> On Jun 25, 2015, at 9:10 PM, Mike Archbold <[email protected]> wrote: > > The issue of proving some AGI will *not* work seems to be at the forefront.
Yes, that was the point. There are usually plenty of ways to show that a proposed AGI architecture cannot express the claimed capabilities. You don’t need to build a rocket to prove that a hot air balloon will never reach the moon. There has always been resistance to this kind of rigor when it comes to AGI though; does not leave much room for magical thinking. > We already know what doesn't work: everything currently out there! I do not think this is necessarily true. A minuscule subset of recent models have no obvious design defects in terms of what they can tractably express. They may or may not be AGI capable but there is no definitive way to prove this, just that the possibility cannot be excluded. AGI falls under the class of software systems where a proof of concept is approximately worthless for demonstration of capability. There is a lot of execution risk sitting between a potentially sound design and a demonstration of capability, so many things have to come together. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
