Open in TheBrain

Bostrom 2012 SuperIntelligent Will Motivation

<p>Nov 01, 2012 11:25 PM

&#160;

Nick Bostrom 2012

&#160;

The SuperIntelligent Will: Motivation and Instrumental Rationality in Advanced Artif Agents

&#160;

Future of Humanity Instit. &#160;2012 &#160;Oxford...

&#160;

Orthogonality thesis: &#160;intel and final goals (purposes are orthogonal.

&#160;

Instrumental convergence thesis: &#160;will pursue similar intermediary goals.

&#160;

all human minds ... a small, tight cluster.

&#160;

Hannah Arendt and Benny Hill.

neural archit... identical.

&#160;

Orthogonality Thesis: &#160;intel and motivation are orthogonal and can freely vary.

&#160;

David Hume : belief and motivation are separate.

&#160;

arbitary high intel does not entail the acquistion of any such beliefs as are motivating on their own.

&#160;

&#160;

arbitar high intel but constitution so alien as to contain no function analogies to beliefs or desires.

&#160;

&#160;

1.3 &#160;Bizarre goals/motivations: &#160;sand-grain counting or paperclip-maximizing.

&#160;

synthetic minds can have utterly non-anthropomorphic goals;

&#160;

1) &#160;stabel pursuit of goals... it will simply get better at pursuing those goals.

&#160;

2) &#160;predictability thru inheritance: &#160;yes... but could be corrupted.

&#160;

3) instrumental convergence

&#160;

2.1 &#160;Self-preservation: &#160;&#160;longevity helps in achieving goals

&#160;

2.2 &#160;goal-content integrity -

software agents easily switch bodies and create exact duplicats, might swap memories,

download skills and radically modify thier cognit architecture and personalities.

might be better individuated as telological threads.

&#160;

2.3 creative enhancement: &#160;

an agent that has access to reliable expert advice may have little need for its own intel/ know.

&#160;

&#160;

suppose an agent is in a position to become the first superintel and thereby obtain a deisive advantage

enabling the agent to shape the future of Earth-origninate life and cosmic resources.

&#160;

2.4

&#160;

more efficient woys of transforming a set of inputs into valued outputs.

&#160;

improved engineerring, &#160;running more quickly, reliably... cheaper and with less energy.

&#160;

technology must be evaluated with ref to some agents final values.

&#160;

von Neumann probes: &#160;automatic, self-meding, self-replicating spaceships.

&#160;

2.5 &#160;Resource acquisiton: &#160;tech and resources facilitate physic construciton projects...

&#160;

value of resources dpends on uses to which they can be put...

&#160;

time, space, matter, energy &#160;can be put to almost any goal. &#160;cud be convered to life

run the superintel at greater speed, longer duariont, &#160;or create physical or simulated lives and&#160;

civilizaitons. &#160;backups, perimeter defenses, &#160;enhanced secutiry.

&#160;

Once von Neumann probes can be built... a large portion of the observ. universe cud be grdually colonized.

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;

&#160;</p>

Attachments