Computer programs at the heart of artificial intelligence (AI) often use neural networks that, embedded in machines and programs, learn through example. They follow the logic of experience, but are restricted to the behaviours and facts that are provided to them.
Thus, AI grows through experiential learning and problem solving. Its efficacy depends on the breadth of the experience and the variety of problems that the neural networks have access to.
Dr. Simon Haykin is a longtime McMaster Engineering faculty member who has written the most popular postgraduate text book on neural networks. The third edition of the book, Neural Networks and Learning Machines published in 2009, explains how machines can learn.
Dr. Haykin’s research team studies how the process of structuring the digital world can be patterned on how the human brain works. They view this problem as a neural system that perceives information from multiple sources and thereafter instantly makes decisions on how best to respond, which is the basis for AI.
Progress in AI could provide social and economic benefits by enabling consistent, repetitious actions which would avoid pitfalls associated with human fatigue, emotions and time management.
Predictions about the impact of artificial intelligence imply that, just like the ubiquitous influence of the Internet, AI will also soon transform our world. Canada’s AI leadership is forecasted to potentially add $17.5-billion to annual GDP and 170,000 new jobs by 2025.
Dr. Tim Davidson is chair of McMaster’s department of electrical and computer engineering. In his opinion, “most of what has been popularized as ‘AI’ in the media falls into the machine learning and neural networks area. As the training of neural networks improves, which it will through access to large volumes of data, and as the algorithms that drive the machine learning processes advance, there is considerable potential for that type of AI.”
He thinks that machines might be able to reason before long, although he clarifies that humans will likely establish the rules under which that reasoning would occur.
Dr. Davidson explains that “if a self-driving car has to make a choice in a crash situation to risk the lives of its occupants or some pedestrians, it will likely run an algorithm designed by a human, rather than genuinely make the decision itself.
Therefore, automated decision making programmed by a human is possible, which places a lot of power and responsibility in a designer’s hands as the parameters of the algorithm are decided upon.”
The research of McMaster Engineering professor Dr. James Reilly and his collaborators uses electroencephalograms (EEG) to monitor the electrical activity in the brains of schizophrenic adults.
After enabling machine learning through the EEG data, they have been able to reasonably predict how these adult subjects would respond to a specific therapy. This research has been extended to predict the efficacy of patient–specific antidepressant therapy.
To solve both problems, trained clinicians had to first identify schizophrenia or depression. Hence, machine learning was essentially based on subjects who were identified through human experience and judgement. The results of the research could lead to a valuable tool for physicians,
McMaster professors, such as Dr. Chris Swartz, have also developed methods to retrieve data, e.g., from temperature sensors, pressure-flow monitors, chemical-concentration indexes, and digital cameras to develop AI solutions for industry. This data is then projected to a computer system that analyzes the information. As a result, automated changes to the controls ensure that better products are created and cost-savings are realised.
While human reasoning can be very strong, AI reasoning is still relatively weak although examples of stronger reasoning are growing. Using the IBM Watson AI, it has become possible to better understand complex data, and use the learning to improve the apps and computer programs that lie at the heart of smart systems.
Robust and reliable AI must be based on clean and dependable data, which is the focus on McMaster Engineering professor, Dr. Fei Chiang. She is working with key partners such as IBM (and Watson) to improve data quality.
Since poor quality data leads to financial loss and operational inefficiencies, the data must be improved by designing and building intuitive software tools to keep personal information private while automating this process for organizations, which saves them time, money and improves the data analysis and decisionmaking results from AI. Dr. Chiang’s research should be helpful to institutions such as banks and other large enterprises.
Autonomous systems are making the jump to mainstream commercial products, with applications ranging from autonomous vehicles to delivery drones. There will be many other spillover effects from AI advances.
Canada’s innovation ecosystem is well poised to take advantage. McMaster Engineering has been in this game for a long time.