A long, rambling but often interesting discussion among guys at MIRI about how to make an
AI that is superintelligent but not dangerous (FAI=Friendly AI). Here's an amusing
excerpt that starts at the bottom of page 30:
*Jacob*: Can't you ask it questions about what is believes will be true about the state of
the world in 20 years?
*Eliezer*: Sure. You could be like, what color will the sky be in 20 years? It would be
like, “blue”, or it’ll say “In 20 years there won't be a sky, the earth will have been
consumed by nanomachines,”and you're like, “why?”and the AI is like “Well, you know, you
do that sort of thing.”“Why?”And then there’s a 20 page thing.
*Dario*: But once it says the earth is going to be consumed by nanomachines, and you're
asking about the AI's set of plans, presumably, you reject this plan immediately and
preferably change the design of your AI.
*Eliezer*: The AI is like, “No, humans are going to do it.”Or the AI is like, “well
obviously, I'll be involved in the causal pathway but I’m not planning to do it.”
*Dario*: But this is a plan you don't want to execute.
*Eliezer*: /All/the plans seem to end up with the earth being consumed by
*Luke*: The problem is that we're trying to outsmart a superintelligence and make sure
that it's not tricking us somehow subtly with their own language.
*Dario*: But while we're just asking questions we always have the ability to
just shut it off.
*Eliezer*: Right, but first you ask it “What happens if I shut you off”and it says “The
earth gets consumed by nanobots in 19 years.”
I wonder if Bruno Marchal's theory might have something interesting to say about this
problem - like proving that there is no way to ensure "friendliness".
-------- Original Message --------
Machine Intelligence Research Institute » Blog
The Singularity Institute Blog <http://intelligence.org>
MIRI strategy conversation with Steinhardt, Karnofsky, and Amodei
Posted: 13 Jan 2014 11:22 PM PST
On October 27th, 2013, MIRI met with three additional members of the effective altruism
community to discuss MIRI’s organizational strategy. The participants were:
* Eliezer Yudkowsky <http://yudkowsky.net/> (research fellow at MIRI)
* Luke Muehlhauser <http://lukeprog.com/> (executive director at MIRI)
* Holden Karnofsky (co-CEO at GiveWell <http://www.givewell.org/>)
* Jacob Steinhardt <http://cs.stanford.edu/%7Ejsteinhardt/> (grad student in
science at Stanford)
* Dario Amodei <http://med.stanford.edu/profiles/Dario_Amodei/> (post-doc in
We recorded and transcribed much of the conversation, and then edited and paraphrased the
transcript for clarity, conciseness, and to protect the privacy of some content. The
resulting edited transcript is available in full here
Our conversation located some disagreements between the participants; these disagreements
are summarized below. This summary is not meant to present arguments with all their force,
but rather to serve as a guide to the reader for locating more information about these
disagreements. For each point, a page number has been provided for the approximate start
of that topic of discussion in the transcript, along with a phrase that can be searched
for in the text. In all cases, the participants would likely have quite a bit more to say
on the topic if engaged in a discussion on that specific point.
Page 7, starting at “the difficulty is with context changes”:
* Jacob: Statistical approaches can be very robust and need not rely on strong
assumptions, and logical approaches are unlikely to scale up to human-level
* Eliezer: FAI will have to rely on lawful probabilistic reasoning combined
transparent utility function, rather than our observing that previously
behaviors seemed ‘nice’ and trying to apply statistical guarantees directly
series of surface observations.
Page 10, starting at “a nice concrete example”
* Eliezer: Consider an AI that optimizes for the number of smiling faces
rather than for
human happiness, and thus tiles the universe with smiling faces. This example
illustrates a class of failure modes that are worrying.
* Jacob & Dario: This class of failure modes seems implausible to us.
Page 14, starting at “I think that as people want”:
* Jacob: There isn’t a big difference between learning utility functions from a
parameterized family vs. arbitrary utility functions.
