What Is Lila and How Does It Learn Like a Child?

TL;DR

Lila is an AI system designed to learn through experimentation, modeled on human cognitive development. It uses a constructivist approach to build complex plans by drawing on statistical knowledge and composable data structures, allowing it to understand hidden states and achieve goals through iterative learning.

Transcript

so I think about half of you know me I'm David Michael Wallace I'm one of the partners at Lila and today since hardly anybody has heard of Lila we've been sort of working away writing code and being accidentally stealthy I start with the 50,000 foot overview of what Lila is going to talk about why we built Lila based on what it is let's start to ju... Read More

Key Insights

• 🥶 Lila is a self-funded project with a team of old AI lab members focused on writing running code and building an AI system based on constructivist principles.
• 🏛️ Lila's constructivist model follows the footsteps of human childhood development, learning through experimentation, reinforcement, and building reliable schemas.
• 😒 It uses statistical knowledge and composable data structures to develop an understanding of the world and build complex plans.

Questions & Answers

Q: What is Lila and what is its purpose?

Lila is an AI system based on a constructivist model that learns through experimentation and builds complex plans. Its purpose is to develop common sense, do reasoning and planning, and provide explanations.

Q: How does Lila acquire language and what role does it play?

Lila acquires language through a teacher that narrates its actions. Language is correlated with activities and used as a heuristic in the learning process. Lila uses language to generate explanations, plans, and reasoning.

Q: How does Lila build complex plans and what are its core data structures?

Lila builds complex plans through experimentation and reinforcement of correlations. Its core data structures are composed of primitive state items, synthetic results, and required computation. Plans and goals are composable, allowing for the development of sophisticated plans.

Q: How does Lila's constructivist model differ from traditional AI systems?

Lila's constructivist model emphasizes learning through experimentation and building reliable schemas rather than relying on preconceived rules. It focuses on understanding hidden state object persistence and uses statistical knowledge to make plans and achieve goals.

Summary & Key Takeaways

• Lila is a self-funded project working on building an AI system based on a constructivist model.

• The system focuses on learning through experimentation and building complex plans based on statistical knowledge.

• Lila develops understanding of hidden state object persistence and uses composability to achieve goals.