The final exam will be a 3-hour open-book on-line exam (Canvas).

Questions will focus on your understanding of the material that has been presented and discussed throughout the course and that which can be found in the assigned readings and lecture notes, with heavy emphasis on the main concepts and topics, and your ability to holistically comprehend these.

(You should know by now (due to your assigned reading About Understanding by yours truly), an agent's (including students') understanding of a phenomenon can be tested by asking them to (a) predict something about the phenomenon, (b) achieving some sort of goal with respect to the phenomenon, © explain something about the phenomenon, or (d) recreate the phenomenon.)

If you are unsure of which topics are the main ones, and which are less important secondary topics, can look at the lecture notes: If something is mentioned there, it is important. If it's mentioned more than once it's even more important. But above all, it is your comprehension (read: understanding) of the relationships between the topics covered in the class, and your ability to put this understanding in context with creating more capable AI systems, that the final exam focuses on.

Below are some example questions formulated in a similar way to (some of) the questions that (may) apprear on the final exam. Note that these are representative of the types of questions, not an exhaustive list. These are provided here to help you prepare for the final exam, refresh yourself on the topics and readings covered, and give you an idea of the scope.

However you answer and whatever you write on the final exam, keep this in mind: Make sure you present strong and clear arguments for your answer, referring back to the most relevant material covered in the course, as appropriate.

1. What is abduction and why is it more complicated in non-axiomatic worlds than in axiomatic ones?
2. How does learning through reasoning work?
3. Is empirical reasoning the same as axiomatic reasoning? Why / why not?
4. In what ways do the mechanisms of a rule-governed world, inhabited by a learning agent, affect its learning process(es)? Illustrate your points with examples.
5. What is the relation between knowledge, information, data, and measurement, and how is this relevant to learning?
6. Give three examples of reasoning methods included in what has been called “ampliative” reasoning?
7. Analogies seem to be an important feature of human cognition. Describe an approach that might achieve transversal analogy making in a reasoning cognitive architecture. What are the main challenges? What are the main benefits?
8. Answer ONE of these questions: (a) Give three arguments why reasoning may be difficult to implement in artificial neural networks. (b) Give three arguments why induction is more difficult to implement in cognitive architectures than other reasoning types.

Note that the answers here below are neither exhaustive nor complete. They are partial and only provided as the beginning of a hint to what the full/complete answers might be (these are not provided). Complete answers can be relatively easily constructed by studying the lecture notes and having read (most of) the papers assigned in the class.

1. 'Abduction' is one kind of reasoning that can be used to handle missing knowledge. In non-axiomatic worlds the rules of the world can never be acquired with certainty and thus conclusions cannot be deemed 'true' or 'false' with absolute certainty.
2. Reasoning methods are used to infer missing information that is used to model (describe the dynamics of) particular mechanisms in the world.
3. No, it isn't.
4. They define the limits of what can and cannot happen; they harbor the (hidden) mechanisms that the learning must model.
5. Measurement produces data; data becomes 'information' when put in particular contexts or used for particular purposes; information is knowledge when it is general enough to be used for several purposes.
6. Abduction, deduction.

2024©K.R.Thórisson