cm3020 Topic 02: Automated Scientific Discovery

Main Info

• Title: Automated Science and Expert Systems

• Teachers: Larisa Soldatova

• Semester Taken: October 2021

• Parent Module: cm3020 Artificial Intelligence

Description

An introduction to automating scientific discovery and expert systems.

Week Six: Introduces scientific discovery systems, and the ‘Robot Scientist’ project.

Lecture Summaries

Can be found in cm3020 Lecture Summaries: Topic 02

Assigned or Discussed Reading

Week Six: Background and History

Essential Reading

• King et al: The Automation of Science (2009)

• Williams et al: Cheaper Faster Drug Develoment (2015)

• Extance: Robot Scientist Discovers Potential Malaria Drug (2015)

• King et al: The Robot Scientist Adam (2009)

Further Reading

• Development of Maholo ->

• Ross King video lecture ->

• Another King lecture ->

• Langley: Scientific Discovery, Causal Explanation, and Process Model Induction (2019)

• Langley and Arvay: Scientific Discovery, Process Models, and the Social Sciences

• Todorovski et al: Modeling and Prediction of Phytoplankton Growth with Equation Discovery

• Zutkow et al: Automated Discovery in a Chemistry Laboratory

• Burger: A Mobile Robot Chemist

• BBC clip on robot scientist ->

• Steinruecken: The Automatic Statistician (2019)

• Kitano: Nobel Turing Challenge: Creating the Engine for Scientific Discovery (2021)

• King et al: Automating Sciences: Philosophical and Social Dimensions

Week Seven: Rational Agents

Perception

• Seminara: Active Haptic Perception in Robots

• Mendez: Electronic Tongues

Adverserial Search and Multi-Agent Systems

• Kitano et al: RoboCup: The Robot World Cup Initiative

• Silver et al: Mastering the Game of Go

Agent-Based Systems

• DeAngelis and Diaz: Decision-Making in Agent-Based Modeling: A Current Review and Future Prospectus

• Adenuga: Agent-based Control System: A Review and Platform for Reconfigurable Bending Press Machine

• Bonabeau: Agent-based modeling: Methods and techniques for simulating human systems

Week Eight: Reasoning

• (repeated): King et al: The Automation of Science (2009)

• (repeated): Williams et al: Cheaper Faster Drug Develoment (2015)

• (new): Soldatava et al: Representing hypotheses ->

Week Nine: Planning

• (repeated): King et al: The Automation of Science (2009)

• (repeated): Williams et al: Cheaper Faster Drug Develoment (2015)

Week Ten: Meta-Analysis and Communication

• Closed Loop Cycles -> (focus on communication)

• Meta QSAR paper ->

• Meta QSAR lecture ->

Lab Summaries

The week 7 lab has you working with the rdflib Python library to parse and manipulate some triples. The ‘drugs’ seem to need to be hard coded at this point as their drugginess is not represented in the ontology at the moment.

The week 8 lab has you work with kanren to adapt the hypothesis generation of the toy scientist. This lab seemed broken at time of doing it. It suggests creating a list of lists for the rules, which get passed along to the Kanren is-a function. There seemed to be a couple of issues with that. Kanren didn’t seem to like lists as it wanted hashable objects, so switching to tuples worked ok. It then didn’t seem to like the two element members of the tuple, so it had to just be a tuple of possible values for y in is-a(x,y) and then it seemed to apply a logical and to only return true if all patterns were matched. Skimming the documentation didn’t shed much light on what they might have been aiming for here…

Week 9 planning lab just has you change the message of the planning agent if a chemical is missing to output the name of the missing chemical. You can use the set difference operation to create a set of missing materials and then print them.

Week 9 analysis lab has you parse the csv output of the experiment and calculate the mean potency to select potential drugs.

Related Textbook Notes

DuCharme Learning SPARQL, esp. chapters 1 and 2.