Curriculum Vitae

Peter Kahl

I work at the intersection of AI governance, risk and institutional accountability. My work helps organisations govern consequential AI-enabled decisions so that responsibility remains identifiable, decisions remain reconstructable, and the resulting governance can withstand regulatory, legal and commercial scrutiny.

My perspective combines more than two decades of reliability-critical electronics and systems engineering, advanced legal training, and original research into evaluative control, delegated decision-making and institutional governance.

The apparent diversity of my career — electronics engineering, systems design, law and research — reflects a single activity throughout: identifying the governing structure of a problem, locating where control actually resides, testing the assumptions on which the system depends, and reconstructing an architecture that can no longer adequately answer for its results.

I began doing research and development at the age of twenty-one. I learned inquiry not, initially, as an academic technique, but as an engineering discipline.

Engineering: reality as the final evaluator

My early career was formed in Silicon Valley's analogue and mixed-signal semiconductor industry. At Sipex Corporation in Milpitas and later at National Semiconductor in Santa Clara, I worked on circuit and systems problems for which conceptual elegance was insufficient. A design ultimately had to function in silicon, across process variation, temperature, manufacturing tolerances and the conditions of actual use.

My work included analogue circuit design at both discrete and integrated-circuit level, operational amplifiers, transimpedance amplifiers, variable-gain amplifiers, references, voltage regulation, mixed-signal systems, design for manufacture and design for test. The applications included high-reliability, medical, industrial and space-related systems.

This was research and development in the substantive Silicon Valley sense. The task was not merely to execute a prescribed method. It was to determine what the problem really was, identify which assumptions were false, construct a model that could survive contact with reality, and accept that the design — however authoritative its designer or attractive its theory — could lose.

That formative discipline remains visible throughout my work in AI governance and institutional design.

An engineering standard is never vindicated merely because it is established. The circuit either behaves as predicted or it does not. Measurement can expose an error that rank, reputation and institutional agreement cannot cure. A competent designer must therefore remain answerable not only for compliance with an existing specification, but for the adequacy of the specification itself.

After Silicon Valley, I lived in Hong Kong for fourteen years and conducted business in mainland China, particularly in Shenzhen. There I also worked as a software developer and systems administrator, building and operating the information systems behind a commercial software product — carrying the same concern with functions, failure and control from silicon into software.

I later returned to engineering work in England, including systems, embedded electronics, failure analysis and high-reliability product design. Across these settings, I repeatedly encountered the same structural problem: standards are commonly settled in one place and applied in another, while those who implement them may possess neither the authority nor the institutional standing to question their constitution.

Law: from technical standards to institutional authority

I later studied law, completing an LLB and an LLM concerned with intellectual property, commercial law, corporate governance and artificial intelligence governance. My LLM dissertation, The Third Enclosure Movement , received a mark of 80 per cent.

Law supplied concepts that engineering alone could not: authority, delegation, fiduciary obligation, legitimate discretion, procedural fairness and institutional responsibility. It also revealed a difficulty that has become central to my work. Formal authority over a decision does not necessarily entail control over the evaluative standards or technical systems that produce it.

A board may formally own a decision while depending upon models, metrics or professional categories it cannot interrogate. A legal or compliance function may oversee an AI-enabled process while the operative judgement has already been constituted elsewhere. An institution may accept responsibility for a decision while being unable to identify who could answer for the standards by which it was made.

My legal research consequently moved from asking whether power had been validly delegated to asking whether the person or institution formally holding it retained evaluative control.

From institutional diagnosis to governance practice

Entering English academia after a career in research and development exposed a contrast between systems whose assumptions remain open to empirical defeat and systems in which recognition can precede substantive examination.

Institutional affiliation, disciplinary fit, journal hierarchy, citation, credential and professional status can help determine which claims receive serious attention. This was not merely a complaint about bureaucracy or conservatism. It disclosed a more general governance problem.

Institutions exercise authority through standards of recognition, yet those standards are often only partially visible to the people they govern and imperfectly examinable by those who administer them. The resulting evaluative order may be dispersed across inherited practices, rankings, metrics, professional categories and technical systems. No single actor necessarily constituted the standard, although many actors enforce it.

