WhenFirstComputerProducedAprilFoolsJoke

On the April Fools’ Day we would like to cover the serious topic (and I believe we should not be the only one today) of artificial intelligence being able to generate humor. After all, the artificial brain will not be equal to the human one until it is able to produce and recognize jokes.

As, at the beginning of the study or artificial intelligence computer science has mainly focused on the learning algorithms, in the recent years with the neural network development, the researchers have showed much more interest towards computers being able to make jokes.

As confirmed by different scientists and journalists at different moments, the issue is in the definition of humor itself. The subject has been tackled by the psychologists, philosophers, biologists, linguists and sociologists for centuries now, and in spite of the numerous theories, there is little agreement. Should we still stick to the version that humor is a part of human’s creativity that cannot be reproduced by the machines so far?

Ilya Leybovich in iQ by Intel[i] and Joel Warner and Peter McGraw in Wired magazine[ii] have mentioned a number of joke-generating algorithms including STANDUP, the System To Augment Non-speakers’ Dialogue Using Puns and DEviaNT, the Double Entendre via Noun Transfer program. these programs are mainly used on natural language understanding and are fed with powerful ontologies or classified corpora in order to identify relationship between words that would provoke funny meaning. Their capacities are really amazing and the results may be compared to some human generated jokes.

At the same time, these jokes are still based on statistics and are not that spontaneous, context based, and… yes, creative. Joel Warner and Peter McGraw agree with this explaining the nature of humor by the “Benign Violation Theory”[iii], and confirming that when it comes to a “good comedy breaking the rules and reveling in the peculiar”[iv] it is not what a learning algorithm could train itself to do so far.

Another explanation of computers failing to produce humor is provided by Julia Taylor, a professor at Purdue, who insists on the insufficient ability of the machine to fully understand natural language[v].

While being totally in line with that, one would also think that computers do not have the whole historical and cultural baggage that humans have. As often, jokes are culturally determined and will not be funny in any language and every culture either.

One can hardly imagine the linguistic corpus that should be marked and provided to the algorithm so that it could learn about jokes, connotations and play upon words.

Another barrier explained by Julia Taylor and her professor Lawrence Mazlack is the surprise effect due to the gap between the expectation and the joke. “Puns and jokes work, Mazlack argues, because of expectations: We expect one thing and we’re surprised when we get another”[vi]. While the algorithm cannot really expect, and thus cannot experience the surprise effect, Mazlack studies are based on teaching computers understand jokes first providing them with the ontologies producing expectations and explaining the gap when there is humorous effect in the texts.

Analyzing all this, one may conclude that humor remains the area for improvement for artificial intelligence studies. It may not be of primary importance, like cyber security or health analytics, it is still of a huge importance for the researchers. As working on humor detection and generation promises great progress in natural language understanding, as well as in big data processing and developing neural network algorithms.

image source: Pixabay.com


 

[i] Is Humor the Final Barrier for Artificial Intelligence? by Ilya Leybovich for iQ by Intel, March 23, 2015, available online at http://iq.intel.com/is-humor-the-final-barrier-for-artificial-intelligence/?sr_source=lift_twitter&linkId=13153384, accessed on March 31, 2015
 

[ii] It’s Comedian vs. Computer in a Battle for Humor Supremacy by Joel Warner and Peter McGraw for WIRED, April 1, 2014, available online at http://www.wired.com/2014/04/underwire_0401_funnycomputer/, accessed on March 31, 2015

[iii] Benign Violation Theory by Dr. Peter McGraw, Humor Research Lab, University of Colorado Boulder, 2014, online http://leeds-faculty.colorado.edu/mcgrawp/Benign_Violation_Theory.html, accessed on April 1, 2015

[iv] It’s Comedian vs. Computer in a Battle for Humor Supremacy by Joel Warner and Peter McGraw for WIRED, April 1, 2014, available online at http://www.wired.com/2014/04/underwire_0401_funnycomputer/, accessed on March 31, 2015

[v] Is Humor the Final Barrier for Artificial Intelligence? by Ilya Leybovich for iQ by Intel, March 23, 2015, available online at http://iq.intel.com/is-humor-the-final-barrier-for-artificial-intelligence/?sr_source=lift_twitter&linkId=13153384, accessed on March 31, 2015

[vi] Artificial Intelligence? I'll Say. Why Computers Can't Joke by Drake Bennett for Bloomberg Business, March 23, 2015, available online at http://www.bloomberg.com/bw/articles/2012-03-23/artificial-intelligence-ill-say-dot-why-computers-cant-joke, accessed on March 31, 2015