Machines are good
at specialized tasks
Computers are still no good at thinking about “ordinary” things
Why is giving computers common sense hard?
“Fred told the waiter he wanted some chips”
Deep understanding requires enormous amounts of knowledge
Applications with common sense
What do these
applications
need to know?
The kinds of things we
need
to teach our computers
How can we build databases of commonsense knowledge?
Collecting Commonsense from the General Public
Collecting common
sense
from the general public
Quality of collected knowledge
Lessons learned from Open Mind
Putting the Open
Mind
Knowledge to Use
Three large-scale
knowledge bases
extracted from Open Mind
ConceptNet:
A Semantic Network Mined from Open Mind Common Sense
ConceptNet:
a giant semantic network
Lexico-Syntactic
Pattern Matching Rules
ConceptNet:
Putting Common Sense
to Practical Use
Using ConceptNet in Applications
Reasoning by linking related concepts
EmpathyBuddy:
Inferring emotion of text
EmpathyBuddy:
Inferring emotion of text
Topic spotting in noisy transcriptions
LifeNet:
A Probabilistic Approach to Commonsense Reasoning
LifeNet:
a dynamic model of human activity
LifeNet is a
Dynamic Markov Network
LifeNet supports
multiple
types of inference
Inferring
context
from sensory input
Generating LifeNet from ConceptNet
Generating LifeNet propositions
Generating rule plausibilities
StoryNet:
Treating Common Sense as a Huge Network of Story-Scripts
Stories are powerful way to organize knowledge
Others reasons for turning to stories
Reasoning with stories by analogy
Collecting Commonsense Experiences
Evaluation of collected knowledge
Putting these systems together
An Architecture for Commonsense Thinking
A Reflective Reasoning Architecture
Integrating these reasoning systems
Commonsense
reasoning
about human life:
a powerful new technology