Government efficiency is a hot topic here in the United States. AI is a hot topic in all governments.

Artificial Intelligence (AI) is exciting and is touted as a great efficiency tool by governments around the world. Let’s compare AI’s potential efficiency benefits to something way less exciting, LegalDocML.

Before a logical comparison can be made between these two, it is meaningful to set the foundation of comparison.

Government efficiency is an operational objective that supports the principle of stewardship. Operational efficiency is also subject to the guiding principles of transparency and legislative clarity (also referred to as accuracy). 

We can root out fraud and bureaucracy, but we cannot diminish transparency or clarity of the law. Businesses focus on operational efficiency, but while protecting overall valuation or profit.

We are comparing apples and oranges; but our goal is to take a look at the operational efficiency benefits of both.

The Efficiencies of Artificial Intelligence in Government

Material science, drug research, supply chain optimization are a few very hard problems that AI not only makes more efficient, but in some cases AI can help us reach milestones that were previously unachievable.

But parliaments are using AI as a co-pilot (not for cancer research). So we’ll focus efficiency on co-pilot benefits by category.

AI Content Creation

Writing summaries of legislation and/or regulation can be helpful. AI can very quickly read and summarize large documents. So I asked Google’s Gemni, what the expected accuracy rate would be for summaries of regulations. The answer was:

• High-level summaries = 80 – 90% accuracy
• Detailed summaries = 60 – 80% accuracy

Why could an AI summary be inaccurate?  Because it does not start with logic, AI starts and ends with language. A human reads language, constructs the logic, and then summarizes in language.

Gemni warned us that human review was essential. 

Why?  Because the stakes are high in regulatory summaries. While they may not carry the authority of actual regulatory text, they are used by corporations to plan. This planning benefits our economies and bad data can result in poor decisioning by corporations (and loss of valuation, jobs, and/or profit).

So us humans are left with a bunch of summaries, which now require us to review. We must check the logic, and edit before it is ready for an additional review and publishing. 

AI Automation of Repetitive Tasks

Automating meeting times, data entry, and report generation is useful.

Accuracy on meeting times is not critical and this can save us a few minutes here and there. When it doesn’t work, it is usually because something was not connected to our calendar (kids at football or wrong account, etc).

Data entry errors might not have a huge impact on a civil servant, but could have life altering impacts on a citizen. So there is efficiency to be had, but we have to consider the impact to customer service. Governments don’t ship the wrong package, but might affect someone’s ability to provide for their family. 

AI Building Software

Yes. Copilots are very helpful when building new government software. We expect inaccuracies (or bugs) and we build quality assurance processes to discover these bugs before launch. With a safety net, we can take the efficiencies and know that any problems will be caught.

Safety nets cost money. They cost money not only when building software, but in all operations.

AI Language Review

Copilots are fantastic at spellcheck and grammar recommendations. We can feed it our drafts for review, and AI makes good suggestions. We get to decide which suggestions to take and which to ignore.

This is very efficient for the human author and eliminates the need for a reviewer (depending on the impact of our document).

Let’s evaluate LegalDocML against the same categories.

The Efficiencies of LegalDocML in Government

AI is a multi-purpose tool. By contrast, LegalDocML is a single-purpose tool. It is very good at defining the structure and metadata of a legal document.

This is an unexpected comparison, AI is a dynamic and exciting tool with the promise to free humans of mundane knowledge work. While LegalDocML is an XML document structure; a static definition of terms.

The goal of this comparison is to evaluate practical, operational efficiency. Here we go.

LegalDocML Content Creation

LegalDocML creates nothing. It simply classifies data and metadata to make it machine-readable. Computer programs can read and reuse data associated with LegalDocML definitions.

We can write programs to create content based on LegalDocML data and metadata stored in our documents. For example, a list of amendments can be automatically generated and inserted into a bill document. That list might be numbered by date or author; or maybe we require one of each. LegalDocML makes that automation efficient and possible. We have automated the creation of content that does not require creativity but instead requires accuracy.

Another example is the creation of amending language like “On line 15; delete the word ‘exact’ and insert the word ‘approximate’.” When we draft in LegalDocML, we can automatically generate these types of documents and many others with exact and deterministic precision.

LegalDocML Automation of Repetitive Tasks

We might consider the above examples of content creation as repetitive tasks. Another example of LegalDocML making our work more efficient is in the publication and printing of legislative documents.

We can write software (and we do) that takes a LegalDocML document and converts it into a print-ready document with CSS. This can be done with programming, so we are gaining the efficiencies long-term. Governments can eliminate the need for a human to spend hours ensuring that printed documents are accurate and designed correctly.