The same structure now appears in AI-enabled organisations. A model may produce the recommendation, a human may formally approve it, and a board may ultimately own the outcome, while none of them possesses effective control over the evaluative architecture that determined what the system could recognise.

This connection led from epistemic justice and institutional responsibility to AI governance, evaluative sovereignty, recognition systems and the constitutional design of answerable institutions.

AI governance and legal transformation

I direct Lex et Ratio Ltd as an independent research and advisory practice focused on AI governance, risk and institutional accountability.

The work is directed towards boards, legal services organisations, research institutions, public bodies and technology companies seeking to deploy AI while retaining clear responsibility, organisational control and legally defensible governance over consequential decisions.

In practical terms, this means examining how an AI-enabled decision is made, what standards govern it, where those standards were constituted, who may revise them, what evidence is retained, and whether the organisation can reconstruct and defend the decision afterwards.

The objective is not merely transparency. A system may disclose its operation while leaving its evaluative premises untouched. Nor is explanation alone sufficient if no person or institution can be required to stand behind the standards that generated the result.

The more demanding objective is answerability: ensuring that consequential decisions remain reconstructable, that responsibility does not disappear into a distributed technical system, and that the standards governing the decision remain open to examination and correction.

This is also a problem of legal transformation. As AI enters legal, compliance, advisory and administrative workflows, organisations must redesign not only their technology but the allocation of judgement, authority, evidence, escalation and review around it.

Research and governance frameworks

Over recent years I have developed a connected body of work spanning AI governance, institutional accountability, delegated decision-making, evaluative control and innovation governance. It includes peer-reviewed publication, openly licensed working papers, policy analysis and formal submissions to regulatory and public institutions.

Its central proposition is that institutions exercise power not only by making decisions, but by controlling the standards through which possible decisions, persons, risks and forms of knowledge become recognisable.

The work examines several connected governance problems:

The distinction between formal authority and evaluative control
The difference between applying a standard and constituting it
The possibility that responsibility may terminate in no answerable person
The use of metrics to make judgement scalable while obscuring its authorship
The allocation of responsibility across complex socio-technical systems
The inability of a system fully to represent the conditions of its own evaluation
The conditions under which inherited standards may legitimately be reopened
The requirement that an authoritative standard remain capable of losing

These ideas have developed across several related bodies of work.

Institutional accountability and recognition

The Epistemic Architecture of Power develops the proposition that power operates through the organisation of what may be known, credited and acted upon.

Epistemic Clientelism Theory examines relationships in which access to recognition depends upon those who control the channels of institutional legitimacy.

The Recognition Game treats institutional recognition not as a collection of isolated defects, but as a stable governance structure. Horizontal rivalry and vertical dependence combine within a single recognitional order. Metrics do not merely record value; they help determine the forms of conduct, knowledge and capability that an institution can recognise.

Answerability and evaluative control

Delegated Discretion distinguishes authority over a decision from the ability to interrogate the evaluation that drives it.

Distribution Is Not Answerability argues that spreading responsibility among multiple actors does not itself produce an answerable arrangement.

Conformance Is Not Constitution distinguishes responsibility for following a standard from responsibility for why that standard has the form it does.

The Capture of Answerability examines the circumstances in which procedures intended to secure accountability are absorbed by the same evaluative order they were designed to supervise.

The Answerability Fuse translates these concerns into a proposed statutory mechanism: a trigger requiring consequential automated or institutionally complex decisions to become reconstructable.

Artificial intelligence and the evaluative layer

My work on artificial intelligence examines the governance consequences of systems that can act with increasing agency while remaining dependent upon inherited objectives, evaluative structures and institutional permissions.

How Continuity Distinguishes Autonomy from Agency in Agentic AI , published in Discover Artificial Intelligence, argues that autonomy requires more than the capacity to pursue objectives. It depends upon a system's relationship to the continuity, rupture and generativity of the evaluative structures under which it operates.

Objective-Layer AI develops the distinction between systems that optimise within an inherited objective and systems capable of participating in the constitution or revision of the objective itself.

Representational Sealing Theory, developed in The Inside of Generatedness , addresses a deeper epistemic limit: a representational system cannot straightforwardly step outside the conditions through which its representations are generated and exhaustively validate those conditions from within.

