On Wed, Dec 26, 2012 at 12:45 AM, Ben Goertzel <[email protected]> wrote:
>> Watson is able to understand complex sentences and reason without any
>> embodiment. To compensate, it includes code to reason about space and
>> time. For example, if it knows that X is north of Y and Y is north of
>> Z then it can reason that X is north of Z.
>>
>
> Yes.  That kind of reasoning is quite simple and  many other systems
> can do that too.   However, Watson has a large and carefully curated
> knowledge base.
>
> My intuition is that approaches similar to Watson are not going to
> succeed at general-purpose language understanding...
>
> It may be that IBM will succeed at general-purpose language
> understanding.  But if so, I predict it will be via taking an
> architecture vaguely like Watson, and inserting into it a component
> more like OpenCog, that does much deeper and broader forms of
> generalization than anything in Watson...
>
> -- Ben G

Watson's knowledge base is 4 TB of text. That is 4000 times larger
than an average person would hear and read in a lifetime. That, and a
fast reaction time on the buzzer, compensates for it other weaknesses
such as lack of vision and embodiment.

Reasoning about space and time are fairly simple, but these are only 2
of over 100 modules, each of which can answer only a few percent of
the questions. The intelligence comes from putting all of these
together.

What kind of test would be appropriate for comparing Watson with OpenCog?


--
-- Matt Mahoney, [email protected]


-------------------------------------------
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

Reply via email to