In the past decade, there have been staggering progress in the technological level of artificial intelligence (AI). The progress started from 2010 where Google performed tests on “voice recognition” technology via Deep Neural Network (DNN), followed by their acquisition of the startup company DeepMind in 2014 and its subsequent launch in 2017 of the Go match software AlphaGo Zero containing “Adaptive Algorithm" that simulates the learning process of a human being. Starting from zero without inputting any go game records, after 40 days of learning through matches against itself, it defeated the most advanced version of its predecessor (the previous generation required input of 100,000 Go game records in advance; and the game of Go has a total 10^170 variations), going beyond the total knowledge of Go throughout human history! This epoch-making milestone in the history of AI means that the life of human beings will enter a brand new era in the next few decades. We are dealing with AI on a daily basis as we speak, including searching for information on the Internet, clicking on videos, shopping online, check-in and likes on social media, mobile phone voice assistants, etc. Certainly, it will also extend to automatic medical diagnosis, home robots, smart farms, smart warehousing, stock trading algorithm, biometrics, self-driving and miscellaneous types of tasks with high repetitiveness. Among these daily life applications contains the most complicated and difficult matter of autonomous driving which does not only the sensing and computing of unpredictable road conditions in the surrounding but also requires extremely high-speed and powerful computing capabilities to make real-time decisions and responses, so as not to cause irreparable errors.
The essence of autonomous driving is similar to that of AlphaGo Zero, which the “brain” of the system cannot respond to the actual situation only by inputting Go game records (road conditions); it has to continuously learn from road tests and accumulate all possible road conditions that may occur in the future with an attitude like a human being learning how to drive a car. Over the past few years, the capabilities of self-driving "vision" aids (LIDAR, camera, radar, ultrasonic sensing, etc.) have been rapidly improved, along with exponential growth of arithmetic units (the surge of quantum computer technology development can be expected in the coming decade), making “big data” the element in urgent demand in the process of learning. The accumulation of big data requires time and money, which is the greatest obstacle hindering the fully autonomous vehicles from commercialization. At present, the two main contenders of self-driving technology i.e. the dual powerhouses of AI - China and the United States, are making every effort to collect huge quantity of road test data for facilitating “safety index” of self-driving cars. I’d refrain from considering the road tests solution to everything, as it is too time-consuming and expensive. Sooner or later, a powerful computer simulation system should be developed to expedite the testing of self-driving system by creating a virtual world with road conditions containing more obstacles and anomalies compared to the reality based on the overlapping of conditions collected in previous road tests; this resembles the "accelerated durability" in the traditional car industry, which more severe test conditions are applied for tests to gain the results of normal tests which are more time-consuming in actuality. In addition, the "objective function" of the AI self-driving system is rather complex, as it not only requires the fastest and most comfortable route and manner to destination but also the considerations such as "ethics", "order of yielding", etc. to prevent or minimize the severity of an accident when it is imminent or expected to happen. These topics shall be engaged by more than technology companies and automakers, as this is something mandatory for government organs, academic institutions, social leaders, and insurance companies. Hence, I would conclude that self-driving is more than conquering a mountain called “technology”, for there are various non-technical challenges requiring solution by the mass before fully self-driving cars can be mass-produced and popularized.