Marvin Minsky’s Frames and Roger Schank and Robert Abelson’s Scripts

While Marvin Minsky's story is so much more complex than the villain of neural networks, in the early 1970s, fresh from his Perceptrons mic drop on the entire field of neural networks, Minsky was still in the villainous symbolic AI phase.

The odd thing about rejecting neural networks is that Minsky was still trying to model what brains did. Rather than modeling based on neuroscience, he made something up called frames. In a June 1974 MIT memo, “A Framework for Representing Knowledge,” Minsky proposed that a mind meeting a new situation, such as a child’s birthday party, selects a remembered structure called a frame and adapts its details to fit.

A frame’s slots come filled with default assumptions that new information can displace. That I guess we can all mostly agree with. But how do they acquire that new information? Can they learn it through training or must it be hard-coded? Minsky went with hard coding.

In 1975 Yale’s Roger Schank and Robert Abelson presented scripts as a specialization of frames: stereotyped action sequences for familiar events. SAM, a program by Chris Riesbeck, Richard Cullingford and Wendy Lehnert, used a restaurant script to fill in a short story’s unstated steps. Their 1977 book Scripts, Plans, Goals and Understanding expanded the theory.

These theories were not entirely without use but they can be considered more cul de sacs than highways.