Actions

Ontolog Forum

The Ontology Summit 2019 Theme

Explainability

An explanation is the answer to the question "Why?" as well the answers to followup questions such as "Where do I go from here?" Accordingly, explanations generally occur within the context of a process, which could be a dialog between a person and a system or could be an agent-to-agent communication process between two systems. Explanations also occur is social interactions when clarifying a point, expounding a view, or interpreting behavior. In all such circumstances in common parlance one giving/offering an explanation.

Some Views from Philosophy and Science

A brief history of why explanations provides some context and includes the observation that among the first known attempts at understanding the why of explanations were those documented among Greek intellectuals and philosophers. For example, to understand and explain the why there was a Peloponnesian War Thucydides defined explanations as a process where facts (indisputable data), which are observed, evaluated based on some common knowledge of human nature. This was then compared in order to reach generalized principles for why some events occur. In the writings of Plato (e.g. Phedus and Theaetetus), we see explanations as an expression using logos knowledge composable by Universal Forms, which are abstractions of the world's entities we come to experience and know. Facts, in this view are occurrences or states of affairs and may be a descriptive part of an explanation, but not the deep Why. Aristotle's view, such as in Posterior Analytics provides a more familiar view of explanation as part of a logical, deductive, process using reason to reach conclusions. Aristotle proposed 4 types of causes (aitiai) to explain things. These were from either the thing's

  • matter,
  • form,
  • end, or
  • change-initiator (efficient cause).

Following Descartes, Leibniz and especially Newton, modern deterministic causality using natural mechanisms became central to causal explanations. To know what causes an event means to employ natural laws as the central means to understand and explain why it happened. As this makes clear some notions of the nature of knowledge, how we come to know and the nature of reality are part of explanation.

John Stuart Mill, provides a deductivist account of explanation as evidenced by these 2 quotes:

 “An individual fact is said to be explained, 
by pointing out its cause, that is by stating the law or 
laws of causation, of which its production is an instance,” 

and

 “a law or uniformity of nature is said to be explained, 
when another law or laws are pointed out, of which that 
law is but a case, and from which it could be deduced.”

The Ontology Summit 2019 is concerned with the role of ontologies for explaining the reasoning of a system. More specifically, the summit will focus on critical explanation gaps and the role of ontologies for dealing with these gaps. The sessions will examine current technologies and real needs driven by risks and requirements to meet legal or other standards.

Inspired by the current DARPA Explainable AI (XAI) Project, (see: https://www.darpa.mil/program/explainable-artificial-intelligence ) the Ontology Summit theme considers the general problem of explanation. We are interested not only in AI systems that can explain their actions and what they believe but also in other smart engineering systems which may cooperate with and aid humans. With the increasing amount of software devoted to industrial automation and process control, it is becoming more important than ever for systems to be able to explain their behavior to humans. Explanations include expressing rationales, characterizing strengths and weaknesses, and projecting their behavior into the future.

Ontologies could play a significant role in explanations since such agents must represent the conceptual framework that supports explanation. Such explanations would include terms for domain and natural world concepts, relations, and activities. Some version of natural language may be used to describe states and actions in terms that people easily understand, as well as the conceptual structures within which dialog, plans and actions take place.

A benefit of the use of ontologies in support of explanations is the potential for improving interoperability between systems that otherwise would not have a common framework for interoperation. The danger is that current efforts for explainability will be brittle as well as siloed, which will produce a large variety of incompatible explanation techniques that individually satisfy the requirement of providing explanations but which are of little use when explainable subsystems are integrated into large scale systems.

In the usual sense, Ontologies are designed knowledge artifacts but exist in computational (operational) environments which allow reasoning and so should also include the ability to reason about/explain what they know and how they have reasoned with this knowledge. In particular, they should be able to express the rationale for the selected use of the relevant parts of an ontology or suite of ontologies; explain the strengths and weaknesses of the ontology; and, when the ontology is in use, explain data that conforms to the ontology.

Introductory Sessions

Explanations and help facilities designed for people

Co-Champions: John Sowa

Basic principle: It's irrelevant how much you know about computer hardware, software, and systems design. When you are faced with some system designed by somebody else, you are as much a novice as anybody who walks in off the street.

Help facilities for any kind of system are abysmal, and they are getting worse -- primarily because the complexity of hardware/software systems is growing much faster than any help facilities that explain how they work.

People who buy a new car, for example, never learn how to use all its so-called "features" before they buy its replacement. Anybody who rents a car will be driving on a highway before discovering features that may be critical for safety.

