Overview and Summary

unknownlippsΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 4 χρόνια και 23 μέρες)

79 εμφανίσεις

Overview and Summary

Michael Brand


Manchester town hall


The 12
“Manchester
Murals”

Color
-
lit

16
-
foot

pipe organ

Stars &
planets
depicted in
mozaic

City &
country crests

Victorian
-
era
neo
-
gothic
architecture


The Manchester baby


World’s first stored
-
program computer (1948)


Followed by Manchester Mark
-
1 (first w/ fast random
-
access two
-
level store) (1949)


Prototype for Ferranti Mark 1 (first commercially
-
available general
-
purpose computer) (1951)


Manchester coding


Phase encoding, developed for Manchester Mark 1


Used in Ethernet, RFID, etc.


Manchester carry chain


Fast adder with minimization of gate numbers


Virtual memory


Compiler compiler


For the Ferranti Atlas (1962)



Apple

University
of
Manchester

Gay village

Pablo Picasso


Turing’s official biographer


In addition to


On Computable Numbers, with an application to the
Entscheidungsproblem
,
Proc.
Lond
. Math. Soc. (2) 42 pp
230
-
265 (1936); correction ibid. 43, pp 544
-
546
(1937).


Introduction of “The halting problem” (Universal computing)


Computing Machinery and Intelligence,

Mind 49, pp
433
-
460 (1950)


Introduction of “The Imitation Game”/”Turing test” (AI)


The Chemical Basis of Morphogenesis
, Phil. Trans. R.
Soc. London B 237 pp 37
-
72 (1952)


Biological theory of individuation, symmetry
-
breaking and
pattern
-
forming




There is also


Intelligent Machinery

(Written 1948. Unpublished)


Reviews (Charles Darwin (NPL director)):


“A bit thin for a year’s time off”


“A schoolboy’s essay”


“not suitable for publication”


“smudgy”


It contained:


Logic based approach to problem
-
solving


Intellectual activity is primarily search


Genetic algorithms (“evolutionary search”)


Neural networks (“unorganized machines”)


An early form of the imitation game.


A blueprint for connectionism


Turing award winner:
nondeterminism


Turing and computability


On Computable Numbers, with an application to the
Entscheidungsproblem
,
Proc.
Lond
. Math. Soc. (2)
42 pp 230
-
265 (1936); correction ibid. 43, pp 544
-
546 (1937).


The Word problem in Semi
-
Groups with
Cancellation,

Ann. of Math. 52 (2), pp 491
-
505
(1950)


Critical strip


Solved Hilbert’s 10
th

problem


Turing and number theory


Turing and the Riemann Hypothesis



0

1

Critical line

Values on
critical line
can be
calculated as
real
-
valued
integral.
Approximate
and count
sign changes
for zeroes.

ζ

“Turing’s
method”=

calculate total
number of
zeroes in
critical strip
via
approximated
integral.

“Turing’s method” is still in use
today. His (many) other
innovations on RH have since
been superseded. Examples
follow.


Improved integral calculation for counting of
zeros on the critical line.


Improved finding places for suspected sign
-
changes (a.k.a. Gram points)


Improved bounds for
Skewes’s

number (first
case of
π
(x)>Li(x). See
Littlewood

(1914))


Systems of logic based on ordinals,

Proc.
Lond
. Math. Soc (2) 45 pp 161
-
228 (1939)
[was also Turing's Princeton Ph.D. thesis
(1938)] includes, under section “3. Number
Theoretic Theorems” a proof that



thus placing RH for the first time in the Arithmetical
Hierarchy.


Kreisel

(1958) later lowered this to


Automated calculation


“tide
-
predicting machine” (1939 application to the
Royal Society. Never built due to work on Enigma)


First to calculate zeroes mechanically (Mark
-
1)








Also: invented LU decomposition


“The calculations had been planned some time in
advance, but had in fact to be carried out in great haste.
If it had not been for the fact that the computer remained
in serviceable condition for an unusually long period from
3 p.m. one afternoon to 8 a.m. the following morning it is
probable that the calculations would never have been
done at all. As it was, the interval 2
π.63
2

< t < 2π.64
2

was investigated during that period, and very little more
was accomplished.”


