The threat isn’t that AI will take the legal function’s job. It’s that it may train the legal function to stop doing the part of the job that cannot be delegated: thinking.
This is about the person in the chair—the lawyer, the company secretary, the legal director on the board—and what happens when the friction of thought is outsourced.
In metallurgy, strength is not poured in at the end. It is built by work: rolling, forging, controlled deformation. Skip the work-hardening and you can still have a shiny part. You will not have the same resistance to load. Legal judgment is work-hardened cognition. The messy middle—where a messy fact meets an awkward doctrine and a responsible person has to sit with uncertainty—is where original positions are formed. Convenience skips that middle.
Cognitive Offloading Is the Mechanism, Not a Metaphor
A 2025 mixed-methods study in Societies (666 participants) found a significant negative correlation between frequent AI tool use and critical thinking. The relationship was mediated by cognitive offloading: handing analysis to the tool so you do not have to hold the problem yourself (Gerlich, Societies).
This is not “AI makes you stupid” as a slogan. It is a specific behavioural loop:
- A hard question appears (privilege scope, a DORA concentration issue, a disclosure call).
- A fluent draft appears in seconds.
- The draft is good enough to circulate.
- The person never builds the internal model they would have built by struggling.
- Next time, the struggle feels optional.
- The muscle that boards rely on—independent legal analysis—atrophies on a schedule that never shows up in the KPI pack.
Younger, heavier users showed the pattern more strongly in that study. That should alarm any GC responsible for a pipeline of junior counsel. The people who should be forming professional judgment are the most exposed to skipping the formation process.
The “Messy Middle” Is Where Legal Insight Lives
Creative-process research keeps pointing at the same uncomfortable interval: the unstructured stretch where disparate concepts collide before a clean narrative exists. That interval is inefficient. It is also where unique insights form (Tech Research Online).
Legal work has a named version of this interval. It is called thinking about the case.
- Connecting a factual oddity in the file to a doctrine nobody briefed.
- Noticing that the “standard” clause does the opposite of what this counterparty’s incentive structure requires.
- Sitting with the possibility that the comfortable interpretation is the one a supervisor will reject.
A model is excellent at skipping to a structured answer. Structured answers are what junior lawyers produce when they are anxious. Boards then receive papers that are internally consistent and evidentially thin in the only way that matters: nobody lived with the problem long enough to see the join.
Georgetown neuroscientist Adam Green’s team, as reported by Thomson Reuters, found that human-written essays contributed up to eight times more novel ideas than AI-generated ones—even when AI-era text looked more colourful and was rated more “creative” by human judges (Thomson Reuters Institute; related work in Moon, Green & Kushlev). Polish is not novelty. Legal members of boards already know this from a different disaster: hallucinated citations that looked perfect. Offloading is the same aesthetic trap without the fake case law. The memo is real. The thinking may not be.
What Offloading Looks Like in a Legal Board Pack
You will not see a line item called “cognitive offloading.” You will see symptoms:
- Issue lists that are generic. Every matter gets “data protection, contractual liability, reputational risk,” in that order.
- No trail of discarded theories. The paper arrives fully dressed. There is no record of what was considered and killed.
- Verification theatre. Citations are checked; the framing is not challenged. Framing is where the model did the thinking.
- Juniors who cannot brief without a screen. If the tool is down, the analysis is down. That is not resilience.
- Directors who stop asking “what else could this be?” because the pack already sounds complete.
Habitual reliance reduces the mental engagement that strengthens judgment over a career. The Societies study’s practical implication is that frequent, passive use is the risk; deliberate, structured use is the mitigation. The legal translation is old: non-delegable professional responsibility. You can delegate typing. You cannot delegate the duty to understand.
Why Legal Board Members Should Care More Than Other Directors
Other directors can treat AI as a productivity story in operations. Legal members cannot, for three structural reasons.
The product is judgment. A factory can sometimes accept a more uniform part. A board cannot accept a more uniform legal position if the facts are non-uniform. Offloading attacks the product itself.
Training is a control. If juniors offload the messy middle, the firm is not building the next generation of people who can brief a board without a model. That is a succession and resilience issue, not a L&D nicety.
Liability does not offload. Courts have already treated unverified AI legal work as a professional failure. Offloading is the cultural precondition of those failures: the human stopped occupying the problem. You do not need a sanction to have a weaker board. You only need a room full of people who have not practised thinking.
Controls That Protect the Muscle (Without Banning the Tool)
These are operating rules a legal director can put in a policy and a board can ask to see evidenced.
- Thesis first, tools second. The accountable lawyer writes the core proposition in their own words before prompting. The model may attack it; it may not originate it on material matters.
- A mandatory messy-middle artefact. For high-impact issues: a short note of uncertainties, alternative characterisations, and what would change the advice—produced before the polished pack.
- Supervised struggle for juniors. Some research and issue-spotting stays human-first, on a defined class of matters, so the training pipeline still forges judgment.
- Challenge time that cannot be collapsed. If AI shortened drafting by 40%, do not donate all of that time back to the calendar. Spend some of it on dissent.
- Literacy as a duty. People must be able to explain why the recommendation is right without reading the model’s bullets. If they cannot, offloading has already happened.
I help boards and legal directors design AI use so professional judgment stays non-delegable—policy, training, and challenge rituals that survive convenience. Contact me.
Relevant Sources
- AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking — Gerlich, Societies 15(1) (2025) — https://www.mdpi.com/2075-4698/15/1/6
- Is AI Making You Less Creative? — Tech Research Online — https://techresearchonline.com/blog/is-ai-making-you-less-creative/
- Organizations Are Misdiagnosing What’s Killing Their Innovation — Thomson Reuters Institute — https://www.thomsonreuters.com/en/institute/articles/feature-misdiagnosing-whats-killing-innovation
- Homogenizing Effect of LLMs on Creative Diversity — Moon, Green & Kushlev (2025) — https://doi.org/10.1016/j.chbah.2025.100207
- Practical Rules for Safe AI Use — Goldmanmalka — https://goldmanmalka.com/practical-rules-for-safe-ai-use/