Intelligent Systems

1
Intelligent Systems

• Assignment 1
• Group members Probir K Ghosh
• Prasad Valapet
• Soumen
  Sengupta

2
Implications of Deep Blue for AI

• Fact file of Deep Blue, the IBM super-computer
• IBM RS6000/SP
• 256 processors in parallel

• In 1996, Kasparov remained unbeaten and defeated
  the super-computer thrice
• Deep Blue defeated Kasparov on 11th May, 1997 for
  the first time.
• The tournament ended with 2-1 with 3.5 2.5 in
  favor of Deep Blue

3
Man vs Machine

• Deep Blue a supercomputer that outperforms human
  beings with only its raw computational power
• 200 million moves evaluated every second as
  opposed to only 3 per second by human beings
• Considerable improvement to the evaluation
  function used

4
Features of Deep Blue for future apps

• Advanced search algorithms for large amounts of
  search and evaluation
• Large Scale Parallelization
• Application Specific Hardware

5
New Domains

• Financial Institutions require a lot of
  computations which makes it useful to areas in
  Computational Finance
• Monte Carlo simulation in financial
  applications
• Deep Blue can also facilitate data mining
  applied to medical databases

6
Drawbacks of Deep Blue

• Deep Blue lacked the autonomy and adaptability
• It is not capable of learning from its mistakes
• (experience)
• Lacks intuition
• Not capable of explaining its own model or even
  say why it won.

7
What does the future hold for AI?

• Inclination towards modeling the human brain
• Knowledge needs to be combined with parallel
  processing
• Should get better at pattern recognition and
  hence be more selective in its approach
• Machines like Deep Blue could be extended to
  other applications of AI traffic control,
  production quality control, internet
  applications, data mining, massive computations

8
Intelligence of Current Commercial Translating
Machines

• Machine translation is the process of automating
  the tasks related to translation from on language
  to another
• Types of knowledge involved
• lexical
• syntactic
• semantic
• pragmatic
• The box is in the pen
• The pen is in the box

9
Why is Machine Translation Hard?

• Ambiguity
• Knowledge of the linguistic architecture of the
  languages concerned
• Structure of a sentence varies from one to
  another
• Lexical Differences
• Idioms and Metaphors ( multi word units)
• It is our understanding of the real
  world knowledge that
• helps us to understand and express
  ourselves better

10
Is the machine translator intelligent?

• A language translating machine can be deemed
  intelligent only if it achieves the following
  criteria to some degree
• Quality and speed of Translation Engine
• Intelligibility
• Accuracy
• Handling idiomatic expressions

11
Terminology Management

• TM Identifying, Extracting, Storing,
  Re-ordering, using and updating significant terms
  with their translations
• Definitions
• Example of use
• Multiple translations
• Cross refs to other terms

12
Flow Diagram of Terminology Management
Source
Extraction
Feedback
Verification
Repository
13
A few Language Translators

• Systran
• The most popular language translation system
  with support for a wide array of languages
• LOGOS
• One of the first to introduce machine
  translation system in 1970s. Improved widely in
  performance and linguistics and
• METEO (setup in 1977 at the Canadian
  Meteorological Center in Dorval, Montreal)
• widely used language translation software
  used for weather forecasting (domain specific
  vocabulary)

14
A few Language Translators contd.

• IBM Personal Translator
• A popular language translation tool in most
  big enterprises. Outperforms some translating
  systems with their accuracy
• Globalink LH Translator Pro uses very
  sophisticated linguistic processing mechanisms to
  provide an accurate and intelligible translation
  

15
Intelligence of Current Translators

• Current Translators like Systran are smart
• avoid the pitfall of word by word translation
• Achieves a very high accuracy rate (6080)
• Multilingual (can translate between 90 languages)
  but tend to perform better between certain
  languages

16
Intelligence of Current Translators

• The translators have achieved tremendous speeds
  for example Systran can translate 600 pages per
  hour with 32 bit technology
• Coverage of terms pertaining to subject from
  various disciplines ( 20 domains for Systran)
• Makes provisions for storing many terms and
  expressions and can translate using an analogy
  with similar sentences to come up with the
  meaning ( data mining)

17
Intelligence of Current Translators

• Capable of providing alternate meanings
• Some of them like Babelfish of Systran can
  provide accurate translations of some figurative
  expressions
• Current Translation mechanisms are capable of
  feedback mechanisms by which they can learn from
  the authors as well.

18
References

• Erik Larson Rethinking Deep Blue Why a Computer
  Can't Reproduce a Mind
• urlhttp//www.arn.org/docs/odesign/od182/
  blue182.htm
• Robert Morris Kasparov vs Deep Blue The
  Significance for Artificial Intelligence, 14th
  National Conference on Artificial Intelligence,
  1997 urlhttp//citeseer.nj.nec.com/cache/paper
  s/cs/12269/httpzSzzSzwww.cs.fit.eduzSzmorriszSzP
  SzSzwksum.pdf/workshop-summary-kasparov-vs.pdf
• The Implications of Kasparov vs Deep Blue
• Various viewpoints.
• url http//www.cs.strath.ac.uk/fabioc/02-i/doc
  s/DeepBlue.pdf
• Drew Mcdermott How Intelligent is Deep Blue?
• Article published in New York Times on 14th May
  1997 urlhttp//www.nyu.edu/gsas/dept/philo/cour
  ses/mindsandmachines/mcdermott.html

19
References contd.

• How Well Does Computer Translation Work?Find Out
  for Yourself url http//language.home.sprynet.com
  /lingdex/comptran.htm
• Veronica Dahl Understanding and Translating
  Language Challenges of the 90s
• url http//www.cs.sfu.ca/fasinfo/cs/CC/413/v
  eronica/material/
• challenges90s.html
• D.J. Arnold, Lorna Balkan, Siety Meijer, R.Lee
  Humphreys and Louisa
• Sadler Machine Translation an Introductory
  Guide, Blackwells-NCC, London, 1994.
• url http//www.essex.ac.uk/linguistics/clmt
  /MTbook/