Challenge statement:
The article is over optimistic on AI in transportation. It predicts that
autonomous transportation will be commonplace in the near future just based 
on limited information and underestimate the difficulty of self-driving.

The paper lists many achievements such as GPS, ABS and TCS in the cars.
They are huge progresses in the automobile industry, but it is hard to 
link them with Artificial Intelligence. These techniques are all about
sensing and controling. Even though sensing and controling can be 
regarded as a kind of intelligence, it is relatively simple. 
These functionalities can be implemented by determining some if 
conditions and have nothing to do with complex analysis and reasoning.
These functionalities are a success of control theory and sensing
technologies, but not a success of AI.

The real application scenes to demonstrate the power of AI are tasks like
self-driving. It is much more challenging than GPS, ABS and TCS. The paper
also mention the self-driving car techniques and show that Google's and
Tesla's self-driving cars have logged 1.5 million miles with a few 
disengagements. It also predicts that in the near future, advanced sensors 
and sesning algorithms will surpass human's performance and cites a report
saying self-driving cars will be widely adopted by 2020. 

However, the information given are very limited to draw a conclusion that
self-driving cars would become a reality soon. The paper does not research
the challenges facing self-driving. From my perspective the most challenging
problem is understanding the varying and complicated environments. There
are many corner cases. Failing to handle them will cause the car to misbehave,
which is extremely serious for the safety of the car. Also the self-driving 
cars' vision systems needs to consider many factors, illumination, location,
objects and distances. To my best knowledge, although the state-of-the-art
computer vision algorithms outperform human in some benchmark dataset, there
is no evidence showing that computer vision systems can perform as well as 
human in reality. In fact, computer vision systems are sometimes easily 
fooled by change of illumination, motions and some other conditions. 
I do not know where the authors got the confidence that self-driving cars 
will be achieved by 2020.

Regarding the experiments of self-drving cars from Google and Tesla, even
the self-driving cars performs excellently in the city road with zero 
accident rates, there are many other terrain and road conditions. 
Working well in cities does not necessarily mean it will also work in other 
cases. Let alone there are crashes and disengagements in the city road 
tests. Right now, Tesla's self-driving cars still needs human monitoring,
which is far from the goal of self-driving.

The authors strengthen on how advanced the sensors will be while focus 
on how to tackling challenging AI problems, like objective 
detections/tracking and scene analysis. They are on the wrong
track. Self-driving cars cannot be implemented by just super-human
performance sensing technologies. On the other side, it seems the
authors conduct a careful research about the current bottlenecks in
computer vision, or try to find a breakthrough. They underestimate 
the difficulty of self-drivng.

The authors also lists benefits of self-driving cars. Some of them 
are not so harmonic with commensense need verifications, including
lengthening people's life expectancy and eliminating the need to own 
a vehicle due to car sharing. 

