What is Intelligence? What is AI? How can we create it?
The Future of Intelligence Fellowship is a 10-week long survey of attempts to answer these questions, with a focus on deep learning, arguably the most successful form of AI yet. It is meant to be accessible to all students regardless of past background.
Syllabus: We start off with an introduction to deep learning. What are deep neural networks (DNNs) and how do they learn? We’ll dive into both the theoretical and practical foundations of this question. We will apply DNNs to partially solve complex computational problems such as vision. We’ll then study how advanced models take inspiration from biological visual systems to improve even further. From vision, we pivot to the problem of natural language processing (NLP). We’ll learn about the Transformer and its powerful text generation abilities, and consider the social and ethical impacts of models like GPT-4. Lastly, we’ll consider how deep learning has been combined with reinforcement learning to produce some of the most stunning results in AI to date, from superhuman play in chess, Go, and Atari games, to solving protein folding, to even high level code production. We’ll discuss how these deep reinforcement learning models have pushed the boundaries of AI, and also think about what “the next big thing” for AI will be? In doing so, we’ll look to the brain for inspiration in redesigning the fundamentals of AI. And if we do end up building a superintelligent machine, we'll discuss how we might ensure it won’t pose an existential risk to humanity.
Note that this syllabus is not set in stone. This fellowship is for you - if a critical mass of fellows wish to cover something different to the syllabus in a later week, we will try our best to amend our plans accordingly.
- The HUMIC Board