Transcript Artificial Intelligence at Imperial

Artificial Intelligence at Imperial
Dr. Simon Colton
Computational Bioinformatics Laboratory
Department of Computing
Dr. Simon Colton
• Lecturer:
– Artificial Intelligence & Bioinformatics
• Researcher:
– Computational Creativity
• In maths, science (bioinformatics) and arts
• Administrator:
– Next year’s admission’s tutor
What AI Isn’t
• It is not what you read in the press
– Robots will take over the earth [Prof. Warwick]
– Computers will never be clever [Prof. Penrose]
• These are two extremes
– Real AI researchers and educators believe in the
middle ground:
• Computers will increase in intelligence, but not be a threat
AI in General
• AI usually seen as problem solving
– Problems would require intelligence in humans
– This is the way AI is taught
• Some of us see AI more as artefact generation
– Producing pieces of music/theorems/poems, etc.
A Characterisation of AI
• As answers to:
– “How can I get my machine to be clever”
• Seven answers over the years:
–
–
–
–
–
–
–
Use logic
Use introspection
Use brains
Use evolution
Use the physical world
Use society
Use ridiculously fast computers
Elementary, my dear Watson
• Logical approach
– Idea: represent and reason
• “It’s how we wish we solved
problems…
– Just like Sherlock”
• Very well respected
– Established
• 3000 years of development
– Techniques for reasoning
• Deduction & induction
– Programming languages
Introspection
• Logic has limits
– Combinatorial explosion
• “Maybe we’re not logical
– But we are intelligent”
• Use introspection
– Can be highly effective
– Can be problematic
• Heuristic search
– Using rules of thumb to guide
the solving process
BrainWare
• “Maybe we don’t know our
psychology
– But it’s our brains which do the
intelligent stuff”
• And we do know
– Some neuroscience
• Idea is to build:
– Artificial Neural Networks
– Simulate neurons firing
• Networks configuring themselves
• Mostly used for prediction
– E.g., stock markets (badly)
Evolve or Perish
• “Our brains give us our smarts,
– But what gave us our brains?”
• Idea: evolve programs
– Simulate reproduction and survival
of fittest
• Problem Solving:
– Genetic algorithms (parameters)
– Genetic programming (program)
• Artificial Life
– Can we evolve “living” things
The More the Merrier
• “We live and work in societies
– Each of us has a job to do”
• Idea to simulate society
– Autonomous agents
• Each has a subtask
– Together solve the problem
• Agencies have structure
• Agents can
– compete, co-operate, haggle, argue, …
The Harsh Realities of Life
• “But we evolved intelligence for a
reason”
• Idea: get robots to do simple
things in the physical world
– Dynamic & dangerous
• From survival abilities
– Intelligence will evolve
• Standing up is much more
intelligent than
– Translating French to German
– In Evolutionary terms
Brute Force
• “Let’s stop being so clever and
use computers to their full”
– Processor/memory gains have
been enormous
• Can solve problems in
“stupid” ways
– Relying on brute force
• The Deep Blue way
– Little harsh on IBM
A Good Example
• Robotic museum tour guide
– Robot + computers
– And worried researchers
• Who didn’t intervene
• Highly successful
– 18.6 kilometres, 47 hours
– 50% attendance rise
– 1 tiny mistake
• No breakage/injury
• Great science
– Using many approaches
– Won best paper award
AI at Imperial
• Mainly in Computing and Electrical Engineering
– Also in biochemistry, maths, …
• AI in the Department of Computing
–
–
–
–
–
Introduction courses
Logic courses
Advanced courses
Programming courses
Application courses
Logic
• Logic is taught for two reasons
– To enable students to think analytically and at an abstract level
• The mark of good computer scientists
– To give them tools for AI techniques & other areas
• Logic courses
–
–
–
–
–
First year introduction
Computational Logic
Automated reasoning
Modal and temporal logic
Practical logic programming
Advanced Courses
•
•
•
•
•
•
•
•
•
Advances in Artificial Intelligence
Decision analysis
Knowledge management techniques
Knowledge representation
Multi-agent systems
Natural language processing
Probabilistic inference and data-mining
Robotics
Vision
My Research
• Computational Creativity
– Getting computers to create artefacts
• Which we say require creativity in humans
• Past/ongoing
– Automatic generation of mathematical concepts,
conjectures and theorems (theories)
• Current
– Machine learning in bioinformatics
• Future
– Automating the creative aspects of graphic design
Bioinformatics Research
• Computational Bioinformatics Laboratory
– Head: Prof. Stephen Muggleton
• Robot scientist project
– Robot attached to an AI system
• Performs experiments, analyses the results, designs better
experiments, starts again
– Published in Nature (& reported everywhere)
• Metalog project
– Looking at biochemical networks
– Filling gaps, making predictions
– Funded by the DTI
Student Projects
• Students gain a great deal from undertaking
projects
– Abilities to research
– To be self sufficient
– Understanding of a particular subject area
• Projects can also be fun…
Student Projects - Mathematics
• Automatically generating number theory exercises
– Try to beat his classmates
• Inventing integer sequences
– For entry into an encyclopedia
• Making graph theory conjectures
– Try to beat a program called Graffiti
Student Projects - Bioinformatics
• Bioinformatics for the web
– Set of tutorial web pages with little programs in
• Evolving protein structure prediction algorithms
– Using nature-inspired techniques to mimic nature
• Substructure server
– Predicting the toxicology of drugs
Student Projects - Creativity
• Anomaly detection in musical analysis
– Learning reasons why melodies are different
• Automated puzzle generation
– Next in sequence, odd one out, A is to B…
• Pun generation via conceptual blending
– What do you call a vegetable that you wear?
• Evolving image filters
– Growing graphic design algorithms
Evolving Images © Machedo