Bradford W. Miller

I usually describe my profession as cognitive science, cognitive systems architect, or more recently "computational psychologist" - my early work was in computational linguistics and knowledge representation (artificial intelligence), my mid career has been in understanding better the limitations of humans and trying to compensate with technology and current work is really embodied artificial general intelligence motivated by testing some psychologically based ideas for how humans aquire knowledge about the world and apply it to solving problems. This has led me to explicit cognitive models (in the Simon and Newell sense), though I should also thank my graduate advisor James Allen for pointing me in the direction of using plans to represent an agent's intentions (both at a domain and discourse level), as well as the centrality of Problem Solving (which ultimately stems from Simon's work on GPS). Note that I distinguish between the goal of cognitive science - understanding natural intelligence and using systems to test theories - as distinct from artificial intelligence - building systems that 'act intelligent'. Some of the methods are similar, but a cognitive scientist isn't going to be happy with a model unless it matches some modeled population, while AI generally doesn't care so long as it 'did the right thing.'

I'm very interested in the relationship between different kinds of (human) memory: episodic, semantic and procedural; child knowledge acquisition; distributed problem solving (particularly under time constraints) and "intuition, imagination and insight" to quote Simon.

My current gig at GE Global Research is in the robotics group, where I'm working on a cognitive architecture with an explicit attentional model. I'm really encouraged with recent developments on the hardware side (From Nvidia, Xilinx and others), as it will allow us to build systems that can do online learning without resorting to communication with infrastructure outside of the robot itself. But I'm also still interested in building graph-based computational systems as our team investigated under DARPA ACIP (back when I was at Raytheon). Brains are really really good at pattern matching; it makes a lot of sense to build hardware that would also be unusually good at doing that (in both the subgraph isomorphism sense, and the unification/predicate constrained logic variable sense).

In addition, I've been helping out with FRA funded studies intended to improve automation in a Locomotive Cab. This started out with building cognitive models (from a Human Factors perspective) and currently involves a new UI that is more collaborative with the Engineer rather than "manual" vs. "automatic." The hope is to allow the system to better understand the Engineer's intentions and to make its own model of the current state and operating model more transparent than the existing system (specifically GE's .. now Wabtec's Trip Optimizer).

LinkedIn profile