1
Lethal Autonomy.Should there be a law against it?
Sander Rabin MD JDThe Center for Transhuman Jurisprudence, Inc.The Future of our Minds, Bodies and Genomes
legal #heet
at
Bots and Brains NYC15 December 2014
Nuclear Engineer
Ophthalmic Surgeon
Litigator
The Center for Transhuman
Jurisprudence
journey to #heet
Patent Attorney
2
3
Introduction
Binary Thinking | Ethics |Philosophy |Questions
4
1. Should there be a law against Lethal Autonomy?
5
2. Is it intelligent to grant lethal autonomy to artificial Intelligence?
6
3. Will Lethal Autonomy become Anthropomorphically Lethal?
7
4. What are we teaching our child?
8
An Artificial Intelligence Insight Structure
Seeing through the Hype
9
The mind is a set of mental faculties that enables cognition, i.e.
Imagining
Memorizing
Thinking
Intending
What is the Mind?
10
Although intuitive, eludes definition; maybe the sum of:
Awareness: Ability to Experience the World
Perception: Transduction, Interpretation
Sentience: Ability to Sense, Feel, or Experience
Self-Awareness: Ability to Experience the Self
What is Consciousness?
11
▪︎ The Self is an idea: an integrated system for representing a human
sustained over changing patterns of neurosynaptic connections and activity
What is the Self?
12
Qualia & NCC: How are phenomena experienced?Mental states are experienced subjectively in different ways
by different people, e.g., seeing red or feeling pain
13
The Explanatory GapExperience arises from a physical basis, but there is no explanation
of why and how the physical becomes mental
14
The hard problem of consciousness
How and why do we have qualia?
Is consciousness Turing computable?
The Hard Problem of Consciousness
How does a lump of fatty tissue and some electricity give rise to the experience of perceiving, meaning, or thinking?
15
Knowledge is UnderstandingKnow-how that informs Do-how
16
Understanding is Explanation is Data Compression, Rules & Modeling
17
Explanation is Description that fits the New into the Prevailing
Paradigm
18
▪︎ Turing Test and The Chinese Room?
Problem: Information = Representation + Interpretation(i.e., meaning)
Turing Test Intelligence: Symbolic knowledge (representation) plus logical manipulation
Turing Test: Machine intelligent if responses to questions indistinguishable from human – a successful masquerade
Chinese Room: thought experiment against conclusion that computer passing Turing Test is intelligent
How does Information acquire Meaning?
19
▪︎ No consensus on definition
▪︎ Intelligence needed to:
▫︎ communicate in natural language
▫︎ memorize ▫︎ learn ▫︎ plan ▫︎ solve ▫︎ judge adapt
▫︎ perceive to acquire knowledge ▫︎ think to create & represent knowledge ▫︎ reason to use knowledge
may be defined as ‘efficient cross domain optimization’
What is Intelligence?
20
AI AGI WBE
WBE: mind uploading approach to AGI - scan & map bio brain & copy data to computer that runs indistinguishable simulation model
AGI: system that performs any intellectual task that a human being can perform
AI: system that senses environment & takes action maximizing chances of it’s success
Artificial Intelligence
21
Emergent ▫︎ Reductive ▫︎ Contextual Contingent?
22
What is Artificial Emotion?Simulation of motor-driven behavior (laughing, crying) is programmable.
But can AI have feelings without basis in organic neural sensation – the evolutionary origin of emotion?
Is Artificial Emotion Supportive or Subversive?
23
What is Conscience?Moral compass faculty: Distinguishes personal right from wrong,
with either remorse or satisfaction following action
MIND
CONSCIOUSNESS | COMPUTABILITY | LAW OF PHYSICS
SELF & SELF AWARENESS
FROM SENSORS TO SENSATION | QUALIA & SUBJECTIVE EXPERIENCE
WAY OUT OF THE CHINESE ROOM: NONBIOLOGICAL ATTRIBUTION OF MEANING
EMOTIONAL INPUT INTO DECISION-MAKING & JUDGMENT
WHOSE CONSCIENCE?
