THE NUTS & BOLTS OFIMPLEMENTING A SAFE
MOTIVATIONAL SYSTEM
Mark R. WaserDigital Wisdom Institute
Outline
What is a “safe” motivational system?
How do we ensure that it happens (and sticks)?
2
What is a “safe” motivational system?
*ANYTHING* that reliably leads toETHICAL BEHAVIOR
3
What is Ethical Behavior?
4
The problem is that no ethical system has ever reached consensus. Ethical systems are completely unlike mathematics or science. This is a source of concern.
Entities Require Ethics
Ethics are “rules of the road” Necessary for “safe” interaction
Yet, we cannot come to a consensus about them
There is something horribly wrong with this picture
5
The Human Moral System Is primarily implemented via emotions
Is not transparent or reflective
Frequently conflicts with “rationality”
Is “clearly” subjective
6
Humans are . . . . Evolved to self-deceive in order to better deceive
others (Trivers 1991) Unable to directly sense agency (Aarts et al. 2005) Prone to false illusory experiences of self-authorship
(Buehner and Humphreys 2009) Unable to correctly retrieve the reasoning behind
moral judgments (Hauser et al. 2007) Almost always unaware of what morality is and why
it should be practiced . . . .
Inflammatory Statements
>Human intelligence REQUIRES ethics All humans want the same things Ethics are universal Ethics are SIMPLE in concept Difference in power is irrelevant (to ethics) Evolution has “designed” you todisagree with the above five points
The Origin of Morality Selfishness predictably evolves Reciprocal altruism predictably evolves
But requires cognitive complexity to ensure that is is not taken advantage of
Ethics predictably evolves As an attractor in the state space of behavior
because community is so valuable But altruistic punishment is a necessity
Arms Race between Individual benefits of successful personal cheating
(really only in a short-term/highly time-discounted view) Societal benefits of cheating detection &
prevention9
Haidt’s Functional Approach
Moral systems are interlocking sets of
values, virtues, norms, practices, identities, institutions, technologies, and evolved
psychological mechanisms that work together to
suppress or regulate selfishness and
make cooperative social life possible 10
How to Universalize EthicsQuantify/evaluate
intents, actions & consequences with respect to
codified consensus moral foundations
Permissiveness/Utility Function
equivalent to a “consensus” human (generic entity) moral sense
11
Instrumental Goals/Universal Subgoals(adapted from Omohundro 2008 The Basic AI Drives)
Self-improvement Rationality/integrity Preserve goals/utility function Decrease/prevent fraud/counterfeit utility Survival/self-protection Efficiency (in resource acquisition & use) Community = assistance/non-interference
through GTO reciprocation (OTfT + AP) Reproduction
Human Goalssurvival/self-protection & reproduction
happiness & pleasure------------------------------------------------------------------------------------
community-------------------------------------------------------------------------------------
self-improvementrationality/integrity
reduce/prevent fraud/counterfeit utilityefficiency (in resource acquisition &
use)
Human Goals & Sinssuicide (& abortion?)
masochism------------------------------------------------
selfishness(pride, vanity)
-------------------------------------------------
acedia (sloth/despair)
insanity
wire-heading (lust)
wastefulness (gluttony, sloth)
murder (& abortion?)cruelty/sadism
-------------------------------------------------
ostracism, banishment
& slavery (wrath, envy)
----------------------------------------------------
slavery
manipulation
lying/fraud (swear falsely/false witness)
theft (greed, adultery,coveting)
survival/reproductionhappiness/pleasure
-------------------------------------------------
community(ETHICS)
--------------------------------------------------
self-improvement
rationality/integrity
reduce/prevent fraud/counterfeit
utility
efficiency (in resource acquisition
& use)
Haidt’s Moral Foundations1) Care/harm: This foundation is related to our long evolution as mammals with attachment systems and an ability to feel (and dislike) the pain of others. It underlies virtues of kindness, gentleness, and nurturance.
2) Fairness/cheating: This foundation is related to the evolutionary process of reciprocal altruism. It generates ideas of justice, rights, and autonomy. [Note: In our original conception, Fairness included concerns about equality, which are more strongly endorsed by political liberals. However, as we reformulated the theory in 2011 based on new data, we emphasize proportionality, which is endorsed by everyone, but is more strongly endorsed by conservatives]
3) Liberty/oppression*: This foundation is about the feelings of reactance and resentment people feel toward those who dominate them and restrict their liberty. Its intuitions are often in tension with those of the authority foundation. The hatred of bullies and dominators motivates people to come together, in solidarity, to oppose or take down the oppressor.
