Discussions about artificial intelligence (AI) often fall into simple, polarized stories. AI will bring great prosperity or bring the end of civilization. It enhances human capabilities or destroys them. People who use it are innovative – or lazy, dishonest, or intellectually deficient.
People in the dispute resolution field should recognize this pattern. We have a long history of creating simple binary categories and attaching strong value judgments to them. Consider familiar distinctions between “facilitative” and “evaluative” mediation and “interest-based” and “positional” negotiation. Real dispute resolution practice is much more complicated and contextual than these traditional dichotomous models suggest.
Similarly, some folks in our community make categorical judgments about AI as if it is all good or bad regardless of the context or how people use it.
There are plenty of legitimate reasons to be critical of AI. I discussed many of them in my post, Bummed About AI? Bots autonomously hacking security systems. Drone warfare. Surveillance. Environmental damage. Control by unaccountable corporations and wealthy individuals. Employment dislocation. Hallucinations. Reinforcement of biases. Threats to privacy. Undermining of human judgment. AI slop. And more.
When used appropriately, AI also can be remarkably useful. Advances in medicine, science, and engineering. Increased productivity and improved quality of work. Reduction of repetitive work. Assistance with research and analysis. Improved communication and education. Generation of ideas and alternatives. Development and exercise of good human judgment. And more.
People use AI well and badly, thoughtfully and carelessly, openly and covertly. The key challenge is to promote uses that enhance human judgment and discourage inappropriate reliance on AI in place of human judgment.
AI is already widely used, and its use is likely to increase. Legal institutions increasingly are developing rules and norms that permit AI use while requiring lawyers to exercise independent judgment and comply with their professional responsibilities. Many lawyers already use AI, and many law graduates will enter workplaces where competence in using it effectively is expected. Some legal employers already assess AI competence in recruiting, giving graduates who have developed that competence a competitive advantage.
I suggest that we approach AI with a mindset of “responsible realism.” We should recognize the world as it is, not focus only on how we wish it would be. We should recognize both the promise and perils of AI.
We should think like mediators about AI. This involves carefully considering situations, interests, and options, recognizing complexity, asking good questions, analyzing tradeoffs, checking our own biases, respecting different plausible perspectives, and avoiding all-or-nothing thinking. Responsible realism requires us to pay attention to how people actually use AI and evaluate particular uses rather than rely on overbroad generalizations about “AI use.”
One implication of responsible realism is particularly important: we should resist stigmatizing AI use as a category. Serious concerns about AI deserve serious attention. But treating mere AI use as discrediting ignores enormous differences in how people use it. It also can discourage precisely the transparency, judgment, and responsibility that critics of AI should want to promote.
Stigmatizing AI Use Is Counterproductive
Some people seem to assume that AI assistance necessarily diminishes the value of someone’s work. That assumption can lead people to dismiss AI-assisted work without considering the human contribution or the quality of the product.
Whether particular AI uses are appropriate depends on the context. This is especially important in legal education. Beginning students may need to develop foundational skills with little or no AI assistance before learning to use AI effectively. As students develop those skills, educators may appropriately authorize greater AI use and teach students how to integrate it into their work while continuing to exercise independent judgment. Different courses and assignments also may call for different rules depending on what students are expected to learn and demonstrate.
When AI use is appropriate, stigmatization creates bad incentives. If disclosure of AI assistance causes people to dismiss work without considering how it was created, they have an incentive to hide their use.
We should want people to be candid about appropriate AI use. Disclosure can help readers, teachers, editors, colleagues, and others understand what AI and humans each contributed, whether important information was verified, and who exercised judgment and took responsibility for the result. It should help us evaluate AI-assisted work rather than operate as an automatic disqualification. Requirements may vary among courses, institutions, professions, and publications.
Stigmatization can also discourage beneficial uses of AI. When people avoid AI because its use is stigmatized, they and others lose benefits it can provide, including more and better ideas, better analysis, improved communication, and higher-quality work.
Some critics may wish that we could put the “AI genie back in the bottle,” but that’s not going to happen. An abstinence-only approach could actually aggravate some problems by driving AI use underground rather than encouraging good “AI hygiene” involving verification, confidentiality, attribution, transparency, and responsibility. Just as avoiding candid sex education can contribute to unsafe sex practices, failing to provide good AI education can leave people more likely to use AI like zombies. I discussed this analogy in my post, What Do AI and Sex Have in Common?
People who are concerned about problems related to AI can promote a variety of strategies to reduce the risks, including regulation, responsible industry practices, professional norms, and education. In the legal field, lawyers, courts, bar organizations, dispute resolution programs, technology companies, educators, and others can contribute to a broader “public legal health” effort.
Education about AI is especially important. People need practical knowledge and experience to use AI effectively and responsibly. They also need to understand it well enough to participate effectively in public debates and advocate for responsible policies, practices, and norms governing AI use.
We also should be very cautious about making strong generalizations from empirical research about AI, especially in education. Results may depend on the students, educational setting, particular tasks, amount and type of training, experimental conditions, and AI models used. The technology itself also is changing rapidly. A study of students using AI for one task under particular conditions may provide limited evidence about what students can accomplish when they are taught to use AI thoughtfully for substantially different tasks.
