Advances in Artificial Intelligence (AI) and Machine Learning (ML) will have a strong impact on people, society and the planet in the future. GESDA has worked towards a first mapping of scientific breakthroughs over time frame of 5 to 25 years. This mapping has been developed by consulting with some of the leaders in the field and will be released as part of of a major report later this year.

The mapping considers the series of recent breakthrough enabled by deep learning techniques, sometimes also referred as the "second wave" of AI/ML or "statistical ML”. The breakthroughs have touched all areas of ML, i.e., supervised, unsupervised and reinforcement learning. Those breakthroughs have been substantially aided by the exponential increase in computational power that is available and the access to ever larger data sets. It is expected that in the coming 5 years this trend will continue, and more and more specialized application areas will benefit from this technology. Despite all this progress, a series of limitations remains for deep learning. These techniques work considerably less well for problems of combinatorial nature with many categorical (discrete) data; training requires substantial computational power and access to data, and allows only a few companies or organizations to pursue the most ambitious developments; the transferability of “learning” is limited; it is far from clear how to incorporate complicated “real world” objectives; the results are brittle by nature and easily fooled; and successful models are basically impossible to interpret or understand, making their certification for reliability a real challenge.

Looking into the future at 10-25 years, the "third wave" of AI will consist in integrating contextual information, common sense knowledge and high-order reasoning into machine learning algorithms. This will enable machines to learn from much smaller datasets than what is possible today, increasing their applicability substantially to a much larger and diversified set of real-world problems. Those algorithms will understand and perceive the world on their own and will be able to perform basic forms of reasoning. In perhaps 10 years, ML will also benefit from combining 'classical' ML techniques with dedicated chip architectures, possibly also with biocomputing and quantum algorithms.

The step beyond the third wave of AI may be the development of truly intelligent machines and Artificial General Intelligence (AGI - fourth wave of AI). It is debated within the community whether this will happen in a timeframe of 25 years. However, a recent survey conducted by the Future of Humanity Institute revealed that experts in the AI estimate that AGI will happen with a probability of 50% in 45 years and 10% in 9 years. AGI will have fundamental implications ranging from our understanding of basic science questions to new applications in virtually all areas of human activity.

Those developments inevitably ask a series of moral and governance questions: How likely are they and what will be their impact on the future human society? What are the questions around co-development, access and use of advanced AI technologies? Will we move towards more human-machine cooperation or will machines gradually take over more and more human tasks?

The purpose of the session is twofold:

1. Anticipate the future of machine learning and AI with leaders from the field, from a science and engineering perspective.
2. Explore the implications of those developments on the ‘human’ and society as a whole, followed by a debate around governance issues.