Research group overview
Principal investigator: Kenneth Haase
Despite decades of sustained exponential improvement in performance
and capacity, computer applications and services remain brittle,
difficult to use, awkward to learn, and unreliable outside of narrow
contexts. Much of the reason for this failure is that they lack the
common sense --- the ability to adapt and understand --- of a six-year
old child. The Common Sense group at Media Lab Europe is giving computer
applications the adaptability, breadth, and integration to interact
fluently, fail gently, and take initiative reliably. Systems with
common sense learn from experience, draw on deep backgrounds of
understanding, and understand the world in multiple divergent but
mutually supporting ways.
Common Sense T9
Nick Hawes, John Kelleher
Predictive text input using the T9 approach is currently the de-facto
standard for text entry in the mobile domain. One problem it faces is
the selection of which word to present to the user when more than one
word matches an input sequence. The standard T9 approach selects words
using word frequencies, ignoring the context provided by previous
interactions with the system. The goal of the Common Sense T9 project
is to investigate ways in which common sense knowledge can be used to
augment the standard T9 approach to text prediction, improving the
number of words it correctly predicts first time.
Common Sensing connects everyday context and activity to a range of
online knowledge and information. The idea is to expand individual
awareness of the world in ways which are congruent with our natural
sensing abilities. Common sensing takes large knowledge bases and
determines what information --- both momentary and persistent --- is
relevant to the situation and activities of the user.
Common Sensing is the knowledge-based piece of the laboratory's
initiative on Intimate Interfaces, combining context and background to
present relevant information in natural non-disruptive ways.
This project is actively creating rich semantic knowledge bases to
support both knowledge-aware applications and automatic machine
understanding. These knowledge bases consist of both general world
knowledge and specific knowledge about individuals, geography, and
Experiential computing is a new model for programming intelligent,
adaptive, and autonomous systems. In experiential computing, most of
the structure and logic of the computer's activity is driven by
interactions between the current situation and descriptions of
previous situations (experiences) stored in memory. Abstraction and
generalization emerges mostly from interaction between current context
and past experience.
Grounded Code introduces common sense semantic knowledge into the
development, deployment, and application of computer software. Most
programs already include substantial amounts of common sense knowledge
about themselves and their applications, hidden in comments,
identifier names, and debugging code. The idea behind "grounded code"
is to make this articulation systematic and tie it to deep background
SearchEurope is a prototype interlingual search engine which uses
interlingual concepts for describing web sites and other resources.
As a first pass, it uses the DMOZ web directory, developed by
Netscape/AOL and links that directory to the interlingual BRICO and
XBRICO knowledge bases. These links allow users in different
languages to browse interlingually with BRICO and then find websites
related to the concepts the select from BRICO.
SearchEurope will also allow interlingual annotation, so that
individuals can annotate and categorize the web in their own languages
while allowing users in other languages to use those annotations.
The Viscous Display
John Kelleher, Nick Hawes
The ability to judge similarity is one of the core attributes of human
problem solving. Analogy extends the traditional remit of similarity
to include the structure inherent in many situations. Taking the
structure-mapping theory of analogy as a starting point, the Visual
Analogy project investigates similarity and analogy, and their
efficacy for solving different problems. In particular, the results of
this research have been used to develop a system that solves visual
analogy IQ test problems.