Beneficial AI

←All Research

Making AI work for humans, now

Beneficial artificial intelligence is a term I came up with to encompass the a broad push to make robotics, machine learning, and related autonomy more useful for humans. This encompasses a broad push to integrate work from AI Safety, Ethical AI, Fair Machine Learning, and Human-in-the-loop autonomy. What started as a blogging direction is now an active area of my research.

Robots interact with humans in the physical world, so any deleterious effects of algorithms will be magnified (and even more concentrated on lower-income groups that engage with companies more likely to automate, those with low margins).

One of my big hopes is that model-based RL proves useful in pure robotics, and in addressing some of these sociotechnical concerns.

Open areas of study:

  • interpretable RL, guiding policy for RL,
  • AI “clinic”
  • Mitigating harms of intelligent consumer drones (e.g. slaughterbots)

(Papers accepted and awaiting publication)

When you "measure data", you quantify its characteristics to support dataset comparison & curation. You also begin to know what systems will learn. Many ML systems don't reason with this, we posit you should.
We propose a new type of documentation for dynamic machine learning (and reinforcement learning) systems!
We detail why reinforcement learning systems pose a different type of (dynamic) risks to society. This paper outlines the different types of feedback present in RL systems, the risks they pose, and a path forward for policymakers.
We present a concise set of directions for understanding the societal risks of new directions of AI research.
We study three developing subfields of AI research and their growing relationship with the sociotechnical: AI Safety, Fair Machine Learning, and Human-in-the-loop Autonomy.