Innovation and research governance

Recognising New Capability in Innovation Governance compares the institutional conditions of Shenzhen, Hong Kong and Britain.

Answerable Generativity asks how an institution can reopen inherited evaluative standards without converting innovation policy into unstructured discretion.

Sealed Recognition in Innovation Governance examines why institutions can become closed against evidence that their existing categories cannot recognise.

Standards that can lose

The most recent synthesis is Standards That Can Lose: An Epistemic-Fiduciary Constitution for the University .

Its defining proposition is that a legitimate evaluative standard must remain defeasible by what happens to those it governs, through a route controlled neither by the standard's holder nor by those whose authority depends upon its continued validity.

The paper develops a model of the university as an epistemic fiduciary in respect of its evaluative powers: admission, assessment, credit, appointment, promotion, funding and recognition.

It uses the term ‘epistemocracy’ against its established grain. The term does not mean rule by those certified as knowing. It describes a constitutional order in which epistemic authority remains institutionally answerable and no recognised knower possesses finality.

The method: inquiry before answer

My work begins before the formulation of an answer. I do not assume that a problem has been correctly framed merely because an institution, profession or technical system presents it in a particular form.

I ask:

What standard made the problem appear in that form?
Who constituted that standard?
Who merely transmitted or applied it?
What does it permit the organisation to see?
What does it render unintelligible?
Where does effective control reside?
Who bears the consequences when the standard is wrong?
Who is entitled to require an account?
What evidence is retained for reconstruction and challenge?
What would count as evidence against the standard?
Can the standard itself lose?

These questions move inquiry from the visible decision to the architecture that generated it. Once that architecture is made available for examination, problems that appear unrelated — corporate governance, legal services, academic recognition, administrative discretion, automated decision-making and innovation policy — disclose common structures.

A single career, not a change of direction

My career is sometimes described as a transition from engineering to law and then to scholarship. That description is chronologically accurate but intellectually misleading.

The object changed. The method did not.

In electronics, I examined the structures through which physical systems acquired, conditioned and processed signals.
In systems engineering, I examined the allocation of functions, tolerances, interfaces, failures and control across interacting components.
In software engineering, I examined how abstraction, representation, interfaces, inheritance and layered architectures determine how state, behaviour and authority propagate through a system — and whether the provenance of a result, the location of control and the basis for later revision remain intelligible and answerable.
In law, I examined how legal systems attribute intention, duty, reasonableness, fault, authority and responsibility, and how those attributions determine who may be required to justify a decision or bear its consequences.
In AI governance, I examine how technical systems alter the allocation of judgement, authority, evidence and accountability.
In institutional theory, I examine the standards through which organisations perceive, evaluate and recognise reality.

The common activity is analytical reconstruction: identifying the layer at which a result was actually determined, distinguishing that layer from its interface or public representation, and asking whether the governing architecture remains intelligible, controllable and answerable.

Current practice

I am Director, AI Governance & Risk at Lex et Ratio Ltd, based in Reading, England.

I develop governance frameworks for organisations deploying AI and other complex decision systems. The work combines technical systems engineering, legal analysis and institutional design to help organisations retain control over consequential automated and AI-assisted decisions.

My current work is directed towards organisations whose formal authority has become separated from the standards, models, metrics and technical structures through which their decisions are produced.

I help identify the resulting governance gaps: where responsibility has been distributed without becoming answerable, where a human decision formally remains in the loop but no longer controls the operative evaluation, and where an organisation cannot reconstruct how a consequential result was reached.

The practical objective is to build answerability into the decision architecture before these weaknesses become regulatory, legal, operational or reputational failures.

How I help organisations

AI governance and answerability reviews — mapping where consequential AI-enabled decisions are actually made, which standards govern them, who may revise those standards, and whether the decision can be reconstructed and defended afterwards.
Control frameworks — practical controls for provenance, attribution, standing, escalation, review and remedy across automated and AI-assisted decisions.
Board and executive advisory — making responsibility for AI-mediated decisions identifiable and defensible before regulators, auditors, courts and clients.
Deployment governance — redesigning the allocation of judgement, authority, evidence and review as AI enters legal, compliance, advisory and administrative workflows.

View research frameworks View publications Contact