Anca Dragan on Human-Robot Interaction, Reward Engineering, and Understanding Human Rationality

This podcast episode features Anca Dragan, a professor at Berkeley specializing in human-robot interaction (HRI) algorithms, discussing the complexities of designing robots that can effectively coordinate and interact with human beings. Dragan emphasizes that HRI moves beyond robots functioning in isolation, requiring them to account for human actions and preferences in shared environments. She highlights two main challenges: robots coexisting with humans in the same physical space and robots optimizing for the end-user's preferences rather than just a programmer's predefined objective. The conversation delves into how robots can understand human intent and anticipate behavior, often drawing parallels to inverse reinforcement learning (IRL) and utility maximization models.\n\nA significant portion of the discussion revolves around the challenge of modeling human behavior. Dragan explains that while simple models like Boltzmann rationality (a noisy version of utility maximization) have seen some success, they often fail in complex tasks where human behavior appears 'irrational' or 'messy.' This leads to a deeper exploration of behavioral economics' insights into human decision-making. However, Dragan proposes a hopeful alternative: instead of labeling humans as irrational, robots should assume humans are rational but operating under different assumptions, beliefs, or simplified world models (e.g., intuitive physics). By inferring these underlying models, robots can better interpret human commands and provide more effective assistance, as demonstrated in experiments like the 'lunar lander' task.\n\nDragan also introduces the concept of 'information gathering actions,' where robots actively influence human behavior to learn more about their preferences or driving styles, rather than passively observing. This proactive approach allows robots to engage in a dynamic 'dance' with humans, updating their internal models based on human responses. Examples include autonomous cars subtly 'nudging' to gauge another driver's aggressiveness. This paradigm shift suggests a more collaborative and adaptive future for HRI, where robots are not just executing tasks but are actively learning and adjusting to the nuances of human interaction.\n\nThe broader implications of this research touch upon the ethical and social dimensions of AI. Dragan discusses the importance of robot expressivity for fostering human connection and teaching children appropriate social behavior (e.g., not being rude to Alexa). The episode underscores the philosophical challenge of reducing human complexity to mathematical models, advocating for an empathetic approach that seeks to understand the 'why' behind human actions, even when they seem illogical from a purely robotic perspective. This perspective extends beyond robotics, suggesting a valuable lesson for human-to-human communication: assuming a lack of empathy or understanding rather than inherent irrationality." "concepts": [ "Human-Robot Interaction (HRI)

I thought that robotics is a really cool way to actually apply the stuff that I knew and loved like optimization.

I felt like I fell in love or something like it because I thought I know how a spot many works right it's just I mean there's nothing truly special it is it's great engineering work but the anthropomorphism that went on into my brain they came to life like a head little arm and like and looked at me he she looked at me you know I don't know there's a magical connection there and it made me realize wow robots can be so much more than things that manipulate objects they can be things that have a human connection.

I think that when robots move they can communicate so much about internal states or perceived internal states that they have and I think that's really useful in an element that we'll want in the future because I was reading this article about how kids are kids are being rude to Alexa because they can be rude to it and it doesn't really get angry right it doesn't reply it in any way it just says the same thing.

now you kind of have to expand the notion of state to include this human internal state what is the person actually perceiving what do they think about the robots something's better and then you have to optimize in that system.

if I'm a programmer I can specify some objective for the robot to go off and optimize you can specify the task but if I put the robot in your home presumably you might have your own opinions about well okay I want my house clean but how do I want it clean then how should robot how close to me it should come and all of that and so I think those are the two differences that you have your acting around people and you what you should be optimizing for should satisfy the preferences of that end user not of your programmer who programmed you.

inverse reinforcement learning which is the notion of someone acts demonstrates what how they want this thing done what isn't inverse reinforcement learning you said it right so it's it's the problem of take human behavior and infer reward function from this figure out what it is that that behavior is optimal with respect to.

people are not rational people are messy and emotional and irrational and have all sorts of heuristics that might be domain-specific and they're just they're just a messy mess.

maybe we can give them a bit the benefit of the doubt and maybe we can think of them as actually being relatively rational but just under different assumptions about the world about how the world works.

the robot doesn't need to sit there and passively observe human behavior and try to make sense of it the robot can act too and so there's these information gathering actions that the robot can take to sort of solicit responses that are actually informative.

it's not irrationality precise doing the right thing under the things that they know.