Over the years, a number of mathematical theories has been stated on how to reason in the presence of general principles, guidelines and rules of thumb, that can potentially conflict with each other. We have also seen a neat number of implementations, where these theories have formed the basis for computer software for decision support and such. One particular challenge, however, is for these systems to explain to the end user the correctness of their analysis. After all, the underlying theory is highly mathematical, and not necessarily clear to those not having a background in this kind of maths. My job is therefore to try to transform these theories into something normal people can understand. At the moment, I’m working on explanations that are based on interactive discussions between man and machine that are very similar to the kind of discussions people have with each other.
Monthly Archives: February 2013
Computer systems that write in English have come a long way in the last 20 years. Now the most advanced computer systems can do things like write summaries of a hospital patient’s progress for the attention of the nurse coming on duty. In this example, until computers took up the burden of writing, the nurses had to spend time writing that they could have spent nursing: soon they will be able to spend all their time on the nursing. When computers write in human languages like English for the attention of people, we would like them to write in a way that is clear, concise, and understandable. The computer scientists who build the programs that do the writing have some say on how the writing is done: however we still do not know how to find out whether people do find the writing clear and intelligible. In my work on sentence processing I use an eye tracking machine to show how people’s eyes move through sentences as they read them. Methods like this are used to find out how sentence processing works, and these methods can tell us if readers find a sentence easy or difficult to read. I hope to adapt eye tracking methods so that they can tell us which of several ways to write a sentence for the best effect when it comes to writing sentences about plans.
I’m Nir, an investigator on SAsSy. I’ve always been a tinkerer, spending a bit of time on some problem I find interesting before moving to the next shiny thing (not something I actually recommend to anyone). Most of my tinkering has focused on various aspects of artificial intelligence, with the aim of “building machines that think”. While we’re a long long way away from achieving this goal, a good first step seemed to be thinking about how we think. It seemed to me that people tend to think in arguments (e.g. I should do this because of X,Y,Z; X because of reasons A and B, etc), and this led me into studying this topic. Along the way, I discovered that we’ve already got some very powerful AI tools, but that these are typically impenetrable to people. Given this, it seems to make sense to try describe how these tools operate in the same way people reason, i.e. using arguments, and here I am, working on SAsSy and trying to build arguments to explain how machines think. If you want to find out more about me, see this page.
I am interested in how computers can adapt themselves to humans. People differ in many ways. For example, we differ in knowledge, experience, and personality. We differ in which information interests us, how we prefer to process information, and whether we prefer pictures or words (or even logic). The Sassy project is trying to explain to people why the autonomous systems -agents- have agreed upon a particular plan. Explaining every step in the agents’ reasoning can be very cumbersome. Some people will just want a quick summary, and particularly of the part of the plan that is most relevant to their job. They may not want to be explained details they already know. Together with Nava, my role in the project is to investigate how best to explain for a particular person. That involves doing studies with people to see what they like best, what allows them to do their job most effectively and most accurately. I will also help with the public engagement activities of Sassy, trying to explain the science. (For an example of a public engagement website of a project I have worked on see the Joking Computer: www.abdn.ac.uk/jokingcomputer)
Hi, I am Roman and I am really interested in programming and building computer systems (software engineering). The point of the SAsSy project is to make it easier for people to understand what is happening inside a computer program when it tries to create some plan. Computers normally “think” in terms of various symbols and this is quite difficult to understand. My role on the SAsSy project is to write a program that can explain what the computer was “thinking” when it was creating a plan. The way I intend to go about it is to create a translator that translates the program’s symbols into English so that people can understand how and why a particular plan was chosen. This process of translating symbols into words is called Natural Language Generation (http://en.wikipedia.org/wiki/Natural_language_generation).
Apart from software engineering, I also like swimming, photography and most of all rock climbing.
I am Wamberto Vasconcelos, and I have always been curious about how us humans think. Thinking is, some say, what makes us humans, and we all do it, all the time. I am not on my own: for many centuries mathematicians and philosophers have studied logics trying to understand and capture some aspects of thinking – specifically, reasoning, or how we can make inferences, as in, “if the grass is wet, it must have rained or someone watered it“. There are many kinds of logics, for different kinds of reasoning (about time, beliefs, etc.) and some of these logics are very useful to support decision making, information management, and software engineering (designing, implementing and testing computer software), to name a few activities. I am keen to find out what kind(s) of logics we will need (to create?) for SASSY. Find out more about me on my web-page.