Week 5

This week and the next were spent primarily reading up on how the system could be made more intelligent. I would personally classify this 'intelligence' to be of two types.

  • Perceived intelligence
  • Built-in intelligence

Perceived intelligence

In this case, the user comes to think that the system behaves intelligently although the underlying algorithms may not have been modified much, if at all. In this regard, some of the ideas I had mentioned in the entry last week might fall into this. For instance, the conversation mode, while not requiring a major improvement in the current code base, could give the user the impression that the system is able to handle and follow conversations. Similarly, in the new proposed poetry mode, having a structured poetry mode makes the system look less 'stupid', especially when the text sometimes seems to go outside the bounds of the screen display.

Such display improvements would readily make the system appear more intelligent in the poll mode as well. If the system is able to calculate, aggregate, and display the votes received (again, something that does not require a major upgrade), it would mark an improvement in terms of intelligence.

Another suggestion I would like to add to the previous list is a question mode. In this mode, the blogwall would display a question on screen, and await answers. One way to make this intelligent is to have questions related to different areas, so if a user answers one question correctly, the system would pick a harder question from the same field to test his or her knowledge further. Otherwise, the system could go to another field.

Built-in intelligence

Improvements in this category would require a substantial undertaking in th research and the implementation of new smart algorithms that would 'truly' imbue the system with at least some human-like intelligence. For the Blogwall, I believe this intelligence can come most readily in the following forms:

  • Natural Language Parser. The ability of the system to understand (at least to a certain degree) the context of the text message that has been sent and then use that information when selecting a poem is one. However, another option that could be done is conversation threading in conversation mode, again by analysis of the text in the message. The system could then support multiple conversations at the same time, and then 'know' which conversation the received text message was intended for.
  • Intelligence through affect. I think this is one very promising field that could apply well to the blogwall. I had mentioned this briefly in the last entry, by using sound and colours to convey affective content to the user. The system could attempt to interpret the emotional weight of the chosen poem using these means. However, another interesting application would be to let the user influence the system, but sending words that would make the blogwall 'sad' or 'happy'.

What this means

A combination of both would probably be the best way forward. Built-in intelligence in terms of language parsers have their challenges. It is extremely difficult to have a well-functioning system that will understand the context of all text messages. The very performance of natural language parsers depends on extensive knowledge about the outside world. It would be well nigh impossible to consolidate all this knowledge. The concept would however would work well in a limited context, where this amount of external knowledge is minimized.

I will talk more about how this intelligence could be implemented in the next update.


Any comments?

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