MUST THE HARD PROBLEM BE SOLVED?
Threat AssessmentAuthentic Non-biological Intelligence
25
Deployment of Lethal Autonomy
26
Opponents point to:
▪︎ unconscionably changing quintessential meaning of war
▪︎ amplifying extremist power & political destabilization ▪︎ increasing propensity to wage war
anti-Americanism ▫︎ losing battle for peoples' hearts and minds
What are Cons of LAWs?
27
Like UAVs, LAWs are effective weapons that comply with IHL with no risk to US soldiers
Are Proportional ▫︎ Do not inherently cause unnecessary suffering
Are Non-Indiscriminate in Targeting ▫︎ Distinguish Civilians from Combatants
Satisfy Military Necessity: ▫︎ kill leaders ▫︎ disrupt terrorist networks ▫︎ instill sense of insecurity
What are Pros of LAWs?
28
International Humanitarian Law seeks to limit war (Rules of War) :
LAWs not inherently unlawful under IHL
by restricting combatants’ means & methods of warfare:▫︎ Discrimination ▫︎ No Needless Suffering ▫︎ Proportionality
by protecting persons who are not combatants: ▫︎ Distinction
Is Lethal Autonomy Legal?
Governing Lethal Behavior: Embedding Ethics in a Hybrid Deliberative/Reactive Robot Architecture
People outsource wide range of moral questions to friends, peers, experts, writers, and public figures. Can a machine do any worse?.
Goal is to create ethical-decision programs for LAWs that perform better ethically than human soldiers in combat.
LAWs can be used in self-sacrificing manner
LAWs can be designed without fear, hysteria, rage, frustration, shell-shock, etc. that clouds human judgment; may perform better than humans in fog of war
LAWs can avoid human problem of ‘scenario fulfillment’ - distortion or neglect of information that does not fit pre-existing beliefs
LAWs can program ethical decisions that govern LAWs consistently with Laws of War & Rules of Engagement
LAWs may be capable of independent, objective monitoring of combat behavior by all parties and reporting ethical infraction
Machine Morality v. Human Morality
30
Arguments over legitimacy of weapons ancient. Arms race makes LAWs inevitable and ban unworkable
Ban won’t stop back market sales
Also assigning potentially lethal tasks to nonmilitary machines, e.g., driverless cars
People accepting lethal autonomy in nonmilitary machines will expect same tech in war.
Best way to adapt IHL to LAWs is global dialogue for common standards and best practices.
1
2
3
4
5
Human Morality as Measure: Unworkable Bans and Unenforceable Treaties
31
Superintelligent Self-Replication
32
▪︎ Intellect smarter than best human brains
If bio brain is physical system, simplest superAI may be WBE operating faster than bio brain
GE, SynBio, Neurotech may create bio superAI
Bio constrains processing speed & size of human brain
Bostrum: SuperAI simply dominant at goal-oriented behavior. Avoids questions of intentionality (Chinese Room) or consciousness (Hard Problem)
What is Superintelligence?
33
34
35
“What is the answer?” “That depends on why you’re asking the question.”
3. Will Lethal Autonomy become Anthropomorphically Lethal?DOUBTFUL
2. Is it intelligent to grant lethal autonomy to artificial Intelligence?PROBABLY NOT BUT LIKELY NECESSARY
1. Should there be a law against Lethal Autonomy? NO
4. What are we teaching our child?COHERENT EXTRAPOLATED VOLITION
ANSWERS
36
Merge with the Child?
37
Do what we would have told you to do if we thought as fast as you do and could consider many more possible lines of moral argument
Do what we would have told you to do if we knew everything you know
Program the AI to:
Do what we would tell you to do if we had your ability to reflect on and modify ourselves
Initially design AI to learn human values by looking at humans, asking questions, scanning human brains, rather programming with fixed set of imperatives
Lesson Plan
38
A First Word Summit on Human Enhancement Enabling Technology
economic #heet
Coming in 2015
http://www.tranhumanjuris.com @transhuman juris
39
Future’s Greetings