Re: Friendliness and blank-slate goal bootstrap

From: Rafal Smigrodzki (
Date: Sat Jan 10 2004 - 23:45:37 MST

----- Original Message -----
From: "Metaqualia" <>

> > the belief in an objective morality to which you are privy, a belief
> > infrequent at the sl4-level.
> If space and time can be relative, then morality can be absolute, no? :)
> You have not explained why an objective morality is impossible to
> You have just said that entertaining such thoughts is 'unworthy' just like
> heliocentrism was "unworthy of the light of day".

### I do not know if an objective morality (i.e. a system of moral reasoning
which forces itself on any rational, intelligent mind by its sheer
convincing inevitability, regardless of the mind's previous built-in
judgments) can or cannot be found. At the present level of our intelligence,
humans have not been confronted with such a system, or else all of us would
most likely be sharing it (unless the system ejoins its carriers to utmost
secrecy). Whether cognitively enhanced beings will discover such a system is
not likely to be legitimately predicted by us, either. However, while
descriptions of purportedly objective moral systems abound in some social
strata, to my knowledge they are distinctly unpopular on sl4, especially
ones involving the statement "Death is morally neutral". Such beliefs might
be worthy of sl4 discussion if supported by a formidable array of
sophisticated arguments, but so far such arguments have not been presented.


> It's ok to choose creating a morally-reasoning AI versus creating an AI
> a preset moral. But at this point, we do not even have an arrow. Go to the
> east, they will tell you that not killing your wife when she commits
> adultery is immoral. They will tell you that killing cows is more immoral
> than killing monkeys. What is an AI gonna make of all this? Average out
> moral/immoral statements? Would that make barbie dolls slightly more
> than killing your political enemy? Can't make a saint when you maintain
> right and wrong are relative!

### Nick Hay got it right: "help sentients, and others that can be helped,
with respect to their volitions -- maximise self-determination, minimised
unexpected regret". I would see achieving stable function along these lines
as the engineering challenge before Ben, Peter and Eliezer, since this
formulation feels ("feels" - I am waving the red-flag word of subjectivity
here) congenial to me, and whether someone could call it objectively right
or dismiss as objectively wrong carries little weight with me (unless such
pronouncement came from someone whose intellect I trust more than my own).
Similarly, I do not have the emotional need to claim objective morality on
my side if I were to act against a wife-killer - it's enough that I don't
like it on an emotional level and that I know (intellectually) that allowing
such deeds reduces the likelihood of my wishes coming true.

No, really, you don't need to have the one and absolute Truth to reject some
acts as wrong. An appeal to shared volition, some game-theoretic
simulations, and it is possible to come up with working solutions to moral
dilemmas, including a principled rejection of cultural moral relativism.

> There is confusion; as a species, we know nothing. How can we expect to
> create an AI who reasons about morality when we, the creators, can't agree
> on what this morality is?

### Of course, coming up with an AI that would satisfy the moral intuitions
of every single member of our species appears to be an impossible
undertaking - I see the goal of FriendlyAI research as merely building a
device which will stably perform according to the scheme summarized by
Nick - the simplified essence of a complex system of moral reasoning
embedded in the brains of many humans, including mine - which is why I
enormously like the idea of Friendly AI, as expounded by Eliezer. No doubt
some will claim Ben is Satan himself, when he unveils his BabyAI in a few
years - I don't feel much of a need to beg them for forgiveness (although
some bullet-proofing might be a good idea).


> I have already given an example of something that we may have completely
> missed because objective morality has not been talked about: "the AI may
> not reach any kind of moral belief after all, unless it can experience
> qualia". Without experiencing qualia, a rigorously logical system will
> to hard determinism and interpret your screams as a physical reaction of
> neuron activation and air flow from your mouth. A morally neutral physical
> process. Do we know better? Then let's find some _valid_ justifications
> it.

### It is an interesting hypothesis, postulating that a stable expression of
Friendliness (recognized as such by rational humans), requires the existence
of qualia in the AI. What it has to do with objective morality eludes me.
Would you mind telling me what do you mean by objective morality or "real"


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:45 MDT