Machine intelligence and ethics
Copyright By PowCoder代写 加微信 powcoder
Computing machinery and intelligence in Mind, 1950
I PROPOSE to consider the question, ‘Can machines think?’
How to define ‘think’? Too difficult, even meaningless question
Instead: Imitation game
Turing Test
Thinking like a human = Giving outputs like a human
Turing – objections
So, machines can be intelligent and can ‘think’ like humans
The ‘Heads in the Sand’ Objection T: Human hubris and superiority
The theological objection: Immortal souls required for thinking
T: God could give souls to machines
Lady Lovelace’s Objection: No creativity, surprise
T: Factually wrong (DALL-E)
The Argument from Consciousness: No feeling, self-aware, wanting
T: Begs question re. imitation game or solipsism (I only know X is conscious if I can experience X’s experiences)
wikipedia 4
: Machines cannot think
• AI that just computes has no mind
• Brain/mind not a computer
• i.e. think, understand, conscious as humans are
• Turing: ‘thinking’: kind of computation independent of matter it occurs in
• Any material: microchips, vacuum tubes, beer cans
• Strong AI: real mind
• Weak AI: simulates thinking; in specific area
e.g. chess, medical diagnosis
Chinese Room
http://jamesian58.blogspot.com/2017/04/searle-chinese-room.html 6
Chinese Room cont’
• Rule book = computer program
• Writers of rule book = programmers
• Person = computer
• Inputs (question)àcomputeràoutputs (answers)
• Pass Turing Test
• Has person understood?
• No! Only manipulated symbols
• Symbols have no meaning to person – they are not thinking at all
• Computation: Syntax vs. semantics
• E.g. person/computer has no idea what
the symbols are about
• Even if minds perform computationsà
insufficient for thinking, understanding
• Would not help to make system more complex (e.g. parallel processing, neural networks)…
• …nor increase its Turing Test performance to seem more ‘human’ etc. (e.g. quick-witted , funny, knowledgeable etc.)
• AI simulates not duplicates
Biased against non-biological systems (cf. ‘wet slimy stuff’)
• S: must be something about neurobiological systems (brains) vital for causing mind states
• But: S open to non-biological minds that are not merely computational
The whole system is the mind
(person+book+basket of symbols) • S: memorises rules
• Robot Reply
Computers isolated from world – robots causally connected: AI that responds in physical space to objects poss. Different
• Turing Test revisited
Seems counterintuitive that person in CR understands, but counterintuitive to deny a fully conversant future AI understands?
Superintelligence/AGI
Superintelligence
Ray Kurzweil Nick Bostrom
Superintelligence: : “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains”
Once equal to human intelligence: Technological singularity: runaway intelligence (minutes/hours/weeks)
Speed; multiple intelligences; qualitatively smarter
“we may be the stupidest possible biological species capable of starting a technological civilisation”
Fear of AI
Stephen Hawking: “The genie is out of the bottle. I fear that AI may replace humans altogether”
Elon Musk and others: Years to decades
: “may be fundamental obstacles to get much beyond human intelligence.”
Ethical issues 1
Existential threat: : outwit all of us (e.g. off switch)
What if AI reasons that egoism or relativism is true?
Subgoals: Riemann hypothesis & computronium, paperclips
Most goals when coded: unforeseen consequences
Ethical issues 2
Orthogonality thesis: More-or-less any level of intelligence could in principle be combined with more-or-less any final goal (Bostrom)
Control-problem: how to get superintelligence to choose humanistic values (Bostrom)
Asimov’s 3 Laws of Robotics: non-injury, obey humans, self-protection
Disagreement about ethics and theories (U, VE, CE, D) – how to choose?
Avoid building AI with anything close to human intelligence?
Questioning superintelligence
False prophets?
Bostrom: most pressing threat of all – devote much attention
More important than climate change, nuclear threat, etc.?
Searle: AI has no “real, observer- independent intelligence”
AI outputs: “just electrical sequences unless some conscious agent” interprets them
What AI does: “no psychological reality”
All images from Unsplash
Bietti, E. (2020, January). From ethics washing to ethics bashing: a view on tech ethics from within moral philosophy. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (pp. 210-219).
‘Ethics washing’
• Misusing word ‘ethics’ and ethics language
• Big tech companies
• Appearance not reality of ethical
• Self-serving uses
• Justify deregulation, self- regulation, or market driven governance
Digital ethics team
‘Ethics washing’ cont
• Hiring moral philosophers to look good but not actually do much
• Company ethicists may feel or be inhibited in what they do and say
• Tinker round edges vs. asking hard and disruptive questions e.g.
• Eg. ‘Can does not imply ought’ – some AI perhaps should not be developed, rather than tinkering with algorithmic fairness
Ethics bashing
Philosophy and ethics are seen as:
1. a mere communications strategy and form of cover-up for unethical behavior
2. mere “ivory tower” intellectualization of complex problems that need to be dealt with in practice (too abstract, misses details, not relevant enough)
3. opposing political representation and social organizing
Intrinsic vs instrumental value
• Intrinsic value: ethical understanding for its own sake
• Instrumental value: ethics as tool e.g. better regulation, company
reputation etc.
• Not mutually exclusive: Both important
• Ethics/moral philosophy: requires a rational, impartial, and questioning attitude, not predominantly an instrumental one
Exam information
In lecture!
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com