Dr. Mark Humphrys

School of Computing. Dublin City University.

Home      Blog      Teaching      Research      Contact

Online coding site: Ancient Brain

Free course: Online AI programming exercises

Search:

CA114      CA170

CA668      CA669      Projects


Survey of AI

The field of AI is incredibly big, and its edges are not well defined.
Some things could be considered part of AI or not, depending on definition.

There are two broad camps (though hard to define boundaries, much overlap, and many hybrid systems):

  1. Symbolic AI
  2. Sub-Symbolic AI



Symbolic AI

Overview:
  
Topics:



Sub-Symbolic AI

Overview:
  
Topics:
  


Artificial Life on Ancient Brain

A major theme in "Artificial Life" research is simulating systems with large numbers of simple actors.
e.g. Swarms of insects. Schools of fish. The immune system.

Such simulations can sometimes be applied to more intelligent actors. e.g. Human crowd behaviour, or warfare, or economics.

  

Artificial Life demo: Conway's Game of Life.
A simulation of simple actors that can change neighbouring squares, and yet one with amazingly complex dynamics.
Click to run World: The Game of Life at Ancient Brain.
Click "Randomize" and then "Play".




Debates




Elements of both Symbolic AI and Sub-Symbolic AI


Spam detection software, running on the system "alpamayo.computing.dcu.ie", has
identified this incoming email as possible spam. 

Content analysis details:   (25.2 points, 2.0 required)

 pts rule name              description
---- ---------------------- --------------------------------------------------
 1.0 FSL_XM_419             Old OE version in X-Mailer only seen in 419 spam
 2.6 NSL_RCVD_FROM_USER     Received from User
 0.0 FSL_RCVD_USER          FSL_RCVD_USER
 2.4 TVD_PH_BODY_ACCOUNTS_PRE BODY: TVD_PH_BODY_ACCOUNTS_PRE
 0.0 HTML_MESSAGE           BODY: HTML included in message
 0.8 BAYES_50               BODY: Bayes spam probability is 40 to 60%
                            [score: 0.4972]
 0.0 T_OBFU_HTML_ATTACH     BODY: HTML attachment with non-text MIME type
 0.7 MIME_HTML_ONLY         BODY: Message only has text/html MIME parts
 0.8 FSL_UA                 FSL_UA
 0.4 HTML_MIME_NO_HTML_TAG  HTML-only message, but there is no HTML tag
 0.1 FORGED_OUTLOOK_TAGS    Outlook can't send HTML in this format
 0.0 T_HTML_ATTACH          HTML attachment to bypass scanning?
 1.7 FROM_MISSP_MSFT        From misspaced + supposed Microsoft tool
 0.0 T_OBFU_ATTACH_MISSP    Obfuscated attachment type and misspaced From
 0.0 FSL_NEW_HELO_USER      FSL_NEW_HELO_USER
 1.9 AXB_XMAILER_MIMEOLE_OL_024C2 AXB_XMAILER_MIMEOLE_OL_024C2
 2.6 MSOE_MID_WRONG_CASE    MSOE_MID_WRONG_CASE
 0.0 FORGED_OUTLOOK_HTML    Outlook can't send HTML message only
 3.7 FROM_MISSP_TO_UNDISC   From misspaced, To undisclosed
 2.0 FSL_MISSP_REPLYTO      Mis-spaced from and Reply-to
 1.0 FROM_MISSP_USER        From misspaced, from "User"
 1.6 FROM_MISSPACED         From: missing whitespace
 0.0 FROM_MISSP_REPLYTO     From misspaced, has Reply-To
 1.9 FORGED_MUA_OUTLOOK     Forged mail pretending to be from MS Outlook

Bayesian spam filtering at DCU.


Cross-AI topics

There are some fundamental topics which are entire fields by themselves.
They may use:
  1. Symbolic AI
  2. Sub-symbolic AI
  3. No AI
or combinations of all three.

These fundamental "cross-AI" fields would be:

  1. Robotics
  2. Computer vision
  3. Natural Language Processing (NLP)

Each of these fields is an entire career by itself.

  

Links:
  1. Robotics

  2. Computer vision

  3. Natural Language Processing (NLP) (and here)




An example of computer vision plus OCR plus machine translation plus automated "photoshop".
The Word Lens app.
Google bought Word Lens and this functionality is incorporated into the Google Translate app.
See videos.



An example of audio speech recognition plus machine translation.
Microsoft real-time speech translator (integrated into Skype).




