By Dr. Kevin Shepherdson
A colleague recently forwarded me an article with a headline that immediately caught my attention: Judge Learns Lawyers on Both Sides of Case Used AI, Cancels Trial, Kicks Everyone Off the Case.
At first glance, it sounded almost absurd. Lawyers on both sides of a legal dispute had relied on AI in ways that raised such serious concerns that the judge cancelled the trial and removed them from the case. After the initial surprise, a deeper and more uncomfortable question emerged: if both sides were using AI, who was actually doing the legal work?
Was it the lawyers?
Was it the AI?
Was it the humans supervising the AI?
Or was everyone simply assuming that someone else — human or machine — had done the thinking?
This is not just a legal story. It is a warning signal for leaders.
Against this hurried backdrop of AI transformation, many organisations are rushing to adopt generative AI. Boards are asking about productivity. Managers are looking for use cases. Employees are asking which tools to use. Vendors are promising speed, scale, automation and transformation.
But we may be avoiding the harder question: when AI is doing more and more of the work, are humans still doing the thinking?
How we raise this in Gen AI courses
In our Gen AI courses and AI capability work with organisations, we often pose a deceptively simple question: if AI is helping both sides do the work, who is actually doing the work?
This question is no longer theoretical.
In our Apps Design courses for Learning and Development, we discuss scenarios where both teachers and students use AI. The teacher uses AI to design lesson plans, generate quizzes, personalise learning materials and assess student work. The student uses AI to draft essays, summarise readings, solve problems and prepare answers.
So who is learning?
Who is teaching?
Who is assessing whom?
In our Apps Design courses for Human Resources, we explore similar dynamics. HR managers use AI to write job descriptions, screen CVs, generate interview questions and assess candidates. Applicants use AI to optimise their CVs, craft cover letters, rehearse interviews and tailor answers to what the employer wants to hear.
So who is evaluating whom?
Is the HR manager assessing the candidate?
Or is one AI-generated profile being screened by another AI-assisted process?
Now, with lawyers on both sides of a case using AI, the question becomes even sharper. If one side uses AI to produce arguments and the other uses AI to respond — and neither side sufficiently verifies the work — is the court hearing legal reasoning or machine-generated noise dressed up as professional judgment?
The Rise Of AI on Both Sides Of The Table
The legal example is dramatic, but it simply makes visible something we have already been discussing in our Gen AI classrooms: the teacher and student both using AI, the HR manager and applicant both using AI, and now, lawyers on both sides of a case doing the same.
Lawyers and opposing counsel
As described in the Mississippi case, lawyers used AI tools to draft arguments, research cases and prepare submissions, and opposing counsel did the same, leading both sides to file briefs with nonexistent case citations. Used responsibly, such tools can improve access to justice, reduce costs and help lawyers work more efficiently. Used irresponsibly, they can generate fake authorities, weak reasoning and professional negligence at scale.
The issue is not that lawyers use AI. The issue is whether they still exercise legal judgment, verify the sources, understand the reasoning and remain accountable for the work. A lawyer cannot say, “The AI made the mistake.” Courts admit lawyers to the bar, not AI models.
This pattern is not unique to law. It is spreading across professions, institutions and relationships.
We are entering a new phase where AI is not merely assisting one party. It is appearing on both sides of the table. And once that happens, the nature of work changes.
Researchers and peer reviewers
Researchers use AI to draft papers, analyse literature, generate hypotheses and polish academic writing. Peer reviewers increasingly use AI to summarise manuscripts, identify weaknesses and even draft review comments.
This can accelerate knowledge creation, but it also creates a dangerous loop: AI-assisted research is reviewed by AI-assisted reviewers, and AI-generated arguments are assessed by AI-generated critiques. Papers may become more fluent, but not necessarily more rigorous.
The risk is not just plagiarism. The deeper risk is epistemic laziness — a weakening of how knowledge is challenged, tested and validated. If both the author and the reviewer are leaning heavily on AI, who is protecting the integrity of the discipline?
