Trust with Accountability, Explainability, and Quality
My garden is maturing nicely. Below you see the current picture of my garden. The intent is to enable a dialogue between one human and another human, a human and a machine, a machine and a human, or one machine and another machine. Or said more succinctly, a dialogue between two agents (human or machine). A new type of collaborative work system with human and machine teaming maximizing the potential of artificial intelligence.
A "conversation" and a "dialogue" are both forms of communication; but they have very distinct differences. Conversations tend to be more informal, spontaneous, unstructured. A dialogue is a particular kind of conversation which tends to be more formal, purposeful, intentional and focused, and structured. Conversation is exchanging words. Dialogue is jointly exploring meaning.
You can have a conversation without dialogue. For example, two people can exchange facts and opinions without ever trying to understand one another. But every genuine dialogue is a conversation because communication is occurring. A pragmatic knowledge management system enables dialogue.
A conversation is the movement of signals between participants in the conversation. A dialogue is the collaborative creation of shared meaning from those signals. Conversation is exchanging words. Dialogue is jointly and reliably exploring meaning.
A financial statement is a dialogue.
Matt Wood points out, using the metaphor of a garden, that "The most ambitious gardeners of all sit inside the most deliberate boundaries." I have well defined, deliberate boundaries.
Nicholas Figay points out, in his book Inhabiting Babel, that "Meaning is always mediated, situated, and carried by human collectives." Information is always in context; structures containing that information provides the context.
Thomas Holmgren points out, "Semantically interoperable information is the foundation for building cost-effective digital processes." Understood.
Jessica Talisman describes the notion of a "semantic knowledge management system" (SKMS) in her book, Ontology Pipeline: A framework for building knowledge infrastructures. I explain that notion in my blog post, Semantic Knowledge Management Systems. Excellent foundation; but even more is necessary.
Architects of the pragmatic web state, "The vision of the Pragmatic Web is thus to augment human collaboration effectively by appropriate technologies, such as systems for ontology negotiations, for ontology-based business interactions, and for pragmatic ontology-building efforts in communities of practice. In this view, the Pragmatic Web complements the Semantic Web by improving the quality and legitimacy of collaborative, goal-oriented discourses in communities." I explain the difference between a semantic system and a pragmatic system in the blog post, Five Types of Systems. Bingo, discourses (a.k.a. dialogue) in communities (a.k.a. human collective).
I point out, in detail, that accounting and accountability needs a pragmatic system, not a semantic system, in the blog post, Accounting Systems Need the Pragmatic Web. (Again, read Five Types of Systems to understand the specific differences.)
The universal technology of accountability provides what amounts to a global open industry standards based leverageable framework which is semantic-oriented, model-based, and empowers (supercharges) AI which is also a conservation law and mathematics oriented computational substrate.
Putting all these pieces together using the Seattle Method, you get a best practices based attempt to clarify, simplify, formalize, popularize, and systematize digital financial reporting which will potentially be used worldwide.
While this approach is specifically designed for financial reporting which has zero tolerance for error and which demands trust and high quality; it appears to me that this framework is generally applicable to taming artificial intelligence.
More specifically, it is guidance for creating what I refer to as a pragmatic knowledge management system which is built on top of and enhances a semantic knowledge management system. This enables the creation of modular, scalable industrial processes. Key ideas include:
- AI-first, Digital-first, Graph-first, Model-first philosophy as a foundation (e.g. embrace new paradigm); you cannot fix a bad foundation
- Pure logic oriented system which fixes fundamental flaws in the relational model (see The Quantum Field Database)
- Embracing Atomic Design Methodology be enable the creation of interfaces which are approachable and manageable by business professionals
- Proper reasoner, rules engine, theorem prover; both rules-based and probability-based
- Understand the need for models, metadata, rules, schemas (e.g. called reporting frameworks in financial accounting)
- With provenance, traceability/trackability
- With governance; minimizing epistemic risk
- With immutable ledgers
- With discrete stable units of information which are not copies of information, they are projections of information
Here is the current graphic of a modern universal accountability framework or Trust with Accountability, Explainability, and Quality Framework:
The output of this, the product that is produced, is trust. Confidence. An effective dialogue.
The following sections explain the moving puzzle pieces and tries to provide examples where I have examples.
Governance
Governance defines what matters, creates objectives, obligations, permissions, rights, responsibilities, norms, establishes mandates, delegates authorities, sets standards, creates rules, provides oversight, establishes risk parameters; governance is the constitutional framework of accountability
Governance creates authority. Accountability evaluates the exercise of that authority. Double entry bookkeeping provides the evidence needed for that evaluation.
