Table of Contents
T-720-ATAI, ADVANCED TOPICS IN ARTIFICIAL INTELLIGENCE, Fall 2019
See Canvas For Up-To-Date Information, Notifications, Links to Lecture Notes
Instructor: Kristinn R. Thórisson
Teaching Assistants: TBD
8 ECTS Units, full Master's-level course
Readings: Readings in SC-T-720-ATAI-19
Lecture Notes: Lecture Notes in CS-T-720-ATAI-19
Classes: Tuesdays 9:45, Room: V110 | Fridays 9:45, Room: M108
Class Schedule: See Canvas
Discussions: SC-T-720-ATAI-19 on Piazza
The course focuses on the phenomenon of intelligence and how to create a truly intelligent machine. In the past 10-15 years attempts to answer this question has been have been made under the rubric of artificial general intelligence (AGI), developmental robotics and cognitive robotics. Looking further into the future than allowed by mere linear extrapolations of popular technologies being applied in various industries today, the course centers on the issues of intelligence architecture, system autonomy, realtime attention, anytime planning, model-based knowledge representation, and what could be considered holistic integration issues. Ideas from systems theory, constructivist AI, control theory and cybernetics provide a conceptual foundation. The course takes inspiration from the questions asked by the founders of the field of A.I., e.g. Turing, McCarthy, Minsky and others, – What is intelligence? and How can we implement intelligence in a machine? – as well as the ideas of cyberneticians and early pioneers of systems science. Historical background and relevant topics from constructionist AI (“good old-fashioned AI”) provide a contrasting backdrop for our treatment of how to build more autonomous and self-contained intelligent systems than possible with today's methods. Relevance of AGI to autonomous robotics and systems operating in the physical world will be addressed.
Prerequisites & Requirements
- Programming experience necessary (LISP, Prolog, Haskel or related is a plus)
- A prior introductory class in one or more of the following is recommended: Artificial intelligence, simulation techniques, cognitive science.
- Patience, resilience, and focus to read at least 2-3 papers per week.
After taking the course, the following should hold. Note that these bullets being achieved depends directly on students diligently attending the classes and doing the assignments, thoroughly reading assigned materials, and actively participating in discussions.
- Students should be able to:
- Identify key challenging research questions related to advanced machine learning (AML), artificial general intelligence (AGI), and general machine intelligence (GMI)
- List methodological difficulties and proposed solutions to building AML/AGI systems
- Explain key components of some AML/AGI architectures, and how these relate to the creation of truly intelligent machines of the future
- Students should have a good idea of:
- The limitations of current AI methodologies
- How the pursuit of AGI differs from development of “narrow AI”
- Some AML/AGI projects in development
- What the main requirements are for building (more) complete minds
- What methodologies are currently available and applicable for building complete minds
- How software architecture plays a central role in AI, robotics, and AGI
- How to apply presently-known techniques and methodologies for building complex AI systems
- Emergence, self-organization, and synergism
- Students will have had hands-on experience with:
- Selected machine learning methods, notably reinforcement learning
- One programming environment targeting AGI
Note: This assignment outline is indicative; until Sept. 30st some details of these assignments, and their percentage of total grade, may change. Please take note of announcements on Canvas.
- Students should hand in their assignments (using Canvas) on time; if there will be any unnecessary delay in handing in the assignment then students must assume the possibility of a lowered grade or a grade of zero.
These will be given in the first half of the course.
- Total 22% of final grade.
- Control Assignment 1. Counts 5% of final grade.
- Control Assignment 2. Counts 7% of final grade.
- Counts 12% of final grade.
These will be pursued in the second half of the course.
- Count 2% each, total 10% of final grade.
- Regular Discussions of reading material will be held for 20-30 minutes in class. Your participation to counts towards your Discussion Topics grade.
- For discussions we will use the forum on Piazza.com.
- After most Tuesday classes in the second half of the course the instructor will post a link on the online forum to a paper or article on an interesting aspect of AI (more likely than not related to the last 2 week's lecture topics). You need to read this paper or article and by Tursday at noon the week after post, under the same forum thread, two (2) questions about the contents of the material.
- Your posting should arrive by 12:00 noon on Thursday the week after assignment (but see exact hand-in time in Canvas - this could vary).
- The questions can point out concepts that you have difficulty understanding, but preferably they should be questions that provoke discussion from the material.
- In the discussion section of the Friday class in the week after, chosen questions from those submitted will be discussed by the group as a whole and you are expected to participate.
- Participation in class discussions is an important part of the course, and mandatory.
- Grading: 10% of final grade.
- Each topic counts equally towards the grade
- Counts 16% of final grade.
- In-detail description: Final Project
- Final Exam will count 40% of final grade.
- Subject matter / focus: Any of the material and topics covered in readings, assignments, discussions, and in class.
- A grade of 6 or higher on the final exam is required to pass the course.
- Hand-in of the assignments is not a requirement to get permission to take the final exam.
- Expect to draw architectural diagrams, write pseudo-code, and write (short) essays in the final exam.
- Closed-book: No helping material is allowed.