EDRM conducted a terrific webinar on responsible AI use yesterday! Here are five takeaways from the webinar on responsible AI use in legal practice.

EDRM’s webinar titled Training Is Not Enough: Guardrails for Responsible AI Use in Legal Practice (available for viewing on demand on their webinars page here) had a terrific panel, with Judge Ralph Artigliere (ret.) doing an excellent job of moderating the panel that included Professor William Hamilton, Senior Legal Skills Professor and Director, UF Law International Center for Automated Information Retrieval, Suzanne H. Clark, Esq., CEDS, Of Counsel – Mass Torts Discovery Counsel, Beasley Allen Law Firm, Rose Hunter Jones, Partner, Hilgers PLLC and Dr. Varun Perumal Chadalavada, Head of AI Strategy, OrcaWorcs AI.

The discussion was well organized, with each panelist taking a particular topic as part of the discussion. My five takeaways from the webinar are aligned to each of those segments.

Takeaway 1: AI intelligence vs. human judgment is the great paradox

As part of his discussion, Professor Hamilton identified a profound paradox: as AI becomes more “intelligent” at processing text and spotting patterns, the value of uniquely human judgment becomes more indispensable. Lawyers aren’t mere technicians or information processors; they’re “public citizens with special responsibilities” to their clients and the justice system. While a machine can master text prediction, it lacks the capacity to evaluate fairness, honesty, or social value.

As Professor Hamilton stated: “Judgment asks questions that no machine can answer. Is this the honest thing to do? Is this faithful to the facts as I understand them? Is this a position I can defend publicly? How will others reasonably view this conduct of mine? Does this advance justice?”

The modern firm must protect the environment in which this judgment flourishes. We can’t simply make inquiries to a machine and call it legal practice (like what happened in this case). Instead, we must use AI to handle technical tasks so that our capacity for judgment remains at the center of the work.

Couldn’t agree more.

Takeaway 2: Don’t be a “Flyer” or a “Builder”

Clark noted that legal practitioners typically fall into two distinct buckets – each equally dangerous. “Flyers” move at terminal velocity without a parachute, experimenting with AI tools informally and without a structure for safety. On the other hand, “Builders” are paralyzed by fear, standing still and refusing to pick up the tools until they have perfect materials and zero risk.

Both groups suffer from the same execution gap because they lack practical, repeatable workflows. Flyers risk ethical breaches through a lack of oversight, while Builders lose the competitive advantage of efficiency. Both require the same intervention: a set of digestible rules that make responsible use clear and repeatable. Clark also noted that vague guidance from leadership isn’t a safeguard; it’s simply anxiety with a policy attached.

I love the terms “Flyers” and “Builders” to represent the two extremes in AI adoption.

Takeaway 3: Operational safety requires a three-tiered approach

Judge Artigliere discussed the hierarchical model (which he previously discussed in his excellent article here) to move AI governance from abstract ethics to desk-level execution. This framework ensures that technology supports professional judgment rather than bypassing it. The relationship between these tiers is cumulative, creating a vertical line of human oversight.

• Tier 3 – Standards: Ethical rules, court orders, bar guidance, and client requirements that establish the professional floor.
• Tier 2 – Workflows: Specific instructions on supervision, documentation, and review steps for each task.
• Tier 1 – Embedded Guardrails: Point-of-use safeguards like prompts, warnings, and locks that assist the user in the moment of action.

Standards define the duty. Workflows make it repeatable. Guardrails support execution.

I love it! Comprehensive, yet simple. You need all three tiers to maximize the responsible use of AI.

Takeaway 4: Make AI “supervisable” through repeatable workflows

Jones emphasized that AI must be viewed as a “force multiplier” for judgment, which requires it to be integrated into documented, repeatable workflows. This structure ensures the process is defensible when questioned by a client or a judge. Without documentation, AI use becomes untethered, making it impossible for a senior lawyer to fulfill their duty of supervision.

In a workflow such as witness preparation, the team curates a bespoke set of source materials for the AI to organize and synthesize. The AI may build timelines or identify inconsistencies across depositions, but the final witness outline and strategic determinations remain exclusively with the human lawyer. This process must be backed by a “template memo” or audit trail that confirms the lawyer evaluated both the input and the output.

As Jones noted: “If you cannot explain to me how you did this, how can I go and explain it to a client or a court?” This transparency is the difference between a responsible practitioner and a reckless one.

Documentation leads the way to repeatable workflows – they go hand in hand.

Takeaway 5: Move safeguards to the “point of use”

Dr. Chadalavada highlighted the necessity of shifting from “Human Working Memory” to “Embedded Enforcement.” Traditional training assumes a lawyer will remember every client policy and ethical landmine while working under a deadline. Embedded guardrails solve this by preventing the user from doing the wrong thing in real-time through the software itself. The key, of course, is to use software that embeds these guardrails, which public LLMs won’t typically provide.

For example, a tool could be designed to physically block the input of medical data if a specific client policy prohibits it. Instead of relying on a lawyer’s memory at 9pm at night when they’re tired, the tool provides an immediate warning or a lock. This ensures that organizational policies are enforced at the point of use rather than after a costly mistake has occurred.

Other technical safeguards provide immediate verification of AI-generated content. A lawyer can click a generated figure or fact and see the exact trace of source documents used to produce it. This “one-click verification” bridges the execution gap by making it easier for the practitioner to exercise the judgment required to sign their name to the work.

We’re already seeing that for AI implementation within solutions, including eDiscovery solutions. The more guardrails the solutions can provide, the better.

Conclusion

Now you see why the title of the webinar starts with “Training Is Not Enough”! Responsible use of AI in legal practice requires a multi-faceted approach, which includes training, ethical rules and bar guidance, development of repeatable workflows and safeguards within the AI tools as well. Excellent webinar!

There’s a lot that was discussed beyond these five takeaways from the webinar. Go to the EDRM webinars page here to check this webinar or any other upcoming or on demand webinars they have.

So, what do you think? What do you think of these five takeaways from the webinar? Please share any comments you might have or if you’d like to know more about a particular topic.

Image created using DALL-E-3, using the term “robot lawyer wearing a suit sitting in a law office holding up five fingers”.

Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.