Transcript Robotics

Illusions of intelligence
Alan Turing
Lab: Combine sound and light exercises.
Homework: Identify & describe problems with the
automatic solution similar to ‘human’ and problems
with the automatic solution very different
Artificial Intelligence
• Definitions?
– Machine (computer) simulating human reasoning
– Machine (computer) demonstrating surprisingly
human intelligence
• Problem for field: as soon as some AI research
proves practical, it isn’t considered AI!
– American Association for Artificial Intelligence (AAAI)
– AAAI/Nova interview with Cynthia Breazeal
Different strategies
• Work to solve problem, using any
technique that works
• Work to understand how humans reason
enough to implement these ways using
computers (cognitive science) and use
implementations to test ideas
• Symbolic manipulation (as opposed to
numerical calculations)
• Enumerating / expanding trees of
possibility: branch-and-bound search
• Expert systems: states and testing
• Neural nets (and other forms of machine
Expert systems
• Collection of rules: If (A, B, C) then do D
• Using the expert system consists of going
through the rules and doing the actions
• Early example: medical diagnosis.
– Do checks
– Actions may be do another test
– Arrive at diagnosis
Neural nets
• Modeled after how brain may work
• Define graph (nodes, directed edges)
– Nodes conditions
– Edge from A to B if A leads to B
• Different techniques for building and refining net,
including trying many cases and putting weight
on edge if it leads to good result
Pattern recognition
• Need to extract measurable features
• These constitute the signature
• Compare to archive
• Example: facial recognition. Features such
as ratio of spaces between eyes to eyes to
chin. Need to use ratios for such things.
• Theorem proving: technique to assume negative
and see if you reach a contradiction by trying all
• Natural language processing for interface
– My abominable abdomen project vs moon rocks
• Natural language processing for translation:
currently has some success/utility
• Speech recognition
– Depends on size of language, restrictions and/or
• Attempt to make use of outtakes from interviews
used for documentary of Jacques Lipchitz,
sculptor: 1970!
• Use keywords linking to segments to have
Jacques answer questions
– Illusion only of natural language
– System continually massaged
– Updated to work with latest technology
• If [art, Histor] moves you, you must be satisfied.
Quote from Jacques Lipchitz about art in
general. Can apply to using Histor.
• AI?
• 1960s program (parady) by Joseph
Weizenbaum to emulate a therapist (Rogerian)
• Relatively simple manipulation of patient’s
remarks with randomly inserted stock questions.
– “My head hurts.” “Why do you think your head hurts”.
– “I feel bad today.” “What do you think about your
• Fairly successful!
Alan Turing
• Significant theoretical work on computation
(Decision Problem)
• Worked at Bletchley Park during WWII on
decoding German codes (Enigma machine):
done by altering a coding machine
• Worked various places, including Princeton, with
Van Neumann, others, on early computers
• Proposed way to build a chess machine
• Defined Turing test
Decision problem
• Theoretical problem, set by Hilbert (1900)
Entscheidungsproblem: What does it
mean for something to be computable?
• Turing (1936) produced 2 formulations
(Turing machines & recursive functions)
and proved them equivalent (and later
proved these equivalent to a formulation of
Post). Also proved limitations many others
Turing machine
• Infinite strip of tape
• Machine has finite number of states. A state
holds the definition of what to do when reading a
0 or 1 on the tape
• Machine reading a spot on the tape can
– Move (on tape) left or right or stop
– Write something on tape (re-write)
– Change state
• Turing machine computes a function on an input
if it stops. The result is the number on the tape
(technically, answer is number of 1s on the tape
or one more)
Universal Turing machine
• Encode a Turing machine to be a single
• A Universal Turing machine takes as input
the number representing a TM plus input
and produces the result that the TM would
produce from that input
Recursive functions
• Functions from vectors/tuples to
• Built up from basic functions
– Constant functions F(x1, x2, ..xk) = n
– Successor function F(x) = x+1
– Projection Fik(x1, x2, ..xk) = xi
• Using
– Composition
– Primitive recursion
– Inverse (aka μ operator)
Turing test
• Set up a judge to have a ‘conversation’
using text messages back and forth to a
machine and to a human.
• If the judge cannot tell the difference, then
the machine has passed the [Turing] test.
• What about Histor? What about Eliza?
Should the test be harder?
• Robotics has been considered part of AI in
computer science but also in engineering
• Are AI techniques such as pattern
recognition, expert systems, neural
networks, especially applicable to physical
Preview / Commercial
• Fall course Advanced Topics in Computer
Science will include computability, AI and
encryption (including recent news about
vulnerability of current practices), etc.
• Beautiful
• Exercise in logical thinking
• Follow line and turn around when there is a
• Your own idea. Ideas:
– From outside the blue oval, start when someone
claps, move to the circle and follow line.
– From outside (especially track in the back), go in one
direction and keep track of each blue line passed.
Use variable (suitcase). Display count.
– [Needs singing/hum] Start at rest. When sound is
greater than a certain level, go around oval. If and
when sound stops (falls),
• turn to inside of oval and stop
• leave oval
– Add bump sensor (perhaps in back or
combined with light?)
• Your own challenge to the class?
• Identify a problem/task for which the automated
solution/approach is
– Similar to the human way
– Different from the human way
• Postings
• Preview: AI topics are options for research
presentation as are 'real world' robotics, real
robots versus robots in literature, health,
miltiary,ethics, etc.