1966 MIT project targeted computer vision goals
Shubhendu Trivedi described a 1966 MIT Project MAC summer effort under Marvin Minsky that sought to solve computer vision. MIT accounts indicate the goals took roughly 50 years to reach practical realization. Trivedi asked whether added compute power alone could have enabled success without large modern datasets. Luca Ambrogioni replied that the absence of such datasets would have prevented any breakthrough.
#1478Shubhendu Trivedi@_ONIONESQUEThere is this MIT folk story that goes like "Minsky told a student to solve vision over the summer." It refers to an actual [quite ambitious] Project MAC 1966 summer project. Its goals only took ~50 years. If they had the compute but not huge datasets, could they have succeeded?
The actual project, one could say, could have happened. But I am talking of the folk version "attach a camera and describe whatever you see." What alternate tradition could it have led to? It's a non-serious-seeming counterfactual that is also an interesting historical exercise.
Shubhendu Trivedi@_onionesqueThere is this MIT folk story that goes like "Minsky told a student to solve vision over the summer." It refers to an actual [quite ambitious] Project MAC 1966 summer project. Its goals only took ~50 years. If they had the compute but not huge datasets, could they have succeeded?
The default answer is of course no, not feasible. In many ways various ML traditions and schools around AI were different counterfactual rollouts, since they all underestimated (and overestimated) different aspects.
Shubhendu Trivedi@_onionesqueThe actual project, one could say, could have happened. But I am talking of the folk version "attach a camera and describe whatever you see." What alternate tradition could it have led to? It's a non-serious-seeming counterfactual that is also an interesting historical exercise.
#1478Shubhendu Trivedi@_ONIONESQUEBut looking at the abstract you can see where the story comes from (used to be commonly told in Vision, heard it many times as a graduate student).
Shubhendu Trivedi@_onionesqueThere is this MIT folk story that goes like "Minsky told a student to solve vision over the summer." It refers to an actual [quite ambitious] Project MAC 1966 summer project. Its goals only took ~50 years. If they had the compute but not huge datasets, could they have succeeded?
And tbc, I meant only using what could be feasibly implemented at the time.
But some of the works from the time were geniunely quite modern e.g. the first (largely conceptual) statistical ML paper from 1956 by Solomonoff.
Shubhendu Trivedi@_onionesqueBut looking at the abstract you can see where the story comes from (used to be commonly told in Vision, heard it many times as a graduate student).
@_onionesque No
Shubhendu Trivedi@_onionesqueThere is this MIT folk story that goes like "Minsky told a student to solve vision over the summer." It refers to an actual [quite ambitious] Project MAC 1966 summer project. Its goals only took ~50 years. If they had the compute but not huge datasets, could they have succeeded?