Issues: How do current help and explanation facilities work? What are the underlying mechanisms for implementing them, invoking them, and using them? How can they be improved? What kinds of AI facilities could improve them? What are the R & D directions and possibilities for improved design?

Goal: Make help and explanation facilities as useful and usable as a guru or geek sitting next to you. Even better, make them as usable as a kind and forgiving friend who knows what you want.

Overview of Explainable AI

Co-Champions: Ram D. Sriram and Ravi Sharma

Main Tracks and Sessions

Kickoff Meeting

All co-champions introduce their tracks.

Financial Explanations

Co-Champions: Mark Underwood and Mike Bennett

Medical Explanations

Co-Champions: Ram D. Sriram and David Whitten

Explainable AI in Medicine has several facets.

  • One facet is explaining a suggested decision made by an computer system and what justification supports it to the medical provider who is diagnosing a malady.
  • Another is explaining a medical process to the patient who might be a participant and their support team.
  • Similarly, there is the explanation and justification to the billed party of such a process and why it is coded the way it is, whether that billed party is the patient, an insurance company or some other financier.
  • A final facet is explaining the medical record to a future reader tracing the process of care.

Some discussion of this is at https://medcitynews.com/2018/10/how-do-you-make-doctors-trust-machines-in-an-ai-driven-clinical-world/ Another useful set of papers by Peter Szolovitz are at http://groups.csail.mit.edu/medg/people/psz/home/Pete_MEDG_site/Publications.html

Explainable AI

Co-Champions: Ram D. Sriram and Ravi Sharma

Narrative

Co-Champions: Donna Fritzsche and Janet Singer with Mark Underwood as consultant

Commonsense

Co-Champions: Gary Berg-Cross and Torsten Hahmann (U of Maine) An early goal of AI was to teach/program computers with enough factual knowledge about the world so that they could reason about it in the way people do. The starting observation is that every ordinary person has "commonsense" or basic knowledge about the real world that is common to all humans. Spatial and physical reasoning are good examples. This is the kind we want to endow our machines with for several reasons including as part of conversation and understanding. System understanding of human perceptual and memory limitations might, for example, be an important thing for a dialog system to know about.

Early on this was describe as giving them a capacity for "commonsense". However, early AI demonstrated that the nature and scale of the problem was difficult. People seemed to need a vast store of everyday knowledge for common tasks. A variety of knowledge was needed to understand even the simplest children's story. A feat that children master with what seems an natural process. One resulting approach was an effort like CyC to encode a broad range of human commonsense knowledge as a step to understanding text which would bootstrap further learning. Some believe that today this problem of scale can be addressed in a new ways including via modern machine learning. But these methods do not build in an obvious way to provide machine generated explanations of what they "know." As fruitful explanations appeal to folks understanding of the world, common sense reasoning would be a significant portion of any computer generated explanations. How hard is this to build into smart systems? One difficult aspect of common sense is making sure the explanations are presented at multiple levels of abstraction, i.e. from not too detailed to tracing exact justifications for each inference step.

This track will explore these and other issues in light of current ML efforts and best practices for AI explanations.


Other potential subtopics:

  • Natural language and discourse generation facilitated by ontologies & knowledge base
  • Explaining machine learning models using ontologies
  • The role that the Open Knowledge graphs could play in explanations.
  • Explanation for industrial ontologies (e.g., IOF) and systems engineering (e.g., ST4SE)
  • Explanation of mathematical, statistical and physical models, both quantitative and qualitative models
  • Explaining reasoning
  • Common sense reasoning
  • Explaining ontologies
  • Capability of Chat Bots of Better Understand/Explain
  • Conceptual interoperability to achieve pragmatic interoperability
  • Human model interaction
  • Understanding explanations (possible overview topic)

[A preliminary Synthesis of this track and Commonsense Knowledge and Reasoning CSK]

Ontology of Ontologies

There has been a proposal to have sessions in parallel with the Ontology Summit to revisit the challenge addressed by the Ontology Summit 2007 - Ontology, Taxonomy, Folksonomy: Understanding the Distinctions was to clarify what everyone means when they use the term "ontology" or when they refer to these semantic structures. While this earlier summit did succeed in making some progress toward understanding some of the distinctions among the various ontology-like structures, it did not succeed in clarifying the term "ontology". Accordingly, it may be useful to revisit this topic. Some potential subtopics/issues:

  • Can the community agree on a definition of "ontology"?
  • Could more progress be achieved by taking a more operational point of view?

The results of the Ontology Summit 2017 are available at Ontology Summit 2007 Assessment Report.