Faculty considering AI use should exercise their own judgment, drawing on empirical research as well as their own experience and that of colleagues in comparable educational settings. Such experience may be more relevant to their decisions than research conducted in substantially different circumstances. For examples of the wide variety of ways that law professors at institutions across the country are incorporating AI into legal education, see the National Law Review’s 2026 Top 50 Legal Innovators in Academia. For a concrete example, see my recent post about Renee Henson’s use of an AI chatbot named Toby to give students practice negotiating with a difficult counterpart lawyer. Renee designs and supervises the exercises, while students must exercise their own judgment in responding to Toby’s challenges.
Focus on How People Actually Use AI
The important questions generally aren’t answered by asking whether someone “used AI.” We should pay attention to how people use it, how much human judgment they retain, and what they actually produce.
People use AI to do a wide range of tasks, with varying benefits and risks. AI can serve as a writer, editor, researcher, tutor, Socratic questioner, role player, critic, devil’s advocate, brainstorming partner, simulator, feedback provider, or source of real-time coaching advice. It doesn’t make sense to lump all these uses together as if they were the same.
Consider two people who both use AI.
Alex types a question into a chatbot, copies the response, and sends it to someone without checking whether it is accurate or sensible.
Blake spends considerable time interacting with AI. She provides detailed directions, questions its assumptions, accepts some suggestions and rejects others, supplies ideas and information it missed, asks it to test her reasoning, checks important factual claims, revises repeatedly, and ultimately decides what she is willing to put her name on.
Calling both activities “AI use” is accurate, but it masks important distinctions.
Alex used AI in “zombie mode” by accepting AI outputs without carefully examining the assumptions, alternatives, or sources.
Blake used AI in “centaur mode,” combining complementary human and AI capabilities while retaining responsibility for decisions and trying to produce better results than either could produce independently.
A critical distinction is whether people use AI to enhance their judgment or replace it. In many contexts, good human-AI collaboration can involve a great deal of human thinking – questioning assumptions, considering alternatives, verifying important information, and ultimately deciding what conclusions to endorse. Merely knowing that someone used AI doesn’t tell us whether it enhanced or replaced that person’s judgment.
At the same time, not every appropriate use of AI requires human involvement in each individual action. Some routine, low-risk tasks can appropriately be delegated to agentic AI systems with little or no human involvement.
Get Meaningful Firsthand Experience
As members of the dispute resolution community, we should appreciate the importance of experiential learning.
We generally don’t assume that someone understands negotiation because they have read articles about it. Because practical experience reveals things that aren’t obvious from theory alone, students and trainees participate in simulations, observe practitioners, and reflect on their experiences.
We should apply that insight to AI. Effective human-AI collaboration requires practice. Results can depend enormously on the information people provide, instructions they give, questions they ask, corrections they make, and judgment they exercise.
This is particularly important for AI critics. Firsthand experience may help them appreciate the benefits of AI use while giving them additional reasons to criticize it. Their criticisms are likely to be more informed, nuanced, and credible when they understand firsthand what people can and cannot accomplish using AI.
AI proponents also can benefit from experience. Actual use can expose limitations and risks that enthusiastic accounts overlook.
We all are subject to confirmation bias to some extent. Getting firsthand experience and being open to evidence that challenges our pre-existing beliefs can help us reach more accurate conclusions and participate in discussions more constructively.
Evaluate Simple Stories Carefully
Firsthand experience is especially valuable because much of what we otherwise learn about AI comes from stories about it, and stories aren’t a random sample of reality.
Spectacular AI failures are highly visible. For example, lawyers filing briefs containing hallucinated cases deserve the attention they receive. These episodes are concrete and serious.
Successful routine uses of AI are inherently less newsworthy. Thousands of lawyers may use AI successfully for research, analysis, document review, or drafting without producing anything that becomes a news story.
The negative stories are necessary and important. But a collection of individually accurate stories can produce a distorted impression if dramatic failures are much more visible than routine successes.
When we identify problems with AI use, what practical changes in training, supervision, incentives, or procedures might reduce those problems while preserving worthwhile benefits?
AI technology, practices, and social norms are changing rapidly, so we should be cautious about our current judgments and remain willing to revise them as we learn more. We should evaluate simple stories based on a range of information, including research, reliable data, careful reasoning, and firsthand experience. We also should consider whether vivid examples we encounter are representative, what important information may be missing, and whether alternative explanations fit the evidence.
Judge the Work and the Human Judgment Behind It
If the fact that someone used AI isn’t sufficient to evaluate the work, what should we consider? Here are some useful questions:
- Is the reasoning sound?
- Are important factual claims reliable?
- Does the work make a useful contribution?
- What did the human contribute and what did AI contribute?
- Did the person exercise independent judgment?
- Was important information appropriately verified?
- Was AI assistance disclosed when appropriate?
- Did the use comply with applicable rules and expectations?
- Is the person willing to take responsibility for the result?
Answering these questions doesn’t guarantee good AI use. Nothing does. But they focus attention on what matters: the quality of the work and the human judgment and responsibility behind it.
People ultimately must decide what information to trust, what conclusions to endorse, and what work they are willing to put their names on. At the same time, thoughtful AI use can provide opportunities to develop and exercise human judgment.
AI presents serious problems and substantial opportunities. Responsible realism requires us to take both seriously and resist categorical judgments about “AI use.” We should distinguish among very different uses, encourage appropriate transparency, and evaluate AI-assisted work based on the work itself and the human judgment behind it.
That’s harder than deciding that AI is simply good or bad.
Unfortunately, reality has an annoying habit of being complicated.