Turing award winner: RSA, differential
cryptanalysis


Turing and Enigma


Major mistakes (G):


Usually, only inner rotor moves


Most strength is in plug
-
board, which can be bypassed


Plug
-
board connection is trivial


No fixed points


Message
-
keys were chosen badly


Operator errors (see
Tutte’s

reconstruction of
Tunny
)


Never willing to entertain suspicions of breakability


Major mistakes (B):


Never guessed plug
-
board connection



“The mythical man
-
month”; Turing award winner:
computer architecture


Turing and the Pilot ACE


Turing’s 1945 proposal (as compared with EDVAC)


is detailed to the register level (more than von Neumann’s
report)


is more general
-
purpose


5x faster


¼ electronic equipment


3
-
op packed instructions (plus a “next” address)


Fewer instruction fetches (
obsoleted

by larger memory)


Optimal next instruction placement in delay lines


Supports variable
-
length block transfers


Punched card I/O directly attached.


and yet, had little impact on computing history.


Why?


Assumption: HW dear; people cheap


11 Central registers, each with its own
behaviors (properties, side
-
effects, implied
operators, implied targets, multiple names


no accumulator)


No generic multiplication, no conditional
branching. Works in backwards
-
binary


No random access


No subroutine support


= A beast to program



Turing award winner: model checking


Formal verification


Turing:


Of course
entscheidungsproblem
, but also:


Checking a Large Routine
, Paper for the EDSAC
Inaugural Conference, 24 June 1949. Typescript
published in
Report of a Conference on High Speed
Automatic Calculating Machines,

pp 67
-
69.


Proof of termination by transfinite induction (presaging
Floyd (1967))


View as a graph problem


Formal languages for model definition (based
on temporal logic)


Symbolic model checking (storing partial
states)


Bounded model checking (Use SAT solvers to
consider the first
k

steps)


Node clumping


CEGAR: Counter
-
example guided automatic
abstraction



Turing award winner:
Quicksort
, CSP


Can computers understand their own
programs?


Turing: Self
-
simulation + verification + AI


Suggested alternate wording: can a computer
program provide its programmer with
pertinent information about itself?


Where the positive answer is already in use:


Programs can check for buffer overflows


Can generate test
-
cases for recent changes


Can pinpoint cases where changes can make
programs slower


Former world chess champion;


-
3½ against “Deep Blue”


Turing’s paper machine


Turing and chess


At Bletchley park: Hugh Alexander, James
Macrae

Aitken


Turing’s Running Chess


Early imitation game


15 seconds of silence?


(1948) Turing designed the first chess algorithm. He
hand
-
simulated it (and lost) in a match against
Alick

Glennie


Kasparov’s team implemented the algorithm. Found that
Turing inadvertently alpha
-
beta pruned.


Changing the result in 10 of the game’s 24 moves.


Today: “Advanced Chess”
(
GM+Comp

vs.
GM+Comp
)


Kasparov: Cooperation is key.


”The algorithmic beauty of sea
-
shells”


Turing and Morphogenesis


Turing: inhibitor w. longer range (diffuses)

Activator

Inhibitor


Today mainstream, but initial
scepticism


Stochastic results


but live organisms not so


initial state
nonsymmetric


cannot produce axial patterns


negative concentrations in equations (fixed by
nonlinear reactions)


Explains a wide variety of phenomena

Hydra


Periodicity



Gradients



Oscillations



Phyllotaxis

centralization


“...but with three or more
morphogens

it is possible
to have travelling waves. With a ring there would be
two sets of waves, one travelling clockwise and the
other anticlockwise. There is a natural chemical
wave
-
length and wave frequency in this case as well
as a wave
-
length; no attempt was made to develop
formulae for these...”.


