﻿     Artificial intelligence as a branch of computer science has been developed for several decades, though, it is only recently that AI received great attention from the public, and many people, still have limited understanding of what AI is and what AI is capable of. I suppose people may not have that large demand for some of the future AI applications anticipated in “One Hundred Year Study on Artificial Intelligence”. Take transportation as an example, electric cars and fuel cell vehicles are gaining growing attentions by people from around the world, because the climate change and environmental issues are widely recognized as a huge threat to human. However, it is hard to say that people have similar huge demand for self-driving vehicles. The article quotes that the average commuting time in US is 25 min, and people can do other things during the commutes. But “25 min” is not that long that people cannot waste, considering they take the risk of giving the control of their own life to machines. It does not mean that people will never accept self-driving vehicles, since people nowadays take flights without fear in which passengers also have no control of their lives. But I believe it will take much more time than what article claims before people accept this type of transportation tools. Car companies and other developers should take extra effort to prove and persuade people that the autonomous car is safe. It is true that electric cars and fuel cell vehicles are also still not widely used, but this is not mainly because people have concerns about the safety of the car, instead, due to the relatively higher price and lower performance compared to traditional gasoline-fueled cars.  
     The second point that I want to challenge is about the argument of model/physics-based approach compared to data-driven approach. The article claims that the model-based approach has been giving way to data-driven approach. It may be true that data-driven approach is more efficient in the sense that it can learn from a bunch of data sets with no necessity of understanding the principles hidden behind the data. However, sometimes it would be necessary to combine with model-based approach which can provide detailed understanding of the cause and effects. For example, in my research area of materials  science, traditional approach for understanding process-structure-property relationship of materials is achieved by time and money consuming trial and error approach, i.e. experiments. But the developments of physics based numerical simulations has made it possible to  do the “experiments” in the computer, which is very helpful to reveal and understand new physics behind the data. Those new physics are the critical instructions for further studies to produce meaningful data more efficiently. Indeed, the experiments and simulations together have produced the big data sets that can be used for data-driven approach to accelerate the material discovery process. However, it does not mean that the data sets are already big enough nor we have totally understand the physics behind them. In other words, model/physics-based approach is still a useful way to produce meaningful data, at least in my research field.
