Introduction to Artificial Intelligence for Public Health

Lecture 14: Philosophy of AI

Artificial intelligence is a controversial area for many reasons: whether we can create AI, whether we should create AI, and what might happen if we did create AI are all open questions. To help answer the question of whether machines can think, philosophers have devised Gedankenexperiments as "intuition pumps" to help illuminate the issues, with colorful names like the Turing Test and the Chinese Room, and colorful questions like "What is it like to be a bat?" and "Aliens eat fading qualia". As for whether we should create AI, many ethical issues arise in the quest to create slave machines to replace human workers. Finally, we do not know what effect creating true AI would have, both on our views of ourselves as unique, on our jobs, on our economy, and on the future of our species.

Outline

• The Big Questions
  • Can we? Is AI possible?
  • Should we? Is AI desirable or ethical?
  • What if we? What would be the effect of successful AI?
  • Since we are, how could we do it better? How do we pursue AI responsibly?
• Artificial Intelligence as Computational Philosophy

Readings:

• Artificial Intelligence: Chapters 23 and 25
• Machines Who Think: Chapters 8 ,9, 13, and 14

The Big Questions

Philosophical Issues

• Can we? Is AI possible?
• Should we? Is AI desirable or ethical?
• What if we? What would be the effect of successful AI?
• Since we are, how could we do it better? How do we pursue AI responsibly?

Intuition Pumps

• "Images, stories and analogies that give us something vivid and concrete to help us understand what would otherwise be obscure and abstract."
  - Baggini & Fosl
• Originally a derogatory term by Daniel Dennett
• Help us unpack our intuitions into manageable forms
• Can lead us astray
  • Intuition is no substitute for argument
  • Arguments are no substitute for models
  • Models are no substitute for evidence
  • Evidence is no substitute for judgement
  • Judgement is no substitute for verification
• Thought Experiments
  • Originally German - "Gedankenexperiments"
  • Experiments that you can carry out in your head
  • Examples: "What if you were in a train moving at the speed of light?"
• Deep Questions
  • Questions which have no easy answer
  • Designed to expose gaps in our conceptual systems
  • Examples: "What happens if an unstoppable force meets an immovable object?"
• Philosophical Analogies
  • Designed to illustrate difficult to grasp concepts
  • Often used to popularize concepts - e.g., "selfish genes"
  • Examples: "Each stage of life is like a slice of a carrot"

The Debates

• Fundamentals
  • Man vs Machine
  • Reason vs Emotion
  • Syntax vs Semantics
• Functionalism
  • Strong AI vs Weak AI
  • Programs vs Powers
  • Mind vs Matter
• Stylistics
  • Reductionism vs Holism
  • Emergent Behavior vs Following Rules
  • Connectist vs Symbolic
  • Serial vs Parallel
• Methodology
  • Cognitive Psychology vs Artificial Intelligence
  • Computational vs Biological
  • Top-Down vs Bottom-Up

Could we?

Can a machine think?

