How does MIRI Know it Has a Medium Probability of Success?

In the past, peo­ple like Eliezer Yud­kowsky (see 1, 2, 3, 4, and 5) have ar­gued that MIRI has a medium prob­a­bil­ity of suc­cess. What is this prob­a­bil­ity es­ti­mate based on and how is suc­cess defined?

I’ve read stan­dard MIRI liter­a­ture (like “Ev­i­dence and Im­port” and “Five Th­e­ses”), but I may have missed some­thing.

-

(Meta: I don’t think this de­serves a dis­cus­sion thread, but I posted this on the open thread and no-one re­sponded, and I think it’s im­por­tant enough to merit a re­sponse.)