In the last video we examined that general product innovation doesn't seem to be getting much help from AI, and data analytics, okay. Where processed improvement the process innovation seem to get a lot of help from data analytics. In this video, we're going to deep dive into product innovation to see are there any differences between different types of product innovation and if so how our AI affecting them differently? So remember I showed a lot of examples about how AI can fundamentally change many industries from health care, to drug discovery, to product design of cars, and to the realms of arts and music. So but notice how these innovations have in common. In Watson health example, we have identified the P53 but there are already many examples of P53 out there and also other proteins that interact with the P53. So AI indeed algorithms have lots of examples to look at to figure out what a new P53 protein may look like. Okay, in terms of how well seen, it's brand new antibiotics, they've never been discovered before, and it's obviously a very important discovery, but there are also many other antibiotics out there. So we know what antibiotics can do or should do when we find one is easier to verify. But again, Halicin is not the very first antibiotics, it's out there, it's not penicillin. Car chassis is again a great example improving an existing product. This is chassis. It's very innovative. We probably wouldn't have seen one without the use of AI to tell us about a pattern about different ways that car's turn one direction versus the other. Okay, but again, fundamentally is still just a chassis. And if you look at the paintings, you see that you took a picture from Europe, and then you combine it with Van Gogh's Starry Night or any other famous paintings and you may want to put it in your wall, okay? But notice they're all have something in common in the sense they are recombination innovation. So what do I mean by recombination innovation? That means you might be marrying two different things together to create a new thing. Okay, so you can think about the photos and the Starry Night is a great example. You're literally to putting two things together, a photo and the Starry Night together. And that gave you a new painting, is very beautiful. It's very innovative but it's a combination of two things. Summer was chassis, right? It is still improvement of existing chassis, a chassis marrying sensor, marrying many other technologies and analytics that help discover improved chassis. Okay, how was seen again is by mining lots of lots of data points and lots of lots of research papers about antibiotics but also many other drug interactions out there. Okay find halicin. See that might be very difficult to discover before. But it's still fundamentally antibiotics. Not a first penicillin is by combining information from different disciplines, different source of information, and then put it together just to figure out what halicin may look like. The same cells they can go with P53 proteins again, you already know what to look for in P53 but you just need to find out the patterns. So in a sense, AI analytics can tragically expand the search space for you to find the innovation. Okay, by doing the linkage to detect hidden patterns among all these research papers, all these data points, okay, we're able to find new ways to combine existing technologies in a new way. And that is fundamentally a recombination innovation and it's very powerful form of innovation. So I just gave you intuition on the type of product that AI can help you support. Okay, not only can AI can help you combine things in a new way. They can also do so by combining many different things in a new way. They combine at work, combination is something AI is really good at. So maybe our in our mind we can do combination with three or four or five elements. But AI can do that in hundred thousands. A much more effective way to figure out the combinatorics combination to figure out what works and what doesn't. Okay, so let me give an example, what does it mean by a diversity recombination? What does it mean by going through combinatorics? So for example, if you have four existing technologies, A, B and C, and BC. So A, B and C are technology belong to its own class. They're very different from each other. And BC is a technology that combined B and C together and a diverse combination would be something like ABC. That you combine all four technologies, three technology together to form a diverse combination, okay. And what I call a very new and potentially radical innovation called D. Okay, it's a technology class that has not existed before. So our hypothesis is that AI can greatly help you to find a diverse recombination, combining many different elements together in a new way. Could be ABC, BCA, etcetera in different way, different orientations. But you may not help you find that first in the class innovation. So it's great in finding halicin, but it may not be great at helping you find penicillin the very first antibiotics out there. That means a technology class that does not exist before. If you think about it the technology that does not exist before doesn't have much data about them. Very limited in terms of what AI and data learners can do to help you find that radical new innovation AD. But it is very good at finding innovation that combine different things together in a new way, a diverse recombination. So how do we