Week 6

With that distinction on the type of intelligence out of the way, this entry will examine in greater detail how built-in intelligence can be implemented in the blogwall.

Natural Language Parser

The key is to find how much realism is enough and achievable. If realism was plotted on the vertical axis and complexity/amount of external information on the horizontal axis, I would model the resulting graph as something resembling a normal distribution curve. As complexity increases, realism increases up to a certain point. Beyond this, the system has too much external information to consider and is unable to find meaningful contextual explanations to input stimuli. The challenge then is to find the right amount of information where realism can be maximized for the end user.

This can possibly be done by limiting the settings/situations where a natural language parser is employed. The Blogwall does not have to understand every single input stimuli. For instance, users could be instructed to provide simple sentences with a clear well-defined subject, verb and object. The vocabulary could be limited if the system can perform exceedingly well in this space.

I'm also thinking of simple messages like “say [something] in 5 minutes” or “find little star in poems” The first message will prompt the system to display the specified text in 5 mins. The second might tell the system to look for the specified words in a particular location. The command could be extended by changing the location. Different keywords could be used in addition to “say” for the commands for more options. I have seen one service which has implemented such simple commands, and I believe it gives the user a great impression of the intelligence of the system. The crucial point is to make the commands as natural as possible by the correct choice of keywords.

Affect as intelligence

Since poems contain some emotional content, being able to communicate this affective content to the users would be useful to create an intelligent system, particularly if combined with the natural language parser above. However, attempting to “estimate” the user's emotional state is a tricky proposition, and I would not recommend that. What is proposed rather is to make the system sad or happy depending on the poem displayed. This may or may not correspond to the user's emotional state.

This ties in to my next point, the need for spaces for user appropriation.

Spaces for user appropriation

I believe this to be very important. In any system, especially an affective one, the interaction is much more fun when there are spaces for user appropriation. There is a danger that the system tries to algorithmically calculate and know everything about the user. This will only drive users away and make interaction with the system less pleasant.

This is primarily due to the fact that the situational and emotional context of an input stimuli may have way too much external information attached to it, that the system will be unable to make a proper assumption about the user due to the lack of this information.

As a result, I propose the use of appropriable surfaces such as colours and patterns to communicate such affective content to the user. While certain colours may have associated meanings (eg. green as calm or peaceful), they allow the user to make up their own meanings for the most part. As a result, it might be better that the system does not explicitly communicate the emotional content of the poem as text.

The fun factor: going the other way round

For a system like the Blogwall, an element of fun is essential. But this can be used to convey intelligence as well.

One idea I had was to give the Blogwall system a persona. This could be explicit in the form of an avatar or something more implicit. However, the idea remains the same. Instead of the system trying to guess the situational context of the input message, the users are asked to influence the state of the machine. For instance, the current state of the Blogwall could be sad. The users will then have to send in the appropriate words or sentences so that a happy poem gets selected to make the system happy again. If well implemented, this could be extended in the form of a game or separate mode.

Moving forward

The challenge now is to find the right balance in the approach to be taken. What will be implemented and to what extent? The first step will likely involve a relatively small scale implementation for trial and testing purposes. I hope to have all the information compiled and ready for presentation on October 1, when we can discuss about the way forward and what should be done first.

Sources

Boehner K., DePaula R., Dourish P. & Sengers P. (2005). Affect: From information to interaction.
Natural Language Processing (n.d.). Retrieved September 20, 2007, from [http://en.wikipedia.org/wiki/Natural_language_processing]
Artificial Intelligence (n.d.). Retrieved September 20, 2007 from [http://en.wikipedia.org/wiki/Artificial_intelligence]
- and the many links from these two pages
Ståhl A., Sundström P. & Höök K. (2005). A Foundation for Emotional Expressivity, In Proceedings of DUX 2005, San Francisco, CA, USA.
- More information about the eMoto system at [http://emoto.sics.se/]

Comments

Any comments?

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License