My name is Lynn Wu. I'm Associate Professor at the Wharton School with the Operation, Information, Decisions Department. My expertise lies on AI's effects on organizations and specifically from innovations perspective. In this video, we're going to talk about whether AI and data analytics is suited for innovation. You've probably heard a lot of interesting news in the last couple of years about AI-driven innovation. Probably the most famous one was in 2020. AI helped create a new antibiotic called Halicin. This is a big breakthrough because antibiotics, especially new antibiotics, are extremely difficult to find. It has become a big public health issue as the superbugs are becoming a big problem. About 2020, a bunch of MIT research scientists have leveraged data about all kinds of data compound; the antibacterial activities, et cetera. After screening more than 100 million compounds, they were able to discern new patterns on the data that we haven't seen before. As a result, they were able to identify new antibiotic candidate in just about three days. Halicin is unique in several aspects. First, it's discovered through AI platforms with aid of scientists. Number 2, not only is it very effective, the Halicin is also distinct from other types of antibiotics in its chemical structure and is also tolerated in human body. Couple of years ago, IBM Watson has famously used their technology to discover new type of health care innovation as well. In this case, Watson was able to digest about 23 million medical papers across many different disciplines to find information about a tumor suppressor known as p53. P53 is important because it is associated with half of almost all cancers. In a short amount of time, Watson was able to digest all these papers, finding hidden patterns in these data, and to identify six previously unknown proteins that interact with p53. This feat would have taken researchers more than six years to accomplish what Watson did in just under a few weeks. This is not just in health care industry, not just in drugs, not just in compound discovery, this is also in product design. For example, Autodesk was trying to design a new type of chassis. What they did is, they added many sensors to a chassis that can measure stresses, the strains, the temperatures, displacements, and all the other things that are associated with chassis, and to make chassis work. All these sensors are very cheap and they can deploy many sensors on this chassis. After the sensors were installed, a stunt car driver then drove this car and really tried to push the system. They can accelerate really hard, they break very hard, steer as hard as possible. After about 20 million data points, AI, with aid of scientists, came up with a new chassis. If you look at this picture of the chassis, it doesn't really look like a car, it looks like a bone of a mammoth. But notice the asymmetry in this car. The left-hand side and the right-hand side are not mirrors of each other. Typically, you see the left-hand side and right-hand side of the car to be very much the same. This is because if you look at his data, it was found that the car turns much more often in one direction than the other. Of course, one side will have to tolerate more strains, more stress, and more displacements than the other side, and that's why you have asymmetry before. Even in the realms of arts and science, which is what we think as very human act, we see that AI can produce art that we may be interested to buy and hang in our living room wall. If you take a look at this photo on top, it's a beautiful photo, and then you combine it with Starry Night from Van Gogh, and you have a beautiful juxtaposition of these two photos and then create a brand new photo that has element of the Starry Night. You can do that for many other famous paintings, putting your photo and combine it with another famous painting. AI can combine these elements to produce a beautiful painting that I personally would choose to hang in my living room wall. I just gave you some great examples about AI-driven innovation. However, if you look at this graph produced by Nicholas Bloom and co-authors in 2017, you see a different picture. Some have argued that not only have we not seen great innovations coming out, but on average, we have seen innovation decline. This is how you read this graph. The green lines you can think of as how much money or resources were invested in R&D. The blue line you can think of it is per capita spent, basically how much innovation per capita has been produced. You see that we have spent a lot more money and resources measured by effective number of researchers in the green line. But if you look at the productivity from that resource investment, the blue line has shown that our per capita innovation outcome has gone down quite a bit. This is a pretty long trend from 1930-2000. The reason data suggests this trend has continued. What is going on here? Why do we see so many AI-driven innovation, but somehow they do not show up in our productivity and innovation statistics? This is a very interesting phenomena and this get a central question of what AI can do and cannot do for innovation. In our next video, I'm going to explain what AI is suitable for and not suitable for, for various type of innovation activities.