“Father of the Internet”, ICANN chair, Google VP



IP
-
enabled surfboards, light
-
bulbs and toasters


Sensor
-
nets even in self
-
driving cars & wines


Bit
-
rot hazard → legal issues of IP


Data availability → privacy? new social norms?


techno+academic+legal+civil

society+industry



Interplanetary Internet


“Watson”



“Jeopardy!”: Broad domain, speed, precision,
accuracy estimate


Ambiguity, anaphora resolution


Use of existing resources, no spec data


Sentence parsing + statistical aggregation +
context


Score competing hypotheses based on
evidence + recursively


Was known as
Cottonopolis


Presidential Rhyme Time: “Barack’s Andean pack
animals“


Obama’s Llamas


Fables & Folklore: “
Gerda

tries to rescue Kay from this
Hans Christian Andersen title royal”


The Snow Queen


The first named character in “The Man in the Iron
Mask” also to appear in the author’s previous work.


D’Artagnan


Submitted to Lincoln in June 1964 by the secretary of
treasury and accepted


Offer of resignation.


Or was it a friend request?


AI+Robotics
; past president of
RoboCup



Learning by
perception+cognition+action


feedback. Learning by reusing solutions (“by
analogy”)


Turing (
Intelligent Machinery
): computers apt in (
i
)
games, (ii) language, (iii) translation, (iv)
cryptography, (v) mathematics, of which the most
difficult is (ii) and requires sensory input and
locomotion. (“Roaming the countryside”)



centrally/
noncentrally

controlled


Model world, plan to the goal
(
probablistic
, physics
-
based,
variable
-
detail), update on new
info


Add artificial goals to
heuristically approximate pruned
states


Purposeful perception: little of
the image gets processed.



Companion mobile robots


Active since 2009 & 2010
rsp
.


Navigate Gates Hillman Center
(*)

using the
Kinect

depth
-
camera,
WiFi
, and/or LIDAR


Proactively ask for help


Ask humans to press elevator buttons


Follow humans (and each other) along
glass corridor

Glass corridor

(*)

217,000
-
square
-
foot,
9 floors


Turing award winner: PAC, #P, holographic
reductions, CF parsing,
UniqSat∈P⇒RP
=NP


Quantifying evolution


Basic question: humans have 3⋅10
9

base
pairs. How does evolution get there without
4^that time?


What are the possibilities for protein
expression? Algorithms for next generation?
How does evolution navigate the search
space?



Samson
Abramsky

(
game semantics, domain
theory
): What is a process? (which two are equiv?)


Carole Goble (
eScience
, grid computing
):
Universal social machines?


Manuella

Veloso
: Universal robots?


Ron
Brachman

(
description logic; AI; VP Yahoo!
Labs
): If intelligence is like athleticism in that
there is no single sport metric, what is our aim?



Moshe
Vardi

(
model checking, database theory,
constraint satisfaction
): Does the future need us?



Elephants don’t play chess”,
iRobot


Turing never meant the imitation game
as a test. It was meant to show the
theoretical possibility (nullifying the
emotional weight we put on
“intelligence”). He also suggested a “men
vs. women” variation (which people are
not good at) and wondered whether
computers can be told from humans by
their chess
-
play (which they normally
can).


“Intelligence” is the appearance of
intelligence, which is in the ability to
interact. People love to
anthropomorphize (incl. other people).

Kismet


Steve
Furber

(
BBC micro; ARM 32
-
bit RISC
microprocessor
): Higher intelligence is an
unnatural top layer over human intelligence. We
over
-
assume about our own intelligence. The
most rewarded ability is to lead a game of
football. Our assumed intelligence created a
barrier that makes it difficult for us to build AI.


Manuella

Veloso
: Intelligence is about physical
interaction, not abstract cognition. Learning is
also memory, not just adapting classification
parameters.


Turing’s criteria for “winning” the imitation
game is a 30% success rate at fooling the
judge after a 5 minute conversation.


On the day of the centenary (June 23
rd
) the
biggest Turing test ever was staged at
Bletchley Park.


13
-
year
-
old Eugene
Goostman

managed to
fool judges 29% of the time.


Donald Knuth


Roger Penrose


Andrew Yao


George Ellis


Martin Davis


Samuel Klein (
Wikimedia Foundation
)


...


questions?