The blog post, Towards Semantics and an Algebra for Governance, provides information about governance details. Here is a prototype model for governance expressed using XBRL.
Part of the role of governance is the management of epistemic risk.
Accountability
Accountability binds norms to actors; the obligation of agents (humans or machine) to structurally render an account to principles for stewardship and resource allocation; accountability influences behavior; accountability creates stewards; stewardship is the operational and ethical core of the loop
Accounting
Accounting captures and structures raw business events information into financial transactions; builds the semantic ledgers and information model; establishes control which assures actions conform to governance; determines how to measure, assign meaning, and represent what happened using a standard reporting framework; provides provenance, provides traceability and trackability, provides audit trail
Reporting (and Analysis)
Reporting makes claims; structured aggregates, packages, and exposes state of accounts and changes in state caused by business events via formal interfaces (a.k.a. balance sheet, income statement, cash flow statement, statement of changes in equity); reporting is the interface layer; Analysis uses reporting to create insight. Analysis interprets claims and evidence to support decisions. while reporting communicates what happened; analysis figures out what it means; governance rarely acts on raw claims; governance acts on analyzed claims
Audit/Assurance
Audit systematically tests claims; examines accounting rules; Independently verifies information, providence, controls, control activities, evidence, and compliance with governance rules and norms; provides feedback
Double-entry bookkeeping creates accountable records. Financial reporting frameworks give those records meaning. Auditing provides credibility. Together they form the information infrastructure that allows governance and accountability to function at scale. Double entry bookkeeping solved the problem of creating a coherent account. Auditing solved the problem of believing the account.
Assurance addresses the question: "Why should I believe these claims made in a financial report?" Assurance is the process of increasing confidence in information. An audit is the most formal form of assurance. An audit can be viewed as a meta-accountability mechanism. The enterprise presents an account. The auditor evaluates that account by accumulating audit evidence and using their professional judgement. The auditor then becomes accountable for their opinion.
The enterprise is accountable for the facts. The auditor is accountable for the credibility of those facts. The governance system is accountable for ensuring both forms of accountability function properly. The enterprise gives the account; the auditor gives an account of the account; not a guarantee, reasonable assurance based on evidence and judgement.
If accountability required absolute certainty, modern economies would grind to a halt because no report, audit, or decision could ever be completed. The genius of the audit model is that it establishes a socially accepted threshold of confidence: not certainty, but enough reliability that stakeholders can make decisions and hold others accountable.
That's why auditors issue an opinion, not a guarantee. Reasonable assurance is the level of confidence that makes accountability practical. An enterprise is responsible for a true and fair accounting;
Truth is the objective of accountability. Reasonable assurance is the practical limit of verification. The financial reporting framework defines what a true and fair account looks like. Auditing defines how much evidence is required before society is willing to rely on that account.
An Audit Evidence Graph provides a machine readable database of evidence and a queryable record of auditor judgement. As explained in the blog post, Essence of Audit and Assurance, right now, audit is a human‑driven reasoning process. But once you build an appropriate model which is machine interpretable and structure information per that model; each step is represented as a graph node with strongly typed relationships, you can:
- automatically trace how a risk connects to controls
- automatically trace how controls connect to control activities
- automatically trace how control activities generate evidence
- automatically trace how evidence supports audit steps/audit tests
- automatically trace how audit steps/audit tests inform auditor judgment
- automatically trace how auditor judgment updates risk
This turns the audit from a narrative into a computable chain of causality. Further, the record of auditor judgment is preserved. That preserved record of auditor judgement; if global open industry standards are used; can be queried and analyzed across different audit engagements and new insights and wisdom can be gleaned to better understand and improve that chain of causality, that closed loop. A virtuous cycle is created.
This article, PBCs using XBRL, explains now accounting and audit working papers can be turned into reliable queryable knowledge bases.
Accounting & Audit by Design (A&AD) Framework
Solving the Epistemic Problem of Accountability
- Missing transactions
- Incorrect account classifications
- Electronic spreadsheet errors
- Inconsistent information
- Fabricated information
- Hallucinated AI outputs
- Weak audit evidence
- Ambiguous reporting standards
- Poorly documented judgments
Solving the Epistemic Problem Generally
Additional Information:
- Parallels Between CAD/CAM and BIM with XBRL
- True Cost of Quality
- The Tick Mark Becomes a Test
- Electronic Spreadsheets and the Paradox of Flexibility
- The Semantic Spreadsheet
- Work System
- The Challenge
- Parallels Between CAD/CAM and BIM with XBRL
- The Vision
- Come Together
- From Janitor to Curator: Refactoring Accounting for the Age of Artificial Intelligence

Comments
Post a Comment