Neuroscience/ML: Discussion on A Brief History of Intelligence - Breakthrough #2
Details
Join us for a discussion on the second section of Max Bennett's terrific and insightful book A Brief History of Intelligence (2023).
We already discussed the first section in a previous event, but please feel free to join even if you weren't present for that one.
About the book
After the success of Large Language Models like ChatGPT, there's a lot of talk about the impending arrival of artificial human-level intelligence. However, there is a lack of consensus about what else needs to happen before we reach AGI.
Max Bennett's A Brief History of Intelligence sheds light on this topic by taking a principled evolutionary neuroscience approach. Intelligence appeared in small steps, starting with less intelligent creatures (e.g. C. Elegans with ~ 300 neurons and ~ 1000 connections) and working all the way up to humans (billions of neurons and trillions of connections). The book examines the evolutionary lineage of humans and traces human intelligence to five evolutionary breakthroughs, and compares and contrasts these evolutionary innovations to various ideas in Deep Learning.
At the end of the book, one gets a good picture of what human intelligence (and animal intelligence in general) is all about, and where Deep Learning stands in relation to it.
The book is written in simple language and is quite the page turner!
About the event
We will discuss the second section of the book titled Breakthrough #2: Reinforcing and the first vertebrates (page 105 - 169). Participants are expected to have read this section before joining the event.
We will begin at 18:30 with introductions over pizza and drinks. At around 19:00, we will start the discussion.
Our discussion will be open-ended, but could revolve around the following themes:
- The repurposing of dopamine into a TD-learning signal, and similarities/differences with Deep Reinforcement Learning.
- Pattern recognition/memory in the brains of early vertebrates and similarities/differences with neural networks that perform recognition tasks.
- The first models of the world (place/grid cells) and relationship to ML models.
- Any questions you have about the second section of the book.
The end time of the discussion is open ended. We have put 3 hours tentatively, but the space allows us to discuss longer if we want.
Looking forward to seeing you there!
What about the other sections of the book?
We already covered the first section in a previous event.
Depending on the interest, we will cover the remaining sections of the book in future event. The idea is to use alternate months for this book (until it is done), and the other months for Deep Learning related papers.