Machine translation example


Baron und Baronin Willy von Wattenwyl und ihre kinder Yvonne, Gérard und Sigismund geben hiemit tiefbetrübt Nachricht von dem Ableben ihrer innigstgeliebten Mutter, Schwiegermutter, Grossmutter und Urgrossmutter, der hochwohlgeborenen Frau Baronin von Stenksch-Prittag geb. Margaret Livingston-Gibbon, welche am 3. Januar 1911 nach kurzem Leiden in ihrem 93. Lebensjahre in Schloss Lannach bei Graz gottergeben entschlummerte.
Die teure Heimgegangene wird nach Graz zur Aufbahrung in der Leichenhalle des Evangelischen Friedhofes überführt, woselbst Donnerstag den 5. d. M. um 2 Uhr nachmittags die Einsegnung und hierauf die Bestattung im eigenen Grabe erfolgt.
Lannach, am 3. Januar 1911.

1911 death notice in my genealogy research.

Baron and Baroness Willy von Watteville and their children Yvonne, Gérard and Sigismund hereby give deeply saddened by the death of their message innigstgeliebten mother, mother, grandmother and great-grandmother, the woman Highborne Baroness von Stenksch-Prittag born Margaret Livingston-Gibbon, which asleep resignedly on January 3 in 1911 according to recently suffering in its 93rd year of life in Schloss Lannach near Graz.
The expensive home Gone is transferred to Graz for laying out in the morgue of the Protestant cemetery, woselbst Thursday 5th inst. By 2 clock in the afternoon, the blessing and then the burial in his own tomb done.
Lannach, on 3 January 1911th

Google Translate translation.
Uses Statistical machine translation.

Baron and Baroness Willy von Wattenwyl and their children Yvonne, Gérard and Sigismund hereby give deeply grieved notice of the death of their dearly beloved mother, mother-in-law, grandmother and great-grandmother, the most noble lady Baroness von Stenksch-Prittag, formerly Margaret Livingston Gibbon, who on 3 January 1911 after a short illness at the age of 93 years, in Lannach Castle near Graz, passed away.
The dear departed will be taken to Graz to lie in the mortuary of the Protestant cemetery, where on Thursday the 5th of the month at 2 o'clock in the afternoon the benediction and afterwards the burial in her own vault will take place.
Lannach, 3 January 1911.

The correct translation.

Discussion: Do you find this machine translation impressive? Or unimpressive?
Can you spot any specific language constructs the machine has trouble with?



Competitions

Competitions drive progress in AI since you can prove one system is better than another.



Competitions - Robotics

Competitions are important in robotics. How do you prove one walking robot is better than another walking robot?



RoboCup 2006 - Humanoid class - Final


RoboCup 2006 - Humanoid class - Penalties - Final



RoboCup 2009.



Wheeled players in RoboCup 2011.




Competitions - Natural Language Processing

How do you prove that one language-using program (like a chatbot) is better than another? Competitions are again a useful idea.

  

Reading

  1. Turing, Alan M. (1950), Computing Machinery and Intelligence, Mind 59:433-60.
    Introduces the "Turing Test".
  

Chatbots and the "symbol grounding" problem

  


IBM's Watson system competes on the trivia game Jeopardy!




Competitions - AI and board games

AI and board games is a classic arena for AI competitions. Board games are very "computerisable".

To play football in the real world (or indeed chess in the real world) we need robotic vision sensors trying to detect where the football / chess piece is, and mapping what it sees to some model of the pitch / board.

But if we run a board game on computer, with no physical real-world component, then we just tell the program where each chess piece is. There is no sensing problem.

In a software-only world, the program can be given perfect inputs, with no noise or errors. Though there may still be "hidden state". (Inputs not given to the program.)

  

Do humans and machines play chess differently?

AI finally beating humans at chess led to much discussion:
  

Reading

  1. How Intelligent is Deep Blue? by Drew McDermott, New York Times, May 14, 1997.



Competitions - Other

Competitions help drive progress.



Computer game bot Turing Test - Make a video game bot that seems to play like a human.
Above is the winning entry in BotPrize 2010.
See more videos.







Discussion: The "symbol grounding" problem

Recall the "symbol grounding" problem in our discussion of chatbots.

Is there any way to get a program that does "understand" the words it uses?
What would that even mean?
Would the program have to have a body and interact with the real world?
Or would it just need a better dictionary?

  


ancientbrain.com      w2mind.org      humphrysfamilytree.com

On the Internet since 1987.

Wikipedia: Sometimes I link to Wikipedia. I have written something In defence of Wikipedia. It is often a useful starting point but you cannot trust it. Linking to it is like linking to a Google search. A starting point, not a destination. I automatically highlight in red all links to Wikipedia and Google search and other possibly-unreliable user-generated content.