• Questions
  • Can mere matter think, or is thinking a part of a distinct realm?
    • Dualism: Thought is distinct entity from matter.
    • Materialism: Thinking is a property of matter or its organization.
  • Can a machine think, or is thought particular to humans?
    • Uniquely Human: Thinking is something that only humans do - a property of (human) brains, or maybe (human) language.
    • Universalism: Thinking is a phenomenon that could happen in any kind of system.
  • Can a computer think?
    • Uniquely Brains: Thinking is a property of living brains.
    • Computationalism: Thinking is a process that can happen in any medium.
  • Can a program think?
    • Uniquely Bodies: Thinking arises from agents interacting with environments.
    • Functionalism: Thinking is an abstract property of complex processes.
• Intuition Pumps
  • The Turing Test
    • Designed to investigate:
      • Can a machine behave in a way we'd call thinking?
      • Take faces, facings, bodies, housings, brains, wires out of equation
    • The Turing Test
      • The Challenge: Tell whether someone is a man or woman
      • The Setting: A teletype array with a judge talking to a man and a woman
      • The Goal: Can the man fool the judge into thinking he's a woman?
      • The Twist: If you replace one of the contestants with a computer, how will they fare?
      • Examples: Simon's Rock College 2005 Turing Test
    • The Imitation Game
      • The Challenge: Tell whether someone is a human or computer
      • The Setting: A teletype array with a judge talking to contestants
      • The Goal: Can a computer fool the judge into thinking it is a person?
      • Examples: The Loebner Prize
  • The Chinese Room
    • Designed to illustrate:
      • Can't get from syntax to semantics
      • Programs lack "intentionality" (aboutness)
      • Understanding requires an understander
    • Searle vs Schank: What is "understanding"?
    • The Interpreter and his Room
    • Replies to the Chinese Room
    • The Interpreter and his Memory
    • Replies to the Replies to the Replies
  • A Conversation with Einstein's Brain
    • Dennet and Hofstadter
    • Adding Dials to the Chinese Room
      • Substrate of Simulation
      • Accuracy of Simulation
      • Physical Size
      • Type of Interpreter
      • Speed of Operation
    • Einstein's Brain Now Out in Hardback
    • Haugeland's Brain Now Under New Management
• Opponents of AI
  • Lovelace - Computers do what we tell them to do
  • Taube - Computers are formal, but Godel shows thought is not
  • Dreyfus & Dreyfus - Wrong philosophical tradition - phenomenology, not rules
  • Lighthill - Incorrect analogy from physics
  • Weizenbaum - Computers do not exhibit judgement
  • Winograd & Flores - Computers are designed in a context which we are blind to
  • Searle - Cannot get from syntax to sematics
  • Nagel - We can never know what it is like to be a bat
  • Lucas - Rehabilitated Taube's Godelian incompleteness argument
  • Penrose - Mathematical intuition is not formalizable
• Cautionary Voices
  • von Neumann - Don't anthropomorphize the damn things
  • Dijkstra - Theorem proving is OK but "by describing machinery in anthropomorphic terms and man in mechanistic terms, it has only added to the confusion" about our understanding of the mind.
  • Feynman - Psychology may not be sufficiently lawful to be systematized

    If I'm in the tribe and I'm sick, I go to the witch doctor. He knows more about it than anyone else. But I keep trying to tell him he doesn't know what he's doing and that someday when people investigate the thing freely and get free of all his complicated ideas they'll learn much better ways of doing it. Who are the witch doctors? Psychoanalysts and psychiatrists, of course.

• Advocates for AI
  • Turing - Judge mind by its behaviors
  • Simon - If we understand reason in human affairs we can apply it to machines. (We may find emotion is perfectly reasonable!)
  • Minsky - Human minds are societies; our machines should be too
  • McCarthy - Commonsense is actually highly evolved logic; if we understand that, we will understand the mind
  • Hofstadter - Minds are Programs, but most of the action goes on below the level of symbolic processing
  • Haugeland - Humans minds are semantic engines, which we can build with computers
  • Dennett - The Intentional Stance: Mind is whatever is sufficiently complex for us to take the stance towards it that it has a mind
  • Schank - To understand how humans understand well enough to truly understand it, we'll have to implement that understanding in a computer.
  • Moravec - Download me - future computers will be able to handle it.
  • Kurzweil - Ditto.
  • Brooks - If we don't try to capture the evolutionary stages of progress, we'll miss something cruical.
  • Nilsson - Only by pursuing grand unified theories of mind will we progress.

Can a machine have emotion?

• Thought Experiments (Intuition Pumps)
  • Eliza and Parry
  • I'd like to thank the Academy
  • Data's Emotion Chip
• Arguments against emotional machines
  • Emotion is not mechanical or formalizable
  • Emotional behavior can be captured but not feeling
  • Emotion may be possible but it's not needed
• Arguments for emotional machines
  • Simon - Emotion helps us decide what we value
  • Minsky - Emotion helps us figure out what we pay attention to
  • Ortony - Emotion reflects how we view our status in the world
  • Frijda - Emotions are behavior packages attached to our core concerns
  • Sloman - Emotions are a byproduct of older reasoning mechanisms

Can a machine be conscious?

