Amazon is now running live robots inside its distribution centers, drones that rapidly shuttle merchandise across these massive warehouses so that the world's largest retailer can ship you all those holiday gifts without delay. These orange'bots know exactly where they're going, and they know exactly what objects they need to grab. They even plug themselves into charging stations when they need more juice. But not as impressive as SAM, the 450-pound bot that could understand upwards of six trillion spoken phrases and respond accordingly. And SAM was doing it back in 1989. He looked like little more than an enormous mechanical arm.

Rosen, a researcher at the Stanford Research Institute in Menlo Park, California, envisioned a roboperson driven by neural networks, algorithms that mimic the human brain. Much like its biological counterparts, it would have the power to see and sense its environment. As Rosen and his team wrote in a memo (.pdf) to DARPA, the Defense Department's research arm, describing the project, it would "perform reconnaissance missions" that would normally require human intelligence. DARPA eventually "got kind of excited about it," recalls Nils Nilsson, one of the leaders of the project, and the agency granted the researchers $750,000 -- more than $5 million in today's money -- to make it happen. The project didn't include true neutral networks -- in the 1960s, the technology just wasn't up to the sort of visual analysis, planning, and navigation Rosen and team wanted to explore -- but the automaton did indeed happen.