On Sat, Aug 10, 2019 at 9:04 AM korrelan <korrelan...@gmail.com> wrote: > >Legg proved there is no such thing as a simple, universal learner. So we can > >stop looking for one. > > With all due respect to everyone involved this kind of comprehensive sweeping > statement is both narrow minded and counter productive.
While I am being a downer, here are some other things you can't do. There is no general procedure for testing software. (Rice's theorem, a generalization of the halting theorem). You can't prove all true statements (Godel, Turing). You can't prove most true statements (Legg). You can't compress all strings, or even all strings above a certain length (pigeonhole principle). You can't always know if a string can be compressed (Kolmogorov). There is no universal induction algorithm (Solomonoff). There is no universal reinforcement learning algorithm (Hutter: AIXI). You can't both create an agent smarter than you and predict its behavior (Wolpert: mutual prediction is not possible). There is no such thing as recursively self improving software (because intelligence depends on knowledge and computing power and it gains neither). Sorry, it's math. You can have AGI. You just can't have any shortcuts. -- -- Matt Mahoney, mattmahone...@gmail.com ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T1ff21f8b11c8c9ae-Md85271c00281aa33c0fdaf1e Delivery options: https://agi.topicbox.com/groups/agi/subscription