• Intuition Pumps
  • Intentional Zombies
  • Aliens Eat Fading Qualia
  • What it is like to be a bat?
  • Spock in a Box
• Does Consciousness Exist?
  • Rand - Consciousness is a distinct property of the universe
  • Norstrander - Consciousness is an illusion of our information processing
  • Neuropsychology - Different parts of the brain stay active when conscious
• Arguments against conscious machines
  • Searle - Can't get from syntax to semantics
• Arguments for conscious machines
  • Edelman - Consciousness is a tool we use to integrate the blur of life into one experience

Can a machine have free will?

• Issues
  • Physical Determinism
  • Algorithmic Unpredictability
  • The Feeling of Free Will
• Arguments against free will in machines
  • Ada Lovelace - Can only do what we tell them to do
  • Mortimer Taube - Learning must be formal, thus limited
  • Roger Penrose - Quantum effects enable mathematical intuition
• Arguments for free will in machines
  • Feynman - Predictable Predictability Failures
  • Francis - Drawing a Line Around What Explains Behavior

Could you "download" someone's brain into a machine?

• Thought Experiments (Intuition Pumps)
  • Brain Replacement Surgery
  • There's Plenty of Room at the Bottom
• Arguments against downloading
  • Souls are Copy Protected
  • Neural Patterns like Fingerprints
  • The Unconquerable Vastness of the Brain
  • The Lack of a "Copy To" Destination
• Arguments in favor of downloading
  • Franklin
  • Kurzweil
  • Moravec

How should we handle a machine that acts like a person?

• Thought Experiments (Intuition Pumps)
  • The Star Trek Transporter and the Accidental Copy
  • "Don't pull the plug!"
  • Sociopaths: Persons that pretend to be persons
• Arguments against treating machines like people
  • They're not - Eliza and Parry Revisited
  • Even if they are, they're not - Speciesism
  • It's a slippery slope - Moralism
  • Come back and ask me when they unionize - Pragmatics
• Arguments in favor of treating machines like people
  • Sentient life is all that matters
  • Even if they're not, they're still our children
  • We don't want to make them angry. You wouldn't like them when they're angry.

Should we?

• Questions
  • Should we build intelligent machines?
  • Or machines with emotion, consciousness, or free will?
  • If machines have souls, is it right to make them to do our dirty work?
  • Should we build machines to replace human labor?
  • Is artificial intelligence worthwhile?
• Intuition Pumps
  • The Terminator
  • Commander Data
  • Bolo / Ogre
• Arguments against building AIs
  • They'll destroy us
  • It isn't fair
• Arguments in favor of building AI's
  • They'll be like our children
  • We better do it before someone else does first
  • We can make them just as smart as we need to and no more

Effects and Ethics of AI

• Weizenbaum - Human Judgement
• Kurzweil - Human Upgrades
• Schank - Human Learning

What if we?

What would be the effect of successful AI?

"This is a subject about which not much is known. Therefore everyone has an opinion."
- paraphrased from Richard Feynman

Questions

• Could machines take over the world?
• Could machines displace human workers?
  • Assembly line workers?
  • Janitors?
  • Psychologists?
  • Sex workers?
  • Priests?
  • Artists?
  • Politicians?
  • Musicians?
  • AI Researchers?
• If robotic factories replace human workers, who will we pay for their products?
• If people download their brains into machines, how will that affect society?
• Building Our Replacements - If AI is successful, will AIs supplant us?

Since we are, how could we do it better?

How Real Are the Controversies?

