MIST is a quantiative test of humanness, consisting of ~80k propositions such as:
- Is Earth a planet?
- Is the sun bigger than my foot?
- Do people sometimes lie?
Have any AI attempted and passed this test to date?
Yes, although how useful this AI can be is another question entirely.
mpgac is a "minimally intelligent AGI" trained on the GAC-80K corpus of MIST questions. As a result, it should be able to "minimally" pass this test. However, being trained on the GAC-80K corpus obviously make it lacking for any practical purposes. From the README:
Obviously this should only be capable of producing a minimally intelligent signal when ordinary commonsense questions are asked, of the kind depicted above, using questions which would have made sense to an average human between the years 2000 and 2005. On expert knowledge or current affairs related questions it should perform no better than chance.
The point of mpgac is to compare it to other AIs that could be built to pass this test. Or as the writer wrote in the README:
When scanning the skies how can we tell whether the radio signals detected are from an intelligent source, or are merely just background or sensor noise?
Ideally, you would want to build a program that is "better" than mpgac. In much the same way as ELIZA can be seen as a baseline for the Turing Test, mpgac is the baseline for the MIST test.
The GitHub repo of mpgac (as well as the GAC-80K corpus) is available here.
I believe this is exactly the kind of test where Doug Lenat's cyc would do very well at ? But I can't answer the question : how much of that corpus could it answer correctly ? Probably quite a lot ! (and how many humans could pass that test ? probably not all of them, but many can...)
[but is cyc considered an AI? probably not... so I may be out of topic. But imo it's database should be incorporated to any AI that reaches some kind of "intelligence"...]