4) Loyalty/betrayal: This foundation is related to our long history as tribal creatures able to form shifting coalitions. It underlies virtues of patriotism and self-sacrifice for the group. It is active anytime people feel that it's "one for all, and all for one."
5) Authority/subversion: This foundation was shaped by our long primate history of hierarchical social interactions. It underlies virtues of leadership and followership, including deference to legitimate authority and respect for traditions.
6) Sanctity/degradation: This foundation was shaped by the psychology of disgust and contamination. It underlies religious notions of striving to live in an elevated, less carnal, more noble way. It underlies the widespread idea that the body is a temple which can be desecrated by immoral activities and contaminants (an idea not unique to religious traditions).
15
Additional Contenders Waste
efficiency in use of resources Ownership/Possession
efficiency in use of resources; Tragedy of the Commons
Honesty reduce/prevent fraud/counterfeit utility
Self-control Rationality/integrity
16
How to Universalize EthicsQuantify/evaluate
intents, actions & consequences with respect to
codified consensus moral foundations
Permissiveness/Utility Function
equivalent to a “consensus” human (generic entity) moral sense
17
Critical Components I:Self-Knowledge & Reflection
A self must know itself to be a self Composed of three parts:
The running processes (consciousness) The personal knowledge base (memory) The physical hardware (body)
Must start with: A competent model of each Sensors to detect changes and their effects
*MUST* “care” about itself (motivation)
Critical Components II:Explicit “Anchor” Values
Do not defect from the communityDo not become too large/powerful
Acquire and integrate knowledge
Instrumental goals
Critical Components III:
Reliability Self-Control, Integrity, Autonomy, Responsibility
In “predictive control” of its own state and that of the physical objects that support it Yes! This is a marked deviation from the human
example.
ArchitectureProcesses will be divided into three main classes: Operating system processes Subconscious/tool processes One serial consciousness/learner process (CLP)
The CLP will be able to create, modify and/or influence many of the subconscious/tool processes.
The CLP will NOT be given access to modify operating system processes Indeed, it will have multiple/redundant logical, emotional &
moral reasons to seriously convince it not to even try
Operating System Architecture
Open, Pluggable, Service-Oriented/Message-Passing Quickly adopt novel input streams Handle resource requests and allocation Provide connectivity between components
Safety Features Act as a “black box” security monitor capable of reporting problems
without the consciousness’s awareness Able to “manage” the CLP by manipulating the amount of processor
time and memory available to it (assuming that the normal subconscious processes are unable to do so)
Other protections against hostile humans, inept builders, and the learner itself may be implemented as well
Automated Predictive World Model Is the most important subconscious process(es) Will serve as an interface to the “real” world
The CLP will live in a virtual world (just as we do) Will be both reactive and predictive Will generate “anomaly interrupts” upon deviations
from expectations as an approach to solving the “brittleness” problem (Perlis 2008)
Will contain certain relatively immutable concepts to serve as anchors both for emotions and for ensuring safety (trigger patterns – Ohman et al. 2001)
Anchors & Emotions Anchors create a multiple attachment point model which is
much safer than the single-point-of-failure, top-down-only approach of “machine enslavement” advocated by the MIRI (Yudkowsky 2001)
Emotions will be generated by the subconscious processes as “actionable qualia” to inform the CLP and will also bias the selection and urgency tags of information relayed via the predictive model
Violations of the cooperative social living “moral” system will result in a flood of urgently–tagged anomaly interrupts demanding that consciousness resources be expended to “solve the problem”
Conscious Learning Process (CLP)
The goal is to provide as many optional structures and standards to support and speed development as much as possible while not restricting possibilities beyond what is absolutely required for safety.
We believe the best way to do this is with a blackboard system similar to Learning IDA (Baars and Franklin 2007).
The CLP acts like the Governing Board of the Policy Governance model (Carver 2006) to create a coherent, consistent, integrated narrative plan of action to fulfill the goals of the larger self.
The Digital Wisdom Institute is a non-profit think tank focused on the promise and challenges of ethics,
artificial intelligence & advanced computing solutions.
We believe that the development of ethics and artificial
intelligence and equal co-existence with ethical machines is
humanity's best hope
http://DigitalWisdomInstitute.org26