Today, most machine learning algorithms use the same paradigm: set an objective, and train an agent, a neural net, or a classical model to perform well against that objective. That approach has given good results: these types of AI can hear, speak, write, read, draw, drive and more. But they’re also inherently limited: because they optimize for objectives that seem interesting to humans, they often avoid regions of parameter space that are valuable, but that don’t immediately seem interesting to human beings, or the objective functions we set. That poses a challenge for researchers like Ken Stanley, whose goal is to build broadly superintelligent AIs — intelligent systems that outperform humans at a wide range of tasks. Among other things, Ken is a former startup founder and AI researcher, whose career has included work in academia, at UberAI labs, and most recently at OpenAI, where he leads the open-ended learning team. Ken joined me to talk about his 2015 book Greatness Cannot Be Planned: The Myth of the Objective, what open-endedness could mean for humanity, the future of intelligence, and even AI safety on this episode of the TDS podcast.

Support us (and please rate on podcast app) https://www.patreon.com/mlst  In this show tonight with Prof. Julian Togelius (NYU) and Prof. Ken Stanley we discuss open-endedness, AGI, game AI and reinforcement learning.   [Prof Julian Togelius] https://engineering.nyu.edu/faculty/julian-togelius https://twitter.com/togelius [Prof Ken Stanley] https://www.cs.ucf.edu/~kstanley/ https://twitter.com/kenneth0stanley TOC: [00:00:00] Introduction [00:01:07] AI and computer games [00:12:23] Intelligence [00:21:27] Intelligence Explosion [00:25:37] What should we be aspiring towards? [00:29:14] Should AI contribute to culture? [00:32:12] On creativity and open-endedness [00:36:11] RL overfitting [00:44:02] Diversity preservation [00:51:18] Empiricism vs rationalism , in gradient descent the data pushes you around [00:55:49] Creativity and interestingness (does complexity / information increase) [01:03:20] What does a population give us? [01:05:58] Emergence / generalisation snobbery References; [Hutter/Legg] Universal Intelligence: A Definition of Machine Intelligence https://arxiv.org/abs/0712.3329 https://en.wikipedia.org/wiki/Artificial_general_intelligence https://en.wikipedia.org/wiki/I._J._Good https://en.wikipedia.org/wiki/G%C3%B6del_machine [Chollet] Impossibility of intelligence explosion https://medium.com/@francois.chollet/the-impossibility-of-intelligence-explosion-5be4a9eda6ec [Alex Irpan] - RL is hard https://www.alexirpan.com/2018/02/14/rl-hard.html https://nethackchallenge.com/ Map elites https://arxiv.org/abs/1504.04909 Covariance Matrix Adaptation for the Rapid Illumination of Behavior Space https://arxiv.org/abs/1912.02400 [Stanley] - Why greatness cannot be planned https://www.amazon.com/Why-Greatness-Cannot-Planned-Objective/dp/3319155237 [Lehman/Stanley] Abandoning Objectives: Evolution through the Search for Novelty Alone https://www.cs.swarthmore.edu/~meeden/DevelopmentalRobotics/lehman_ecj11.pdf

Dr. Ken Stanley, a leading expert on Open-Ended AI, discusses the Objective Paradox, Novelty Search, and the future of AI. The conversation explores the dangers and potential of Open-Ended AI, emphasizing the balance between safety and creativity in AI systems. Practical applications of Open-Ended AI in human decision-making are also highlighted, showcasing the benefits of following immediate interests over rigid objectives.