Open-Endedness
Most AI systems nowadays are very specialized, being able to do amazing things in a narrow domain, yet that is not how intelligence in animals seems to work. When looking specifically at Reinforcement Learning, it is usual to set our agents into a world with a specific objective and define rewards that will lead these agents to achieve the defined goal. This paradigm again leads to extremely restricted agents, in which a simple change in the objective (such as "grab the red ball instead of the blue") can lead to an agent failing catastrophically and having to spend a large amount of time relearning the new task.
Open-Endedness tackles the problem of creating general agents that are not just extremely good at a narrow domain, but rather make agents that learn about the world around them and adapt to new tasks and settings. These agents have an immense potential for practical applications and may lead us to a better understanding of the differences between artificial and biological intelligence.
Join us in this panel with leading researchers in the field to ponder and ask questions about this grand challenge of Open-Endedness!
- Oct 7th 11:00AM (-3:00 UTC)
- Event link
Google Calendar / Other Calendar Apps - Learn More