In July 1966 the MIT Artificial Intelligence Group issued an internal memo, “The Summer Vision Project,” authored by Seymour Papert and dated July 7, 1966. It is short, and it is famous for one reason: it proposed to make real progress on computer vision over a single summer using students.
The memo describes the goal plainly. The summer project was, in its own words, “an attempt to use our summer workers effectively in the construction of a significant part of a visual system.” The work was chosen partly because it could be split into sub-problems that individuals could tackle independently while still building toward a system complex enough to be “a real landmark in the development of pattern recognition.” The plan covered figure-ground separation, region analysis, and object identification.
It did not work out on that schedule. Getting computers to reliably segment scenes, recognize objects, and describe what they are looking at turned out to be one of the hardest problems in the field. The decisive breakthroughs did not arrive until the 2000s and 2010s - the ImageNet dataset in 2009 and the AlexNet convolutional network in 2012 - roughly half a century after the summer the problem was supposed to be largely handled.
The memo has become a standard parable for AI optimism precisely because it is so concrete. It is one thing to underestimate a problem in the abstract. It is another to write down “construction of a significant part of a visual system” as a task list for summer interns. The gap between that sentence and the fifty years it actually took is the whole story.