* Eliezer: Unless ‘parameterized’ is Turing complete it would be extremely
hard to write
down a set of parameters such that human ‘right thing to do’ or CEV or even
selfish desires were within the hypothesis space.
Page 16, starting at “Sure, but some concepts are”:
* Jacob, Holden, & Dario: “Is Terry Schiavo a person” is a natural category.
* Eliezer: “Is Terry Schiavo a person” is not a natural category.
Page 21, starting at “I would go between the two”:
* Holden: Many of the most challenging problems relevant to FAI, if in fact
out to be relevant, will be best solved at a later stage of technological
when we have more advanced “tool-style” AI (possibly including AGI) in order
us with addressing these problems.
* Eliezer: Development may be faster and harder-to-control than we would like;
time our tools are much better we might not have the time or ability to make
before UFAI is an issue; and it’s not clear that we’ll be able to develop
AIs that are
extremely helpful for these problems while also being safe.
Page 24, starting at “I think the difference in your mental models”:
* Jacob & Dario: An “oracle-like” question-answering system is relatively
* Eliezer: An “oracle-like” question-answering system is really hard.
Page 24, starting at “I don’t know how to build”:
* Jacob: Pre-human-level AIs will not have a huge impact on the development of
* Eliezer: Building a very powerful AGI involves the AI carrying out
(consequentialist) internal optimization on itself.
Page 27, starting at “The Oracle AI makes a”:
* Jacob & Dario: It should not be too hard to examine the internal state of an
* Eliezer: While AI progress can be either pragmatically or theoretically
internal state of the program is often opaque to humans at first and rendered
partially transparent only later.
Page 38, starting at “And do you believe that within having”:
* Eliezer: I’ve observed that novices who try to develop FAI concepts don’t
seem to be
self-critical at all or ask themselves what could go wrong with their bright
* Jacob & Holden: This is irrelevant to the question of whether academics are
well-equipped to work on FAI, both because this is not the case in more
fields of research, and because attacking one’s own ideas is not necessarily
integral part of the research process compared to other important skills.
Page 40, starting at “That might be true, but something”:
* Holden: The major FAI-related characteristic that academics lack is cause
If we can get academics to work on FAI despite this, then we will have many
* Eliezer: Many different things are going wrong in the individuals and in
which add up to a near-total absence of attempted — let alone successful —
Page 53, starting at “I think the best path is to try”:
* Holden & Dario: It’s relatively easy to get people to rally (with useful
behind safety issues.
* Eliezer: No, it is hard.
Page 56, starting at “My response would be that’s the wrong thing”:
* Jacob & Dario: How should we present problems to academics? An
description is sufficient; academics are trained to formalize problems once
* Eliezer: I treasure such miracles when somebody shows up who can perform
them, but I
don’t intend to rely on it and certainly don’t think it’s the default case
academia. Hence I think in terms of MIRI needing to crispify problems to the
being 80% or 50% solved before they can really be farmed out anywhere.
This summary was produced by the following process: Jacob attempted a summary, and Eliezer
felt that his viewpoint was poorly expressed on several points and wrote back with his
proposed versions. Rather than try to find a summary both sides would be happy with, Jacob
stuck with his original statements and included Eliezer’s responses mostly as-is, and
Eliezer later edited them for clarity and conciseness. A Google Doc of the summary was
then produced by Luke and shared with all participants, with Luke bringing up several
points for clarification with each of the other participants. A couple points in the
summary were also removed because it was difficult to find consensus about their phrasing.
The summary was published once all participants were happy with the Google Doc.
The post MIRI strategy conversation with Steinhardt, Karnofsky, and Amodei
appeared first on Machine Intelligence Research Institute <http://intelligence.org>.
You are subscribed to email updates from Machine Intelligence Research Institute » Blog
To stop receiving these emails, you may unsubscribe now
delivery powered by Google
Google Inc., 20 West Kinzie, Chicago IL USA 60610
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
To post to this group, send email to firstname.lastname@example.org.
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/groups/opt_out.