Software developers and code reviewers
Developers now use AI coding assistants to generate code, while code reviewers use AI to inspect it.
This can be extremely powerful. AI can speed up development, detect bugs, suggest improvements and reduce repetitive coding tasks. But software development is not just typing code. It is about architecture, security, dependencies, failure modes, edge cases and long-term maintainability.
If AI generates code that the developer does not fully understand, and another AI reviews it in a superficial way, we may get the illusion of quality assurance. The code compiles. The review passes. The product ships. But hidden vulnerabilities may remain buried until they surface in production.
We have seen this firsthand in our own Gen AI platform, where AI-generated code produced by an outsourced development team disrupted hands-on AI activities and broke several key functions. The issue was not that AI assisted in writing the code. The real failure was that nobody properly reviewed, tested or validated the output before deployment. In the end, we had to remove the team responsible because the incident revealed a deeper breakdown in professional discipline, code review and accountability.
Insurance claimants and insurance companies
Claimants may use AI to draft claims, describe incidents, organise evidence and maximise compensation. Insurers may use AI to assess those claims, detect fraud, calculate risk and automate approvals or rejections.
This creates another AI-versus-AI environment. The claimant’s AI frames the story. The insurer’s AI evaluates the story. The human may intervene only when something goes wrong.
It raises serious governance questions. Was the claim fairly assessed? Was a rejection explainable? Was the claimant disadvantaged by an opaque model? Did the claimant exaggerate the claim because AI helped optimise the narrative? When AI operates on both sides, trust becomes harder, not easier.
Investors and financial markets
Investors use AI to analyse markets, generate trading strategies, digest company reports and predict movements. Financial institutions and markets use AI to price risk, detect sentiment, automate trades and react to patterns.
At some point, the market becomes a conversation between machines. One AI reads the signal. Another AI reacts. A third AI anticipates the reaction.
Humans may still believe they are investing based on judgment, but increasingly they may be participating in a system where machine-speed pattern recognition overwhelms human-speed reasoning.
This does not eliminate human responsibility. It makes it more important.
Students and universities
Students use AI to craft admissions essays, prepare portfolios and optimise applications. Universities use AI to screen applicants, assess fit, detect plagiarism and manage admissions workflows.
This creates a strange possibility. A student may submit an AI-polished version of themselves. A university may evaluate that student through an AI-assisted admissions process. The result may be a match between two artificial representations.
But where is the real person?
Admissions has always involved presentation. AI simply supercharges it. The danger is that authenticity becomes harder to detect, and institutions may reward those with better AI access rather than better potential.
Dating platforms and human relationships
Even dating is not immune. People already use AI to craft profiles, choose photos, write opening messages and maintain conversations. Others use AI to screen matches, detect red flags and decide whom to meet.
In the film Surrogates, humans interact with the world through robotic substitutes. That once felt like science fiction. Today, our digital selves may increasingly be AI-polished, AI-filtered and AI-managed.
Your AI talks to my AI.
Your profile is optimised.
My response is suggested.
The platform ranks compatibility.
But who is actually connecting?
The more we outsource communication, the more we risk turning human relationships into automated performance.
So Who is Doing the Thinking?
Across these examples, we see a common pattern emerging: AI is no longer just helping one party; it is increasingly doing work on both sides of our most important interactions.
It drafts the claims and evaluates them, polishes the profile and ranks it, generates the argument and responds to it, curates the message and screens it. When machines begin to mediate, enhance and even simulate both ends of a relationship — professional or personal — we risk turning genuine human judgment and connection into a series of automated exchanges.
As AI takes on more of this dual‑sided work, we must pause and ask: in all of this activity, who is truly present, and who is actually doing the thinking?
This is Part 1 of a two-part series titled, “Who Is Doing the Actual Work? When AI Is on Both Sides of the Table”. Look out for Part 2 on the DPEX blog next week.
