Transcript Slides available

Discovering The Foundations Of A
Universal System of Ethics As A
Road To Safe Artificial Intelligence
Mark R. Waser
Inspiration
•
•
•
•
Current Culture/Movies (Terminator, I Robot)
Eliezer Yudkowsky/SIAI (Friendly AI)
Steve Omohundro (Basic AI Drives)
Hugo de Garis (The Artilect War)
The Goal
A universal ethical system that follows the 5 S’s
•
•
•
•
•
Simple,
Safe,
Stable,
Self-Correcting, and
Sensitive to current human thinking,
intuitions, and feelings
Points of Departure
• Current Culture/Movies (solely for excitement)
• Eliezer Yudkowsky/SIAI (wrong argument)
– Architecture vs. Behavior
– Fails S1, S4, and (for many people) S5
• Steve Omohundro (correct but fell short)
• Hugo de Garis (extrapolated the wrong thing)
Omohundro’s Basic AI Drives
• Derived from micro-economic theory & logic
• Tendencies which will be present in goal-driven,
self-improving systems unless counteracted
• Include desires for self-improvement, rationality,
utility-function preservation, self-protection, and
the effective acquisition and use of resources
• “Without explicit goals to the contrary, AIs are
likely to behave like human sociopaths in
their pursuit of resources.”
Rebuttal
Any sufficiently advanced intelligence (i.e.
one with adequate foresight) is guaranteed
to realize and take into account the fact that
not asking for help and not being concerned
about others will generally only work for a
brief period of time before ‘the villagers start
gathering pitchforks and torches.’
Fundamental Claim I
Acting ethically is an attractor in the
state space of intelligent behavior for
goal-driven systems
(if for no other reason than because others will ensure
that acting unethically will end up being
counterproductive to your goals)
James Q. Wilson, The Moral Sense
The real questions about human behaviors are not why we are so
bad but . . . . “how and why most of us, most of the time, restrain
our basic appetites for food, status, and sex within legal limits, and
expect others to do the same.”
In fact, we are generally good even in situations where social
constraints do not apply.
Axelrod & the Prisoner’s Dilemma
• Open-ended series of interaction among peers
• Winning strategies are all nice, retaliatory, forgiving (i.e.
selfish individuals for their own selfish good should be
nice and forgiving)
• Applicability of the durably iterated Prisoner’s dilemma is
much less restricted than it may first appear
• Results were extended with biologist William
Hamilton to draw numerous conclusions about
biological cooperation in terms of when, why, and
how it does and does not appear
In nature, almost without fail, . . .
• Cooperation (escalating to altruism) appears
to the extent that the cognitive machinery
exists to support it
– Blue Jays (Stephens, McLinn, & Stevens)
– Unrelated Cotton-Top Tamarins (Hauser, et al)
• The more intelligent a species is, the more
social it is
– Humans are obligatorily gregarious (de Waal)
Self-Interest vs. Ethics
• Higher personal utility • Higher global utility
(in the short term)
• Less risk (if caught)
• More options
• Lower cognitive cost
• Less restrictions
(fewer options, no need
to track lies, etc.)
• Assistance & protection
when needed/desired
Morals & Conscious Logical Reasoning
• Moral judgments are not products of, based upon, or even
correctly retrievable by conscious reasoning (Hauser)
– Allows a better chance of “talking your way out” of a defection
• The unconscious and emotions can seriously bias the
evaluation of moral dilemmas
• The deliberation-without-attention effect (Dijksterhuis)
shows the limits of the conscious mind in complex choices
• Abandonment of the concept of lifetime development of
universal logical moral reasoning (Piaget/Kohlberg/Gilligan)
Linguistic Analogy
• Both language and the moral sense are an innate faculty
with operative principles that cannot be consciously
expressed (Chomsky, Rawls)
• Cognitive Reuse - It is likely that language/morals were
originally mind-internal systems for thought and planning
that were later co-opted for their current purposes
• Chomsky developed a set of fundamental questions and
distinctions to create a model of how such a cognitive
facility can be studied
• Pinker studies cross-cultural invariants in language;
Hauser is doing the same for morality
Top-down design – Take I
Man: Ethics is that which is determined and
more importantly enforced by society
Machine: So . . . . Why shouldn’t I just
harvest you?
Man: Because there might be someone larger
than you out there who knows that how you
treat me is how you’ll attempt to treat them
when you can get away with it.
Top-down design – Take II
Man: Ethical behavior is that which is good for
you.
Machine: So . . . . Why shouldn’t I just harvest
you?
Man: Because we can form a long-term mutual
relationship that will, in the end, dwarf the
utility of simply harvesting me now.
Fundamental Claim II
Ethical behavior is simply that
behavior which promotes
healthy relationships
(Since healthy relationships are always better than the alternative)
The Relationship View of Ethics
• The ethics of an action is based solely on whether
or not it is a defection from the relationship
• Since relationships are based upon perceived
rather than actual utility, ethics are also judged that
way with each entity judging their own utility
– Makes ethics both culture- and context-specific
• Codified relationships are known as “contracts”
(or interface specifications for CS majors )
• Relational commitment is what makes loyalty an
ethical duty and leads to Singer’s circles of ethics
Fairness
• Refusing to exhibit fair behavior is a relationship
defection and an ethical violation
• If you can justify an action to the entities involved,
it is a fair and ethical action PROVIDED informed
consent and “no force, no fraud”
• An entity may not disagree with reasons that it
uses to justify its own behavior
Avoiding Selfishness
• Thomas Scanlon (a “contractualist”) tries to avoid
selfishness in personal relationships by appealing
to those “motivated to reach agreement” and
“reasonable disagreement”
• John Rawls does the same in A Theory of Justice
by using the “veil of ignorance” and “original
position” to extend the moral sense of fairness to
liberty and justice in society (and recognizing this
as a descendant of Locke’s social contract)
Scale Invariance
• Allows for reframing where our moral sense is not
well-evolved to handle relationships with larger
entities
– Self-to-country (taxes, tragedy of the commons); country-to-self
(equity, non-interference, taxes); country-to-country (trade barriers,
non-interference, refusal to negotiate, terrorism)
• Overrides the “contact principle” and incorrect
sympathy for defectors
• Line between an individual and a group will
become blurred with machine intelligences and
other future technologies
Reprise
•
•
•
•
•
Simple,
Safe,
Stable,
Self-Correcting, and
Sensitive to current human thinking,
intuitions, and feelings
• Scale-Invariant
Preparing for take-off
• As ye sow, so shall ye reap (so we really want to
clean up our act before then)
• The bad news – there’s an awful lot of room for
transition trauma to the first world given the
conditions imposed on the third world
• The good news – most enlightened individuals
seem to want to change the world by education
and agreement rather than by force
Next . . . .
Implementing Ethics As Part Of
An Autogenous Attentional
General Intelligence Architecture
Copies of this powerpoint available from [email protected]