#81 JULIAN TOGELIUS, Prof. KEN STANLEY - AGI, Games, Diversity & Creativity [UNPLUGGED]

AI & Culture

The conversation begins with the potential of AI to participate in cultural creation. Julian Togelius shares his background in a family of artists and his interest in machines that can create autonomously or in collaboration with humans. He believes that art and technological development are deeply intertwined, with new technologies driving new art forms and vice versa 1. This symbiotic relationship could lead to AI contributing significantly to our cultural landscape.

I think the history of art is very much a history of technological development, because art has driven so much development in ways of manipulating colors and materials and representing reality and so on.

--- Julian Togelius

Togelius emphasizes that AI's role in art could enrich our lives and help us understand ourselves better 1.

   

AI in Games

The discussion then shifts to AI in game design, where Julian Togelius and Ken Stanley highlight the challenges and opportunities. Togelius notes that most game designs predate useful AI, and integrating AI into games requires rethinking traditional design conventions 2. He suggests that individual AI practitioners and game designers could lead the way by experimenting with AI-driven game designs, despite the industry's risk aversion.

Just like taking an existing game design, putting some kind of really clever agent in there, it's not going to make the game better in most cases, it's just going to make it worse.

--- Julian Togelius

Stanley adds that big tech companies have research wings, but the game industry lacks similar labs to explore innovative AI applications 2.

   

Ecosystem Complexity

Open-ended learning in AI is another focal point, with Julian Togelius discussing the importance of considering the entire ecosystem. He argues that intelligence emerges from the complex interactions within an ecosystem, not just from individual agents 3. This perspective highlights the need for AI systems that can evolve in dynamic environments.

The reason we could build this complicated civilization is the extreme complexity of life on earth, which then depends on previous life on earth and so on.

--- Julian Togelius

Togelius believes that focusing on population-level intelligence could lead to more robust and adaptable AI systems 3.

   

Emergence

The concept of emergence in machine learning is explored, with Tim Scarfe noting that interesting phenomena often occur at higher levels of abstraction. Ken Stanley discusses the balance between specialization and generalization in AI, suggesting that population-driven algorithms could foster more diverse and specialized behaviors 4. This approach contrasts with the current focus on creating generalist AI systems.

I think population driven algorithms sort of implicitly are more about specialization a lot of the time, because, like each member of the population, you want them to be doing some different thing, so they're kind of becoming specialists.

--- Ken Stanley

Scarfe and Stanley agree that fostering specialization within AI populations could lead to more innovative and effective solutions 4.