Vladlen Koltun — The Power of Simulation and Abstraction

Intelligent Systems

Vladlen Koltun, Chief Scientist for Intelligent Systems at Intel, prefers the term "intelligent systems" over "AI" due to its broader scope and less historical baggage. He argues that while AI has a long history of overpromising and underdelivering, intelligent systems encompass both artificial and natural intelligence, focusing on understanding and inducing intelligence in systems. This perspective allows for a more neutral and comprehensive approach to studying intelligence.

We want to understand the nature of intelligence. We are concerned with intelligence, understanding it and producing it, inducing it in systems that we create.

--- Vladlen Koltun

This shift in terminology reflects a desire to explore intelligence beyond artificial constructs, aiming to integrate insights from natural intelligence as well 1.

   

Embodied Intelligence

Embodied intelligence, as discussed by Vladlen, involves training intelligent agents to interact with their environment in a spatially aware manner, similar to humans and animals. He emphasizes the importance of understanding essential features of spatial intelligence without simulating every physical complexity. This approach aims to develop systems that can navigate and interact with their surroundings effectively.

We can also understand the essence of embodied intelligence without worrying about... how to grasp wet, slippery pebbles and how to pour coffee from a particular type of container.

--- Vladlen Koltun

By focusing on core aspects of intelligence, researchers can make significant progress in creating systems capable of functioning in real-world environments 2 3.

   

Robotics Challenges

The challenges in robotics, as highlighted by Vladlen, stem from the need for systems to learn through interaction with their environment. Unlike perception tasks, robotics requires real-time engagement, which complicates the learning process. He compares this to the development of a toddler, where learning is slow and requires a robust system capable of handling failures.

To learn to act, you need to actually act. To act, you need to act in an environment.

--- Vladlen Koltun

Vladlen suggests that identifying essential skills and using model systems, similar to neuroscience's use of squids, can aid in overcoming these challenges 4.