Start the AI roadmap with Minsky's two problem lists

TL;DR: Skip straight to problem two in Marvin Minsky's first Society of Mind homework. It asks which important problems in your field AI already addresses, and which important problems AI could contribute to. Those two lists make a useful opening for an engineering roadmap.

The source is MIT's Fall 2011 course, whose syllabus lists 13 sessions and describes discussion, critical reading and an applied final project. It is a seminar about how minds work, not a contemporary enterprise deployment guide. The assignment is valuable without pretending otherwise.

My extension is to give the two lists different commitments. A problem with an established technical approach needs an implementation plan and local acceptance criteria. A problem whose approach remains uncertain needs an experiment that can change your mind. Putting both under one delivery date hides a material difference in the work.

Name the uncertainty before choosing a model

As an illustration, extracting text from a known document format and deciding whether its claims support an investment recommendation are different problems. The first may have a well-understood method with integration work remaining. The second may require evidence that the proposed judgment is dependable at all.

I would put the unresolved assumption next to each roadmap item. Then write the smallest observation that would move it to the other list. This keeps a research bet visible even when its demo is polished, and stops an ordinary integration problem from acquiring an indefinite research budget.

The split is my planning rule, not one attributed to Minsky. Existing success elsewhere also does not prove readiness in your setting; data, constraints and failure costs can move a familiar technique back into experiment territory.

I read the assignment and syllabus, not the entire course or its two books. Start with the homework question, bring a real problem from your own field, and let the answer decide what to investigate next.

A roadmap gets more honest when an implementation promise and a research bet stop sharing the same label.