Paper Reading #7: Real-time Interaction with Supervised Learning

Comment 1: http://chi-jacob.blogspot.com/2011/02/paper-reading-7-real-time-interaction.html
Comment 2: http://csce436-nabors.blogspot.com/2011/02/reading-7-real-time-interaction-with.html

Real-time Interaction with Supervised Learning
Rebecca Fiebrink
2010 Doctoral Consortium

In this paper the author talks about a new program she is working with that is going to assist in the machine learning process. As technology gets better more and more programs are going to be moved to having stronger AI programs associated with them and this is going to require more sophisticated and easy to learn machine learning techniques. This also means that all machine learning will not have to be programmed in by a computer scientist but will simply be able to be inserted by any average user and then processed and ready to be used by the program. Much of this work was inspired by the Weka system that allowed users of all backgrounds to follow some very simple machine learning premises and enter a lot of information and then translate this directly into a usable form without the need for programming. The author claims that these kinds of systems will have a lot of practical implications from sound identification to other forms of social media and learning. The author has begun development on a system that will allow the user to not only input data but also record live time data into simple training modules and then on a step by step basis allow the user to pinpoint specifics that the system should look for as well as change the way the program is seeing something by adding more models or more training examples of different kinds. She was then able to take these systems and work with musical performers and directors and see how this kind of machine learning tool has helped their performances. She was able to collect data from them as well as program advancements that help to make the program smoother to use and easier to understand, and change. One of the big things she wanted to focus on was the idea of making the algorithm subject to "change its opinion" as the user entered more examples of an item. In many cases once the user as set a learning mechanism they decide to move it in another direction and fluid movements that allow the algorithm to be "reprogrammed" make it a much more user-friendly experience. The author closes by talking about how more sophisticated machine learning software can make experts more comfortable with using these kinds of adaptive tools to more broad ends.

I think these kinds of systems are very cool and I think this is a good field for more research to be applied. It is true that machine learning is a very hard task for computer programmers but as we learn more about AI and have more people in the field that understand the basics we will start to see new trends develop and there will eventually be a call for more advanced machine learning systems to help assist with tasks that humans might otherwise find very difficult. I have always envisioned systems like this making their way into crime labs and being able to analyze tape and sound to see what is happening in a particular crime or point out important details that humans might otherwise lose. If this were the case we might have a whole new way to establish justice and putting the right people behind bars might be less about lawyers and more about who is actually in the wrong. It is an interesting idea to have algorithms that seem to be "willing" to decide to change their mind and go in another direction given more test cases but it makes sense that a user would want to be able to do that as the kinds of data that might be presented might be changing. It is possible that eventually if we were to build up big enough banks about all kinds of input, audio, visual, touch, smell that we would be able to have computer identify nearly anything and there would be no need to try and figure out "what is that blue-ish fruit in the produce section". I would think that if Google can take a snapshot of the internet and the contents of the world every few weeks to months that we might be able to build up a super computer with information about everything and put this on the internet. I would think.