The human brain is complex and is capable of the most amazing and complicated creative tasks. More than 50 years ago, computers inspired by the architecture or learning approaches within the human brain gave birth to the field of Artificial Intelligence (AI). During the last 50 years, computer power and memory capacity has doubled about every two years. This has led to an approximately hundred million times improvement in computer hardware. After suffering from severely limited early performance (due to the hardware constraints at that time) and widespread pessimism on prospects of AI in the 1980s, AI is now starting to make a real impact on society. The realization of virtual artificial neural networks and deep learning machines, built upon massively parallel Graphical Processor Units (GPUs), has enabled AI to start to solve problems across a range of disciplines. These include finance, medicine, manufacturing, robotics, multimedia, social networks, telecommunications, computational linguistics and cyber-security. As a result, the 21st century will likely see the advent of true AI machines.
In 2018, I was amazed to see the human-like games of self-trained AI-based machine learning algorithms (e.g., AlphaZero developed by DeepMind) defeat the previously world-leading chess-playing software Stockfish. Stockfish employed massive search and human-developed position evaluation rules (making use of approximate values for different pieces, typically counting a pawn as worth one point, a knight as 3 points, a rook as 5 points and a queen as 9 points, and with points given in various positional features) but was outplayed by Alphazero which relied less on search-based tactical computations but rather seemed to use positional understanding to defeat the software, similar to the way in which humans can outplay computers in chess.
As the applications for AI become more developed there is increasing demand for AI specialists to address its applications in both local and global employment markets. AI application domains affect all sectors of society; finance, entertainment, education, transport, communications, manufacturing, security and medicine are just some of those likely to be impacted.
Being at the forefront of global AI research, CUHK will offer Hong Kong's first undergraduate programme on AI, starting with the 2019-20 intake. The programme will be offered by the Department of Computer Science and Engineering, which has already achieved some of the highest accuracies in medical image analysis with its spin-off company (Imsight Medical Technology). Research and development of AI technologies is not limited to the discipline of Computer Science in the Faculty of Engineering and is thriving across our different academic disciplines:
• One of the highest-valued local AI start-ups, SenseTime (US$1.6 billion), came from the Multimedia Laboratory led by Prof. Sean Tang (located in the Departments of Information Engineering and Electronic Engineering). The same group hosts the Nvidia CUDA research center.
• The group led by Prof. Raymond Tong (Department of Biomedical Engineering) employs AI in rehabilitation training and developed the start-up CURehab.
• The group led by Prof. Yunhui Liu in the Department of Mechanical and Automation Engineering created a spin-off company, VisionNav, for autonomous industrial vehicles. Professor Liu also leads the RGC Theme-based research project on surgical robotics and is conducting clinical trials of two robot-assisted surgical systems for sinus and hysterectomy surgeries at Prince of Wales Hospital.
• The Information Engineering team led by Prof. Wing-cheong Lau and Prof. Kehuan Zhang achieved award-winning results with their application of AI in network security.
• Prof. Helen Meng's team in the Department of Systems Engineering and Engineering Management has made impressive advances in human-machine interfaces and multilingual speech processing.
• At the Third Chinese Youth Congress on Artificial Intelligence, the team led by Prof. Kam-fai Wong, from the Department of Systems Engineering and Engineering Management, won First Prize for Innovation.
There are many other concrete examples. Figure 1 shows some of the AI-related research areas that are being actively pursued in the Faculty of Engineering. One of the big challenges is AI's current reliance on power-hungry classical computing architectures. I end this article with my own research perspective on one of the sector's future challenges being addressed in the Department of Electronic Engineering. The annual electricity bill for an AI research lab, requiring the massively parallel processors of today's GPUs, can be of the order of many millions of dollars. For the widespread application of AI, we need exponential improvements in the power efficiency of AI computing hardware so that it can approach the power efficiency of the human brain – a need that has opened tremendous opportunities for research on brain-inspired computer hardware architectures that make use of energy efficient optical rather than electronic circuits in the emerging technology of silicon-based photonic integrated circuits.
Prof. Hon Ki Tsang
Associate Dean (Research)
Faculty of Engineering
The Chinese University of Hong Kong