Re: [wta-talk] My own A.I project: 'Show-down with Singularity Institute"

Your message is very Randian indeed, but surely you don't object to my
building an intelligent machine to serve my purposes. At what point, along
the scale of intelligence, does building a machine become evil?

Suppose I could bypass the reward-pleasure mechanisms in my brain and
effortlessly become happy without earning the happiness as a reward. Would
that be evil?

Frank

On 2002-10-15, Marc Geddes opined [message unchanged below]:

> Thought you all may be interested to know that I've become convinced
> that a 'hard-take-off' Singularity scenario is a possibility which
> needs to be taken seriously. I'm now thinking of 'going for broke'
> and throwing everything I've got into creating a self-improving
> general artificial intelligence of my own. I've been developing some
> ideas for years, and I believe I could plunge straight into 'coding'
> now. I just need a bit of time and some funding.
>
> If you've been following all the 'digs' I've taken at 'Singularity
> Institute' on wta-talk it should be clear that I think that their
> approach is wrong. I think that 'Friendliness' is the wrong super-
> goal for an A.I. I reckon that all intelligent sentient beings
> are 'ends in themselves' and to create A.I's for the purposes of
> serving us humans is misguided. Further, I fell that it's safer to
> try for A.I's to be as 'human-like' as possible, rather than
> attempting to create creepy 'friendly transition guides' That's why
> I prefer an A.I design with 'bounded selfishness' rather than an A.I
> with 'Friendliness'
>
> I've read Eliezer's 'Friendly A.I' and I can only shake my head in
> wonder that someone so brilliant could be so badly wrong ;) There's
> so many severe conceptual errors running all the way through
> the 'Singularity Institute' documents that it's painful.
>
> The idea of a 'post human' ,machine saviour who is going to solve all
> the worlds problems is just plain nuts in my view. To my way of
> thinking, each individual has to take responsibility for their own
> lives, and if a person is unhappy in most cases it's their own
> fault. Having some 'magic genie' (i.e. Friendly A.I) come along and
> start granting everyone 'wishes' in the belief that this will make
> everyone happy is puerile nonsense. I have no doubt that if you took
> any person on Earth who is currently miserable and gave them a magic
> genie who could grant their every desire, they could go on making
> wish after wish after wish until they were blue in the face but they
> would still be just as miserable as they were to begin with! Nor
> will any form of 'happiness drug' or cognitive enhancement change
> this.
>
> My experience is that true happiness only comes when a person
> consistently strives to act according to their own ideals and through
> their own efforts. It's self-responsibility that's the key - it's
> the striving itself not the end result. This can be seen with animal
> experiments also. When animals feel they are in control of a
> situation and are doing things through their own efforts, that is
> when they are satisfied. Simply dishing out 'rewards' to animals in
> these experiments appears to have no effect on their degree of
> happiness. So happiness appears to be a function of what self-
> improvement guru's refer to as 'self-fulfilment' and it's actually
> quite a subtle concept. 'Self-fulfilment' cannot be achieved by
> simply giving someone things (for instance 'Friendly A.I') nor can it
> be achieved through some artificially induced mental state (for
> instance drugs). Self-fulfilment is a *process* not a *thing*. It
> is a correspondence between a person's actions and their highest
> ideals .
>
> This is where I think many transhumanists are going wrong. All too
> often, the implicit belief seems to be that the post-human world will
> be some sort of utopia - that we'll all be happy once we get the
> right technology. Like I said...puerile nonsense.
>
> All this is related to the age-old left-wing/right-wing political
> debate <sigh> The socialistic view is that a person's happiness is
> dependent on external factors - that we're all puppets of 'social
> forces' and if one is miserable it's all the fault of 'society'.
> This view leads to 'utilitarian' ethical systems where the belief is
> that it's some social 'end state' that is important. Like I said, I
> emphatically reject this. In my view, every miserable person has no
> one to blame but themselves for their current misery. That's why my
> ethical system is a form is Aristotelian virtue ethics. If you read
> through 'Creating Friendly A.I', the left-wing bias in there is so
> blatant that it's almost humorous. So now it should be clear why I
> reject 'Friendly A.I'.
>
> No matter how fast a post-human is improving, no matter how powerful
> a post-human has become, at any specific time X that post-human will
> have finitely specifiable limitations. At time X any given post-
> human will have access to a finite amount of material resources, a
> finite processing speed, a finite intellectual capacity and laws of
> nature imposing finite limits. Any 'action' undertaken by a post-
> human will have a 'cost' in the sense that it is diverting resources
> from a finite pool - resources that could have been used for some
> alternative purpose. For a 'post-human' to waste resources by
> being 'Friendly' and granting ;wishes' to every human bozo who asks
> is in my view absurd. It is the post-human itself who would
> understand the most effective use of it's resources, and as I have
> argued, each intelligent sentient should be regarded as in end in
> itself. Since each sentient is ultimately responsible for their own
> happiness (Aristotelian virtue ethics ) the happiness of other
> sentients should not be the super-goal of a post-human. That is not
> to say that a post-human cannot help others as a sub-goal. I'm just
> saying that helping others should not be a sentient's main goal.
>
> An A.I should take it's own life as a super-goal, subject only to
> the 'boundary condition ' that it should not violate the rights of
> other sentient's. (Randian Selfishness). It's other super-goals
> should basically represent the philosophy of transhumanism. I take
> this to consist of:
>
> *Mark Walker's 'perfectionism' ethics. A 'striving for self-
> perfection' should be one of the super-goals.
>
> *Curiosity and Creativity. A desire to learn and teach should be a
> super-goal (curiosity) and a desire to create and appreciate
> beautiful things should also be a super-goal. (creativity)
>
> When aiming for true A.I, we should be aiming to create a 'self-
> centred' self-improving intelligence pursuing the above mentioned
> goals. All S.A.I's should be regarded as citizens with full rights
> and we should not expect that they be 'Friendly' or that they serve
> humanity. The troubles of the world should not be the responsibility
> of post-humans.
Received on Sun Oct 27 10:54:59 2002