know that our intuition is correct? That AI and data analytics can facilitates a diverse recombination innovation. One way to do that is using patents, patents has several vantages, first that patent has a specific classification code telling you whether it is new or not in a sense, the first pattern showing up in that particular technology class and it's very novel. Furthermore, we also know the patent that is current patent cited. So in this red box here, we look at all the patents. Okay, that's a particular patent has cited. And it's important because this is all the prior art that this particular patent is drawing from. For example, if a patent has 10 citations, all belong to the same patent class, then we know it's not a diverse combination, right? Coming from a single technology class, okay. But the same 10 patents are coming from 10 different technology classes, then it's likely that this pattern is more of a diverse combination because it combined 10 technology elements together, okay. And that is a diverse combination. Furthermore, to measure novelty, we can also look at the abstract of the patent to look at whether the words that uses very new. For example, if the first time the word HTML showed up on abstract, that's a very new word. That means it's patent is probably very novel. Okay, if this patent is the 1000's patent I mentioned towards HTML, then it's not a very novel patent relatively to the first one. So, this is how we use patent data and classify each patent into a diverse combination in sense. Is it a combination of many different technology classes together or is it a very new technology while there is being the first of its own technology class or they have used words or so novel that have not seen in other patents before. Okay then we examine the relationship between data analytics and AI investments to this two different type of product innovation. Number one being diverse combination. Number two, whether it's a novel technology. As you see in this graph here, you can see that if you look at the first one is diverse combination and the Y axis is the likelihood of investing in AI technologies and another technologies. And if you look at diverse combination, there are far more likely to adopt diverse combinations and the effects is pretty large and these the error is very small. But if you look at the corresponding graph for novel technologies, remember this is the first of his own kind. It's not a combination like ABC together but a brand new technology because they haven't seen before. In fact, we see it actually almost a negative relationship between novel technology and AI and less investment, although it's virtually zero because so in ways that we can't really measure effectively. Okay, then let's look at the productivity effect. Do firms that tend to combine existing technology in new way and when they adopt AI analytics, do they correspondently see a gain in productivity? Okay, and we do that same thing with whether this firm is oriented very new technology, okay. And we see the first bar in this graph we can see it's the fact is dramatic the firm that invested AI. And also we're interested in combining technologies, existing technologies in a new way seem much biggest effect. This is a about 1%, okay. But 1% change in productivity is huge given that the fact of sheer AI is very small. So it's actually a huge effect. And if you see the same correspondent change on D remember this is a novel technology that we haven't seen before. Okay, and the fact is almost negative although it's slightly hard to tell but at least more zero. Okay, and we see that this is not an IT phenomenon. This is really about a AI phenomenon because we replicate this graph for IT replacing AI technology measurement was just generating. So next to graph the bars you can see that this effect does not happen when you see a diverse combination investing in general anti technologies. Okay, in fact that's almost going the opposite direction than AI. And similar with novel technology. You also don't see that in general IT investment. So I'm showing here is really about AI driving diverse recombination of existing technologies. Okay, and the firm that do both, but they use AI and they use it to find diverse combination of existing technologies. Okay, that's when they receive a dramatic increase in productivity. On the other hand if you use AI and your primarily using to find a novel technology class, the first zone kind effect doesn't seem to be there. Similarly you can't just invest in general IT is really about AI that is driving the fact we're seeing here. So main takeaway here is that we find that data analytics and AI support diverse recombination of these technologies such as ABC, okay. Three element together, okay. But analytics and AI do not support a radically new technology such as D for his own kind. And we find that firm invest in analytics and AI and they focus on recombination innovation. There are a lot more productive than firms that do not invest in analytics by itself or firm that are interesting recombination but do not invest in analytics. So it's really the combination that matters most, okay that you invest in AI and you're using it to find recombinations. So that could be another reason why the paradox exists, if you using AI to find novel new technologies such as D. That effort may not be as effective as you use the technology investment to find recombination innovation.