My broad point is that there is only one way to test knowledge ultimately -
physically.
Science demands physical evidence for everything.
It then has in effect a graded system of veracity (although there is no
formalised system). The truest knowledge comes from direct physical observation
and then physical discovery - literally taking something to pieces, where you
find and directly see a new piece, or see it with some new kind of sensor. Then
comes experimental discovery/knowledge, when you in effect rattle the box, and
make the inner piece rattle, but can't be absolutely sure of what you're
hearing. Then comes theoretical discovery, where you infer the inner piece from
the outside of the box and may have only a v. shadowy picture of its nature.
Then comes hypothesis, where you have an idea about the inner piece, but the
evidence may be inconclusive or non-existent - and there's only a deductive
likelihood - "this is the sort of thing you'd expect to be there, on the basis
of what we know about other similar bodies."
There are probably further degrees of truth-testing - such as secondhand
evidence - someone says he saw it. And the ultimate, least true is taking
things purely on faith, where you just take someone's word for it, or your own
intuition, without knowing whether anything was actually seen. "There is a God
called Bubbel." Science takes some things on faith too, e.g. determinism.
The ordinary human, I suggest, isn't as systematic as science but has a
loosely similar graded system of veracity, based on physical evidence (and what
"I saw with my own eyes") - which is applied, if mainly unconsciously, to all
knowledge gathered.
Speaking extremely broadly, of course, I see no alternative to something like
the graded, evidence-based system of veracity, that I'm v. crudely sketching.-
for any real-world knowledge-gatherer, and certainly not for any would-be
superAGI.
Do you? ["Google-worship", for instance, won't cut it :) And even "peer/ expert
approval" is deeply unreliable. ]
Stephen:MW/MT: ... how do you test acquired knowledge?
I have given this problem some thought, regarding the testing of acquired
grammar facts, rules and skills. Here are some points, mostly from my
experience with Cyc.
a.. Before the knowledge is acquired, the mentor (or ultimately the system
itself) should develop a suite of test questions, and expected answers.
b.. These tests not only will subsequently ensure that the factual
knowledge has been acquired, but they can be routinely performed to ensure that
new acquired knowledge, and new contexts, does not unintentionally corrupt old
answers.
c.. Regarding skill testing, I think that we can draw techniques from the
field of human education. For example, when the Texai system learns a new
construction grammar rule, it should demonstrate to the mentor's satisfaction
that it can use the rule on utterances beyond the training set.
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
http://www.listbox.com/member/?member_id=8660244&id_secret=98558129-0bdb63
Powered by Listbox: http://www.listbox.com
