cloud superintelligence

Cloud superintelligence

Anders SandbergFuture of Humanity Institute

James Martin 21st Century School

Oxford University

Austria has around821,000,000 IQ points

Total work = individual work – communications losses

Cheap superintelligence

Sometimes simple tricks produce amazing results

Select best

Parallel trials (different agents)

Select best

Wikimagic

• Basic model: visitors to a page improve it if they think they can do better

Judge best

Evaluating complex results

Judge best

Multiple agents as units

Judge best

Feedback for further improvement

Select best

Scientific community

Evolutionary algorithm

Select best solution(s)

Variation

Information market

Price mechanism

Prices

Aggregated information

Why we need more people

• Why did the Tasmanians lose bone tools, fishing and warm weather clothing?

• Simple model– People imitate other people, innovate rarely– Number of technologies that can be retained

then depends on population size

Why blogs1 get smart

• Managing limited input channels• Model

– Agents read a limited number of information sources and have certain interests (many unique)

– Random agents post stories based on their interests– If one read agent has a piece of interesting news that source is

regarded as more useful by the agent– Agents write about the interesting news

with a certain probability, creatinginformation cascades

– Over time agents randomly shift to othersources if one proves useless

1: Or twitterers, journals, friends, etc…

Problem recogniser:Invoke suitable resource manager

Monitor/trouble-shooter

Resource manager: set up suitable problem solving structure

Superknowledge

• Often information can substitute for smarts

80 million tiny images

• http://people.csail.mit.edu/torralba/tinyimages/• Gather lots of example images, find nearest

neighbours of input image, deduce properties

Superorganisms and the posthuman realm