Astronauts must place total trust in the programming of their flights, which includes numerous self-checking procedures to correct itself, and warning devices to request human intervention. We are now witnessing fully automated flights with passenger-tourists with no specific skills. Human intervention is still possible, but only to a certain extent, in case of unforeseen events, and human intelligence remains assisted by AI. Astronauts are trained to eventually take full programmatic control in case of a major accident, such as a fire, depressurization, breakdown, or the impact of space waste. This has already happened. Would they blindly trust an artificial intelligence?

From smartphones and chatbots to predicting how protein molecules fold, AI is a technology that is already transforming many aspects of our lives. It has already been integrated in a number of applications and we expect it to be even more present in the coming years. How will these applications change our lives? Which areas in our society will benefit most from this? Will we see improvements in the life of all Canadians or will the benefit be limited to a subset?

Today, the topic of artificial intelligence is part and parcel of discussions in many sectors of society and the economy: image and speech recognition, machine translation, image generation, etc. This is in large part thanks to impressive advances in the last decade of machine learning and in particular deep learning applications. With the growing electrification of the energy mix and the diversification of generation sources— intermittent, variable and distributed—we are just scratching the surface of the potential for AI to accelerate the transition to ultra-efficient, low-emission and interconnected energy systems. New AI opportunities are fascinating and innovative, but they also raise many questions. Is artificial intelligence the best solution to accelerate the energetic transition? How do we consider the challenges of implementing an AI solution, such as robustness, bias and explainability, in order to avoid the potential lack of trust that is difficult to offset a posteriori? How do we set up a certification process for AI solutions to allow them to be used in highly regulated environments? How do we optimize the energy and resources used by AI itself, for learning and data storage, as we strive to meet climate change challenges?

The field of artificial intelligence has gone through some tough times, most notably known as "AI winters." But over the past decade, we have seen a dramatic rebound, thanks to new capabilities combined with scientific and technological advances. AI is now seen as an enabling technology, with the potential to make an impact in many, many areas. It is often compared to the technologies that fueled the industrial revolutions. Although more and more concrete applications of AI exist, research in the field is far from over. In what areas has AI not yet been exploited to its full potential? At a more fundamental level, what are the limits of current approaches? What are the promising avenues of research and what are the obstacles that seem insurmountable? In short, what is missing in AI today?