The paradox of the value of verification: Are we overestimating the efficiency that AI can bring to the legal profession?

By Rocio Ramirez

A few weeks ago I read a study from the University of Auckland Law Bar that discussed the verification value paradox in the use of generative AI in the legal sector. Here's the link: The Verification-Value Paradox: A Normative Critique of Gen AI in Legal Practice by Joshua Yuvaraj :: SSRN

And I thought it was a matter of great interest.

As a premise for this paradox, the study refers to the paradox of opportunity risk, which stems from the accelerated and almost hypnotic intrusion of AI into the legal sector. It examines how the legal profession, perceiving the strategic opportunity this technology offers, as well as the potential for improved efficiency, innovation, and competitive advantage, is hastily integrating it into its internal processes without truly understanding the inherent risks of its unsupervised use.

This opportunity risk generates excessive reliance on AI to achieve efficiency and competitiveness. It also leads to overlooking structural problems with this technology, such as hallucinations or a lack of transparency in the generation of its results, with the risks these pose to legal certainty, ethical obligations, and professional reputation.

The growing number of cases containing serious errors, incorrect data, or fabricated references due to AI-generated hallucinations is now widely known. The professional standing and competitiveness of those who used the results without oversight or verification have been seriously compromised.

And here lies the paradox of opportunity risk, which arises when the pursuit of these opportunities ends up generating risks that compromise legal certainty and professional reputation. In other words, the attempt to take advantage of the opportunity that this technology offers creates the very risk it was intended to avoid (not being competitive in the age of AI).

Therefore, the use of this technology implicitly entails the duty to review, supervise, and validate the results generated. This effort will be more demanding the more complex the task we have entrusted to it.

It is certainly difficult to resist delegating part of our workload to a system that drafts contracts in seconds, synthesizes jurisprudence at lightning speed, and anticipates risks with mathematical precision.

However, behind this promise lies an uncomfortable question: are we really gaining as much efficiency as we think? Or, to put it another way, have we properly calculated the hidden cost that reviewing, monitoring, and validating its results entails?

Because it seems the legal profession has underestimated the cost of this verification effort. We've bought into the narrative of efficiency without calculating the price of oversight.

We must not forget that generative systems do not think, they predict. And in that prediction, errors creep in that, in our world, are not trivial: nonexistent normative or jurisprudential references, biased interpretations, conclusions that seem solid, but lack support.

The risk of generating incorrect results requires reviewing every line and every argument generated by AI.

Blind trust is not an option in contexts where accuracy in the interpretation of regulations and case law, considering the specific facts and circumstances of each case, is a fundamental requirement. Where the arguments put forward to support our claims cannot be based on wishful thinking. And where the slightest inaccuracy in these matters can seriously undermine our obligations.

This verification process is hampered by the opacity of the results generated by AI. Because it functions as a black box, we don't know its internal logic or how it arrives at its answers, which complicates monitoring and validation.

The lack of transparency makes monitoring AI indispensable, but paradoxically, more complex than the task we intended to simplify.

This is where the paradox of the value of verification arises. The net value of using AI depends not only on the efficiency it promises, but also on the cost of verifying its results. The formula is:

Net value of using AI = efficiency – verification cost.

But what happens when the verification cost skyrockets? Because the more sophisticated the task we delegate (a complex contractual analysis, the definition of a complicated procedural strategy, or the issuance of a far-reaching legal opinion), the more time and effort we need to verify that the machine hasn't hallucinated.

The supposed savings are diluted by the time and effort that must be dedicated to the review, since the risk of overlooking an error becomes a real threat.

This paradox leads to the following corollary: if we cannot trust any result without human oversight, and there is no way to understand how AI reasoned to generate a particular result, this means that, in practice, the promise of efficiency could become an illusion. This is because the effort of reviewing results has a cost that is rarely included in return-on-investment calculations.

This is where the paradox of the value of verification becomes evident. When the usefulness of artificial intelligence is overestimated because the cost and effort involved in reviewing its results are underestimated, the dilemma arises of whether the net benefit justifies the investment.

Not because technology is useless or lacks value, but because the narrative surrounding it ignores the most critical factor, which is that supervision is not optional, it is non-negotiable.

Does this mean we should reject AI?

No. It means we need to rethink how we use it. The solution isn't uncritical adoption, but a hybrid model that combines automation with rigorous human oversight.

Only in this way can we balance innovation and legal certainty. Because, in law, efficiency without verification is not progress. It is risk.