Ethical

1
Ethical Mathematical Adventures in Coding
Better Children

• Jeff MedinaSenior Systems Engineer, Lockheed
  Martin TSSAssociate Director, The Singularity
  Institute for Artificial IntelligenceFellow, The
  Institute for Ethics and Emerging Technologies

2
But firstA Brief Digressionto Bring
Everyone-Less-eto Near-Agreement onthe
Presumption of the Moral Significance of
Behaviorally-Personish Artificial Beings
3

• Moral Equivalence in Practice MEIPAn
  artificial being should be treated at least as
  well as if she were a natural being in the
  nearest cognitive-behavioral class.
• SO, for example,
• treat artificial people like people,
• and artificial puppies like puppies,
• but treat artificial puppies-with-people-level-m
  inds like people, and so forth.
• More succinctly, TREAT ALIFE LIKE LIFE.

4
Why MEIP? A brief argument

• P(x in Y For nearly all F, yF ? xF) 0,and
  arguably 0.5
• If (x in Y) humans act as if (x not in Y),
  massive suffering ensues, scaling linearly with
  C(X), with C(X) itself plausibly growing
  exponentially.
• In the opposite case, the expected neg. utility
  (from wasted resources) is less than above due to
  ALife per capita resource efficiency.

5
What responsibility do we have for the lives we
create?

• ParentsGodsand A.G.I. developers?All quite
  similar, as it turns out

6
Case 1The responsibility of parents toward
their children

• Procreative Beneficence(Savulescu 2001)Parents
  should select the child, of the possible
  children they could have, who is expected to have
  the best life, or at least as good a life as the
  others, based on the relevant, available
  information.

7
Analogue to Case 1For those who would build (or
birth) A.G.I.

• Programmatic Beneficence!(Medina, just now)The
  A.G.I.-expectant should select the A.G.I., of the
  possible A.G.I. they could build, who is expected
  to have the best life, or at least as good a life
  as the others, based on the relevant, available
  information.

8
Case 2Theodicy Epicurus et al.s problem of
evil

• If (insert favorite deity here) is
• maximally morally good? wants to help us
• omnipotent? is capable of helping us
• and omniscient..? knows how to help us

9
WHY DOES LIFE SUCK?!?(relatively speaking)
10
This reasoning reverse-implicates obligations for
A.G.I. creators
With greater control over a sentient beings life
comes greater responsibility to do right by it.
11
But how much is enough?

• Do we satisfice? Arbitrary! Nooooo
• Do we maximize? Yes! but how?

12
Considerations

• The intelligence of the A.G.I. designer
  constrains dI given n resources and t time.
• Does finding further improvements require similar
  intelligence resources, or more? If more, how
  much more? Is scaling linear, exponential,
• Create the A.G.I. at intelligence equal to or
  greater than the creators, and provide n
  resources to the A.G.I. for use in improving
  herself?(Bad idea in the case of humans, but
  maybe less so for A.G.I. How malleable is your
  design after the initial developmental period?)