4. Computer Room (2008)
8,000 servers housing 40,000 processor
Experiments produce about 15 petabytes of data per year (= 1.500.000 giga) – 5%
of internet in 2001
6. Wim Klein
first computer in action at CERN
(when machine was slower than human)
Wim Klein worked at CERN from 1958 to 1976
7. First computer in CERN (Wim Klein)
When machine was slower than human
CERN -Farewell Party -
http://www.youtube.com/watch?v=urFiv_PQ2FQ
8. Wait a minute..
48 x 65 x 37 x 84 = ?
(yes.. I checked all the results of the
video)
9. Wait a minute..
48 x 65 x 37 x 84 = 9696960
In Science repeatability >> speed:
• Human experts make sometime mistakes
• Machine repeat a mistake forever => much
better
In 1973 the CERN started to use machines.
Klein retired in 1976 (HP vs Klein)
!
Expertise in computation: Machine, winner since
the 70s
(don’t trust
experts)
10. Intelligence: Human expertise/capability vs. machine
• Yes but computation is a “stupid” activity, humans
are more “intelligent” than machine.
• Fair enough, what about Chess?
Chess = Traditional
challenge in Artificial
Intelligence
!
Expertise in Chess:
Machine, winner since
2003
!
But is it a real victory for
the A.I field?Kasparov vs. machine (Deep
blue)
11. Moravec Paradox & The nature of intelligence
“The main lesson of thirty-five years of AI
research is that. the hard problems are
easy and the easy problems are hard”
Linguist and cognitive scientist Steven Pinker
Kasparov vs. machine (2003)
Fake victory of AI (brut force computation)
“It is easy to make computers exhibit adult
level performance on intelligence tests or
playing checkers, and difficult or
impossible to give them the skills of a one-
year-old when it comes to perception and
mobility” - 1988 Moravec
12. 1969 by neuroscientist
Paul Bach-y-Rita et al.
SIDE NOTE:
Cognition and intelligence development
sensorimotor skills and instincts that
we share with the animals require
enormous computational resources
compared to symbolic thinking
13. Turing Test
Modern example, the captcha
Recognizing words is too
hard for machines, but easy
for people.
= Testing the machine's ability to exhibit intelligence human
expertise.
Traditional Turing Test: a conversion.
!
If, during a conversation,
machine can fool the
interrogator into thinking that
it is human, then it pass the
test
14.
15. !
!
The raise of crowdsourcing
and the decentralization of expertise
16. Crowdsourcing not a new practice:
distributed computing in the old days
a computer at work 1940.
(Human) Computer < 1950
The First World War required large
numbers of human computers.
The Mathematical Tables Project
• 450 'human computers' mostly people
with low incomes
• Worker split up according to arithmetic
function (+, -,* , if skilled /)
17. Modern examples
[1]- Jim Giles, « Special Report Internet encyclopaedias go head to head » Nature 438, 900-901
(Dec 2005)
Encyclopaedia expertise:
Amateurs winner since 2005 vs. professionals
(Wikipedia vs Encyclopaedia Britannica - [1])
Image interpretation expertise:
amateurs still winner vs. machine
(e.g. Galaxy zoo’s project: 60M of images of
galaxies classified in few months by 80.000
amateurs with high accuracy, too hard for the
machines
(and the amateurs teached the machine
[2])
[2] The output data was used to improve the algorithm
18. Expert vs amateurs, another kind of Turing
Test..
Croatia's Ruđer Bošković Institute website
• Against spammer and beginners…
• Extra-hard version of a regular CAPTCHA-type
19. micro-task economy
• Amazon Mechanical Turk (2005 – present)
>80.000 HITs (Human Intelligence Tasks)
(Science: low cost living lab for social/behavior
research)
20. and what about school?
Change in teaching expertise/skill)