• Symbolic AI vs Connectionism - the fight over DARPA grants
• Reactive vs Hierarchical Robotics - the fight to be heard

Marketing and AI

• Expert Systems
• Neural Networks
• Genetic Algorithms
• Fuzzy Logic

Emotion and AI

• Whether machines can think isn't relevant to using AI for logistics planning
• Whether machines can feel emotion isn't relevant to game AI development
• Whether machines can understand isn't relevant to effective data mining, machine translation, or automatic text summarization

Artificial Intelligence as Computational Philosophy

Resources

Web Sites

• Issues
  • Wikipedia on AI: http://en.wikipedia.org/wiki/Artificial_intelligence
  • Philosophy of AI: http://www.aaai.org/AITopics/html/phil.html
  • Can Robots Become Conscious?: http://query.nytimes.com/gst/abstract.html?res=F20C11F73A5D0C728DDDA80994DB404482
  • A review of the Lighthill Report: http://www-formal.stanford.edu/jmc/reviews/lighthill/lighthill.html
  • Wikipedia on Functionalism: http://en.wikipedia.org/wiki/Functionalism_%28philosophy_of_mind%29
  • Wikipedia on the Chinese Room: http://en.wikipedia.org/wiki/Searle%27s_Chinese_room
• People
  • Daniel Dennett: http://ase.tufts.edu/cogstud/~ddennett.htm
  • John Searle: http://ist-socrates.berkeley.edu/~jsearle/
  • Marvin Minsky: http://web.media.mit.edu/~minsky/
  • John Searle: http://www.edge.org/3rd_culture/bios/penrose.html
  • John McCarthy: http://www-formal.stanford.edu/jmc/
  • Hubert Dreyfus: http://ist-socrates.berkeley.edu/~hdreyfus/

Related Readings

• Proponents
  • The Mind's I edited by Douglas Hofstadter and Daniel Dennett.
    A collection of essays built around the deep questions of intelligence. Includes essays on Searle's Chinese Room, Einstein's Brain, and the transporter problem.
  • Godel, Escher Bach: An Eternal Golden Braid by Douglas Hofstadter.
    A literate and entertaining case for Strong AI.
  • Mind Design edited by John Haugeland
    A collection of essays built around the deep questions of intelligence. Includes essays on Searle's Chinese Room, Einstein's Brain, and the transporter problem.
  • Society of Mind by Marvin Minsky.
    An attempt to explain the mind in terms of many small mechanisms.
  • Mind Children by Hans Moravec.
    Investigates seriously the consequences of building robots and downloading minds.
  • Brainstorms by Daniel Dennett.
    Essays on the mind, consicousness and free will
  • Consciousness Explained by Daniel Dennett.
    Essays on consciousness
  • Kinds of Minds by Daniel Dennett.
    An extended essay on evolution and consicousness
  • In the Age of Spiritual Machines by Ray Kurzweil
    An investigation of the vast increase of ocmputing power on our society.
  • Reason in Human Affairs by Herbert Simon
    Investigates rational decision theory in humans and machines
  • Sciences of the Artificial by Herbert Simon
    Establishes the theoretical justification for artificial intelligence
  • Scripts Plans Goals and Understanding by Schank and Abelson
    A dissection of how humans understand the world (with application to machines)
  • Dynamic Memory by Schank
    A dissection of how humans remember the world (with application to machines)
  • A Connoisseur's Guide to the Mind by Schank
    Popularization of the previous two books
• Opponents
  • Computers and Common Sense by Mortimer Taube
    Argues that Godel's result shows that thought cannot be formalized.
  • Minds, Brains and Programs by John Searle
    Argues that you can't leap from syntax to semantics
  • Rediscovery of the Mind by John Searle
    Book length version of argument that you can't leap from syntax to semantics
  • Artificial Intelligence: A General Survey (AKA "The Lighthill Report") by Sir James Lighthill
    Argues that intelligence, like turbulence, is too complicated to model
  • What Computers Can't Do by Dreyfus and Dreyfus
    Argues that intelligence is not rule-based and therefore not formalizable.
  • What Computers Still Can't Do by Dreyfus and Dreyfus
    Updated version of the Dreyfus' earlier book.
  • Computer Power and Human Reason by Weizenbaum
    Argues that tasks involving judgement should not be put in the hands of a machine
  • Understanding Computers and Cognition by Winograd and Flores
    Humans are biological entities in an environment, whereas computers are designed in a context which we are blind to
  • The Emperor's New Mind by Roger Penrose
    Penrose's first try to show mathematical intuition can't be formalized.
  • Shadows of the Mind by Roger Penrose
    Penrose's response to his critics, with greater rigor.