LegalDocML doesn’t handle all the use cases we could apply with AI. But we can automate the generation of 2nd-order legal documents that are required for our parliaments to function.

Another meaningful benefit is traceability. We know exactly what the computer is doing because it is “deterministic programming” versus AI which is “non-deterministic”.

AI Building Software

LegalDocML cannot write software. It is just a static standard that creates interoperability.

Interoperability is not about sharing your laws with other legislative bodies. Spain does not need to tie its laws to Canada programmatically.

Interopability is about sharing documents, data, and metadata within your organization.

For example, if an office of a government representative drafts an amendment in LegalDocML, it can be shared with a legislative counsel with ease and accuracy. That amendment can then be added programmatically to the amendment list of the bill. If the bill passes, it can flow easily into the publication office and onto the government’s website.

Documents constructed with LegalDocML contain the necessary information to tie the system together; regardless of who built which application. This means that our whole system can be constructed in components, instead of as a monolith.

Component-based architecture creates massive efficiency long-term. It allows governments to make investments which can be upgraded in part and new applications can easily join the system.

LegalDocML gives us a platform for agile efficiency, along with the accuracy of deterministic programming.

LegalDocML Language Review

Drafting good legislation or regulation is not creative writing. It is about logic, structure, clarity, and intended outcomes.

LegalDocML is always used in the context of XML programs which provide structure validation and “correct by construction” methods.

It doesn’t make our writing or speling better. But it does allow us to focus on logic, clarity, and intended outcomes instead of structure. Perhaps well formed legislation is more efficient legislation. We do know that poorly formed regulation can have bad outcomes for our economies.

Let’s try a comparison table.

Table of Comparison

Feature	Artificial Intelligence (AI)	LegalDocML
Purpose	Multi-purpose tool for various tasks; aims to automate and enhance cognitive processes.	Single-purpose tool for defining structure and metadata of legal documents, enabling machine-readability.
Content Creation	Can generate summaries (varying accuracy), draft content (emails, reports), but requires human review, especially for high-stakes documents.	Does not create content directly; enables automation of content generation based on structured data and metadata (e.g., amendment lists, amending language).
Automation of Repetitive Tasks	Automates meetings invitations, data entry, report generation; efficiency gains must be balanced against potential impacts of errors (e.g., citizen data errors).	Automates content generation, publication processes, and creation of 2nd-order legal documents; improves efficiency in document processing and publication workflows.
Software Development	Assists with code generation and debugging (co-pilot); requires quality assurance processes (safety nets) to mitigate potential bugs.	Does not directly write software; provides interoperability and facilitates component-based system architecture for long-term efficiency and upgradeability.
Language Review	Provides spellcheck and grammar recommendations; facilitates efficient editing but human editorial discretion still required.	Does not directly improve language or spelling; focuses on ensuring logical structure and clarity through XML validation and “correct by construction” methods.
Accuracy/Predictability	Non-deterministic; accuracy varies depending on the complexity of the task; summaries require human review due to potential for inaccuracies.	Deterministic; ensures accuracy of automated processes and data manipulation; provides traceability and reliability.
Interoperability	Primarily through standard data transfer, and APIs, but doesn’t guaranteeguarentee consistency of data structure.	Provides specific document and metadata standards for guaranteed interoperabilityinteroparability of legal documents, specifically within a single government body, not between different governmental bodies.
Focus	Efficiency gains through automating cognitive tasks and content creation.	Efficiency gains through standardized document structuring and automated data manipulation, which lends itself to predictable results.
Risk and Oversight	Human oversight and review are crucial to mitigate errors and potential impacts of AI-generated content.	Less risk associated with automated processes due to deterministic nature; still requires oversight to ensure correct data and coding, but less risk from the automated processing itself.

Full disclosure: when writing this, Genmi was used to generate the table above and you can see the changes I made. All other parts of this post were written by a human after having some interesting conversations with legislative civil servants.

Summary

In business, the only directive is to make a profit and grow value for shareholders.

In government, we must balance values and the well-being of citizens.

When we consider government efficiency, we must consider those values against our decisions.

Artificial intelligence is not going away and is something we must learn. Today AI is a “black box”; meaning we don’t know exactly how it works. That should give us pause when we value transparency.

Static, open standards are powerful technologies to achieve efficient government. LegalDocML, Akoma Ntoso, and USLM create long-term operational efficiency.

Xcential proposes that AI is not efficient. It is attractive, holds promise, and creates fear. Therefore it is prudent for us to engage with AI. Learning is paramount.

LegalDocML creates real, long-term efficiency while supporting our democratic values. It is not exciting but practical, and offers no threat to humans. It might even be considered boring. 

But prudent governance is not always exciting.