On 9/9/21, Matt Mahoney <[email protected]> wrote: > By "understanding" do you mean learning a model that can make predictions, > or do you also mean the positive reinforcement that accompanies learning > when your brain does it?
Mainly what I'm after here is that somebody might claim their AI understands because it "can make predictions" or "positive reinforcement blah blah blah." Now is that really understanding? I don't know. That's the question. It depends how you define understanding. If you mean only that given some input, the correct output results, fine. Just make sure you tell people that. What seems to happen though is that when people say their AI "understands," users imagine non-existent mental features presumably the same as ours present in a machine. But, show up to the event! We can argue more. > > On Tue, Sep 7, 2021, 7:01 PM Mike Archbold <[email protected]> wrote: > >> >> https://www.meetup.com/Northwest-Artificial-General-Intelligence-Meetup-Group/events/280242958/ >> >> It's common to hear claims in AI -- particularly by marketing types >> and other evangelists -- that their AI/robotics based technology >> "understands." They want to persuade. But what does a claim of >> "understanding" actually mean beyond the smoke screen of sales >> rhetoric? >> >> The problem is a tacit claim is often being made that the machine's >> understanding is equivalent to human understanding -- at least in some >> narrow task. That's what they want you to assume. But in what form is >> the understanding? How deep? Does it mean that novel cases can be >> understood and thus explained? At what point does the machine's >> understanding fail? It seems that "understanding" definitions are >> crucial: the boundaries of the machine's functionality appear to be >> the boundaries of its understanding. >> >> The user is invariably left to draw his/her own conclusions about the >> extent to which the machine's understanding resembles human >> understanding -- which is itself difficult to define. >> >> Since this issue is at the very core of artificial general >> intelligence, a two-part series is planned. >> >> For this event, Part I, we will be making an informal survey of the >> topic with this agenda: >> >> 1) Philippe Delmeire has prepared some brief materials (short videos, >> images, text) to get us attuned to the topic. >> >> 2) Mike Archbold (organizer) has prepared a community survey of >> definitions of understanding, particularly with respect to how a >> machine should understand. We will critique the definitions. Jim >> Bromer is working on highlights of the definitions. The current >> version of the survey draft can be found here -> >> >> https://drive.google.com/file/d/1Kij_Ak5uSWTkxdM2vDKBbpEuXdFtJrqj/view?usp=sharing ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T2ee04a3eb9a964b5-M21715cb8ede58fe1de0b03d3 Delivery options: https://agi.topicbox.com/groups/agi/subscription
