Convergent or Divergent Superintelligence

Document Sample
Convergent or Divergent Superintelligence Powered By Docstoc
					Convergent or Divergent
   Superintelligence
    We need a “smart systems
            theory”
• What is the Evolution and Ecology of
  Intelligence?
  – Under what circumstances do we see complex
    organisations of intelligence of different kinds,
    and under what circumstances do we see
    monocultures?
  – Economy and sociology as the ecology of
    intelligence?
        Hard or soft takeoff?
• Runaway self improvement leading to
  single “winners” - a “spike”
• More gradual self improvement leading to a
  broad front - a “swell”
• The scenario assumed will determine policy
  to a large extent
• Do we have any evidence for either
  scenario?
 General vs. specific intelligence
• How powerful and easy to design?
• Likely a strong determinant of hard vs. soft
  takeoff
      Reasons to think general
        intelligence is hard
• No free lunch theorems
• The Bias-Variance dilemma
  – Specific bias vs. slow learning rate
• No strong evolutionary trend towards GI
  – either very expensive or hard to achieve
  Is intelligence increase easy or
                hard?
• Difficulty of improving intelligence might
  remain constant, increase or decrease
  compared to current level
• Different mental architectures have
  different affordances and might be able to
  develop not just at different rates in
  different domains, but might be able/unable
  to solve certain problems.
One kind of intelligence or many
• Evolution shows that the problem of
  survival can be solved in a large number of
  radically different ways
• Intelligence models
  – Top-down AI with recursive search
  – Subsumption architectures of linked behaviours
  – Calculate P and reformulate problems in terms
    of Turing questions
  – Collective intelligence
     Intelligence amplification
• IA does not need to understand the human
  brain except for extreme enhancement. It
  can be an user interface issue
• Gradual acclimatisation
• Effective intelligence might be more
  relevant than “real” intelligence
                A scenario
• Economics drives development of specific
  intelligent systems
• General intelligence as a result of specific
  intelligence
• Intelligence fairly
  hard but useful
• Plurality of intelligent systems
• Soft take-off
• Intelligence amplification
  – Exoselves
• When GI appears, it will be integrated into
  the existing systems rather than become
  something independent

				
DOCUMENT INFO
Description: Brought to you by the peeps at Chez Tay - http://www.cheztay.com/chocolate.html