GESDA Anticipation Workshop - Advanced Artificial Intelligence

Martin Müller

Artificial intelligence algorithms have become ubiquitous in modern life thanks to successes in machine learning research. However, they have very limited flexibility, operating within narrowly defined parameters and unable to transfer knowledge across domains. They also require vast amounts of training data and enormous computational resources. This is driving efforts to develop new deep learning approaches that are more reliable, transparent, and efficient. Promising avenues include graph neural networks , capsule networks and sparse neural networks.

Rüdiger L. Urbanke

Emmanuel Abbé

AI research has traditionally focussed on solving discrete problems that involve a single type of data, such as images, text, or audio. This has led to superhuman capabilities in narrow areas like object recognition, speech recognition and game playing. But humans and animals use multiple senses to navigate the world around them, and there is growing recognition that for AI to become more flexible it will need to work with multiple data modalities at once.

Massive AI models running on servers are incredibly powerful, but to make use of them data has to be streamed back and forth via the cloud. This is not ideal for privacy sensitive-applications such as healthcare and smart homes, or safety-critical ones such as autonomous vehicles or robotics, where network lags could lead to accidents. The sheer volume of data produced by the growing Internet of Things also presents a problem.

Despite progress in deep learning, recreating the kind of flexible and efficient intelligence seen in humans may involve coupling it with other approaches. There are question marks over whether the statistical patterns these algorithms learn can ever equate to true understanding . They also require vastly more data and energy to learn than humans, are notoriously bad at transferring knowledge between domains, and are incapable of continuous learning.

GESDA Anticipation Workshop - Advanced Artificial Intelligence