Matt Johnson-Roberson - Merging Paths: The Shared History and Convergent Future of AI and Robotics

In this talk

1. 20:20

   AI’s early ideas and limits

   Johnson-Roberson revisits early work by Herb Simon, Allen Newell, and John McCarthy, including Logic Theorist and LISP. He notes that researchers explored learning, reasoning, and intelligence decades ago, while limited memory and computing power pushed them toward simpler tasks. The perceptron’s early dismissal shows how promising ideas can be overlooked when the timing or research climate is poor.

2. 34:32

   Robots tackled familiar problems early

   Early examples include the Unimate industrial arm, the Hopkins Beast finding a wall plug to recharge, and Shakey following natural-language commands. Johnson-Roberson connects these projects to modern mobile robots and asks how much current systems improve on earlier approaches. He also highlights the importance of structured knowledge and system architecture.

3. 41:21

   Funding, hype, and research cycles

   The talk follows funding setbacks, the U.K.’s Lighthill report, and the rise and collapse of commercial expert systems. Johnson-Roberson draws parallels between past booms and current demand for AI talent, warning that hype and inflated promises can distort research priorities. He argues that ending a line of inquiry too early can cost the field valuable ideas.

4. 52:04

   Autonomous vehicles and lessons for the future

   Projects such as ALV, Navlab, and ALVINN explored outdoor autonomy, mapped routes, avoided obstacles, and used neural networks to steer. Their progress was constrained by computing limits, while ambitious promises sometimes exceeded what teams could deliver. Johnson-Roberson closes by urging researchers to stay with hard problems, revisit knowledge representation and architecture, and help the next generation find opportunities.