different kind of Artificial Intelligence and this AI itself is created at the very moment when you press the button when you want to have a recommendation.

Let's move away from the car example. Just imagine one evening, you want to watch something and enter Netflix and in the future danube.ai integration in Netflix tells you what you would like to watch. At that very moment when you press the button the Artificial Intelligence is created and everyone's user experience will differ and it will be different from yesterday's version, today's version and tomorrow's version. So that's actually the difference between the learning phases.

So, when you have machine learning, you usually have a learning phase where you have a data set completed and you just say: "Ok with this data I'll learn and then you go into kind of a live phase where you keep learning but there has to be some kind of starting phase where you build the Artificial Intelligence model."

In our case that's a little bit different and that also gives us unique capabilities in terms of technology and how we solve the problem. In terms of Netflix, as you might have noticed, when you watch a cooking show, the recommendations are based on content similarity. So what it will do is recommend one cooking show after another but that's not the way we watch. I mean, at least not me but realistically, after this cooking show, everyone will want to watch this new action movie everyone is talking about and that's just correlated. Artificial Intelligence doesn't know about it yet, but the users know that, because they just want to watch what they like. So what danube.ai actually does is look at the user behavior only and not on the content similarity.

So it showcases that I am 5 percent similar to Markus, another 5 percent similar to you and so it collects the similarities of all users and predicts how much you are similar to all the other users. Then it continues to create a personal profile and in turn create the personal Artificial Intelligence just for you. Not surprisingly, we don't need big data for it to work. After the first, the personalization gets rapidly better with every click.

If we integrate danube.ai on Netflix, after the first movie, the second recommendation will already be personalized for you. Continuing, this personalization will become even better with every third, fourth, fifth movie or series you watch and this matching of the recommendations to your personal likes will increase exponentially. With this technology we have a huge opportunity to reach into applications that have not been covered by Artificial Intelligence yet. And so this enables us to build our decision engines for our customers.

All this is not my own story but it's the story of our team. We've all worked together to do this in the last three years and now we're really ready to move on to the next step. It has really been a pleasure for me to speak in front of you today. Thank you very much.

 

 

Erfahre mehr zum DevTeam von danube.ai