(Source: ccold)
(Source: italdred)
From November 11-18, 2012, we held (what we now call) the *1st MIRI
Workshop on Logic, Probability, and Reflection*. This workshop had four
participants:
- Eliezer Yudkowsky > (MIRI)
- Paul Christiano > (UC Berkeley)
- Marcello Herreschoff (Google)
- Mihaly Barasz (Google)
The participants worked on the foundations of probabilistic reflective
reasoning. In particular, they showed that a careful formalization of
probabilistic logic can circumvent many classical paradoxes of
self-reference. Applied to metamathematics, this framework provides (what
seems to be) the first definition of truth which is expressive enough for
use in reflective reasoning. Applied to set theory, this framework provides
an implementation of probabilistic set theory based on *unrestricted*
comprehension
which is nevertheless powerful enough to formalize ordinary mathematical
reasoning (in contrast with similar fuzzy set theories, which were
originally proposed for this purpose but later discovered to be
incompatible with mathematical induction).
These results suggest a similar approach may be used to work around Löb’s
theorem , but this has not
yet been explored. This work will be written up over the coming months.
http://intelligence.org/2013/03/07/upcoming-miri-research-workshops/>
James Liu , Likes to develop games :)
*6* votes by Sabrine Rekik , Patricia
Troyer , Sagar
Ghoting
, (more)
*Machine Learning* is designed to *optimize* problems which are not obvious
or not solvable by human deduction alone (or we can’t do it fast enough to
make it worth the effort).
Introduce *Zid**,* a happy-go-luck-does-only-what-you-allow-it-to-do *robot*
* **chicken*. You are *Zid*’s master, and it worships your very existence.
As *Zid*’s master you have the ability to allow *Zid* to experience extreme
existential satisfaction to exist along side you. You also have the
ability to have *Zid*experience great remorse.
Human Happiness is a problem that is not solved and not obvious. *Zid* will
try a slew of things to make you happy. Some will fail… most will fail
miserably…. *but once it begins to learn* (i.e. receives score points),
it can begin to maintain a long-term sustained optimized score.
The difference to highlight here is that Machine Learning can be used to
evaluate problems with a changing data-set or changing target goals.
Maybe last year you loved getting a beer at 5pm everyday, but this year you
prefer it to be at 6pm. *Zid* should be programmed with a set of algorithms
that will look for optimal, and then begin to perform boundary tests to try
and achieve higher and higher scores.
One day *Zid* will bring you a Poo. You need to make sure you subtract
points for this… and make sure you don’t laugh or smile…
Replace *Zid* with Website
Replace Beer with Notification
Replace You with Users
Replace Points with User Engagements
“Website gives you a notification at 5pm, but people don’t prefer it at
this hour.”
Talking toys with low batteries sound like Satan.