Literature Survey

Papers

Explainable Planning (1)

2017 ( paper ) Maria Fox, Derek Long, Daniele Magazzeni Abstract: As AI is increasingly being adopted into application solutions, the challenge of supporting inter- action with humans is becoming more apparent. Partly this is to support integrated working styles, in which humans and intelligent systems cooperate in problem-solving, but also it is a necessary step in the process of building trust as humans migrate greater responsibility to such systems. The challenge is to find effective ways to communicate the foundations of AI-driven behaviour, when the algorithms that drive it are far from transparent to humans. In this paper we consider the opportunities that arise in AI planning, exploiting the model-based representations that form a familiar and common basis for communication with users, while acknowledging the gap between planning algorithms and human problem-solving.

Keywords: Planning Explanation Summary Interactive`

• Types of possible XAI Questions:

  • Q1 Why did you do that?
  • Q2 Why didn’t you do something else
  • Q3 Why is your plan more efficient/safe/cheap?
  • Q4 Why can’t you do that?
  • Q5 Why do I need to replan at this point?
  • Q6 Why do I not need to replan at this point?

• The paper provides ideas, challenges and a road-map for tackling each of these question types

• The paper illustrates this on two examples

  • 1 The Rover Domain
  • 2 The AUV Domain

• Conclusion:

  • No clear way to define a good explanation
  • XAIP should not explain the obvious
  • Defining a good metric for explanation is an important issue

• Comments:

  • These 6 question types could be a good starting point for further developing or categorizing an explanation approach

tExplain: Information Extraction with Explanations (2)

2024 ( paper ) Pedro Cabalar, Adrian Dorsey, Jorge Fandinno, Yuliya Lierler, Brais Muñiz, Joel Sare Abstract: We present a narrative understanding tool that takes as an input a narrative in natural English language that contains both sentences describing actions as well as questions and outputs answers to those questions together with natural language explanations justifying those answers. We evaluate our tool on the several tasks of Facebook’s bAbI challenge where it achieves 100% accuracy.

Keywords: Explanation Actions NLP

• Takes a sequence of natural language sentences / questions

• Maps these sentences into an action representation

• Answers the questions

• text2alm is used to extract for the NLP information extraction and mapping to ALM

• The ALM model is then translated to ASP and solved using clingo

• xclingo is used for annotations to provide explanations in natural language

• tExplain is evaluated on the bAbl dataset (task 1, 2, and 6) and achieves 100% accuracy

• Conclusion:

  • For bAbl provides correct answers and also sensible explanations

• Comments:

  • These results are possible through formulaic action narratives that can be parsed using action verbs
  • The explanations are possible though hand-crafted xclingo annotations for the domain rules of the logic program

Explaining Plan Quality Differences (3)

2024 ( paper ) Benjamin Krarup, Amanda Coles, Derek Long, David E. Smith Abstract: We describe a method for explaining the differences between the quality of plans produced for similar planning problems. The method exploits a process of abstracting away details of the planning problems until the difference in solution quality they support has been minimized. We give a general definition of a valid abstraction of a planning problem. We then give the details of the implementation of a number of useful abstractions. Finally, we present a breadth-first search algorithm for finding suitable abstractions for explanations; and detail the results of an evaluation of the approach.

Keywords: Planning Explanation Interactive

• Explaining the difference between the quality of plans
• This is done by abstracting away the details of these planning problems
• Focus on Mixed-Interactive Planning