Editorial Team By 2026, AI wasn’t just a buzzword in legal tech circles. It was actively predicting settlement outcomes, and the results were immediate. For Sarah Jenkins, a personal injury lawyer in Atlanta, the new AI settlement tools seemed like a godsend at first. Her client, Michael Vance, was a warehouse worker with a nasty back injury from a faulty forklift at a distribution center out by Hartsfield-Jackson Airport. The case, Vance v. Global Logistics, Inc., was complicated, with messy liability and huge future medicals. Sarah had a good feel for the settlement range, but a new AI platform called “LexPredict Pro” was sold as a way to nail the number with near-perfect accuracy. The problem was, she was immediately stuck between using this powerful new tool and upholding her basic duties as an attorney. The real question became: how do you make sure the AI is pushing for justice, not just a quick, efficient number?
Key Takeaways
- You can’t trust an AI’s number blindly. Your professional responsibility means running it against your own experience and the specific facts of the case. It’s your license on the line.
- Telling your client you’re using AI to value their case isn’t optional. It’s about informed consent. You have to explain what it is, what it does well, and where it falls short.
- Your firm needs written policies for using AI. Who gets to use it? What data goes in? How do you document when the attorney’s final valuation differs from the AI’s suggestion?
- The Georgia Rules of Professional Conduct, especially those covering competence and communication, are the ethical guardrails for using any predictive tech.
- To fight bias, you have to demand transparency from vendors about training data and always have a human lawyer review the output. An AI trained on biased historical data will give you biased results.
The Lure of an Easy Number
Sarah first saw LexPredict Pro in action at the Georgia Bar’s annual tech symposium, and the demo was impressive. This Silicon Valley startup claimed its algorithm was trained on millions of personal injury cases, including verdicts from Georgia’s Superior Courts and mediation results from across the state. The sales pitch was that it could digest case facts, medicals, police reports, and even juror demographics in Fulton County to spit out a probable settlement range with a 90% confidence interval. For a smaller firm like Sarah’s, where every hour counts, the idea of automating case valuation for clients like Michael was incredibly tempting. It would mean more time for actual client meetings and hard-nosed negotiation, plus freedom from hours of manual research.
The pitch worked. Sarah decided to run a pilot using LexPredict Pro on Michael’s case. She fed it everything: Michael’s MRI scans that showed a herniated disc, the incident report from the warehouse on Cascade Road, witness statements, and the report from Dr. Eleanor Vance, a top orthopedic surgeon at Emory University Hospital. A few minutes later, LexPredict Pro came back with a number that was way off. It projected a settlement between $150,000 and $200,000. Sarah’s own conservative estimate, which accounted for Michael’s lost wages, his suffering, and the very real possibility of future spinal fusion surgery, was closer to $250,000 to $350,000.
An Ethical Gut Check: Competence and the “Black Box”
That lowball number immediately threw up a red flag, bringing her straight to her duty of competence under Georgia Rule of Professional Conduct 1.1. That rule says a lawyer has to provide representation with the necessary legal knowledge, skill, and preparation. Could she really claim to be competent if she just took an algorithm’s word for it? She knew the AI’s brain is just its training data. The real question was who audited that data for hidden biases.
A 2025 study in the ABA Journal showed exactly what she was worried about: predictive algorithms trained on old case data often just repeated the systemic biases from those past decisions. If past settlements for people in certain demographics were historically lower because of prejudice, an AI would learn that as a valid pattern and replicate it. This isn’t some academic worry. In workers’ compensation and personal injury cases, socioeconomic factors always play a role, and an AI might mistake a history of biased, lowball settlements for a legitimate legal pattern.
Sarah’s approach was to treat LexPredict Pro’s output as just one more piece of evidence, not the final word. She dug into the platform’s methodology, but true transparency was impossible. The vendor, LexPredict Inc. (LexPredict Inc.), gave a general overview, but the core algorithm was a trade secret. That’s the fundamental problem right there. You’re on the hook for diligence under the ethics rules, but the AI’s “black box” design keeps you from seeing how it really works. You’re responsible for a tool you can’t fully inspect.
How to Talk to the Client About the AI
After wrestling with her duty of competence, Sarah immediately had to figure out how to talk to Michael about this. Georgia Rule 1.4 on client communication requires lawyers to keep clients reasonably informed. So does she tell him about the AI’s lowball number? And how? Just saying, “A computer thinks your case is worth less,” would shatter his trust and give him a completely warped view of his claim’s actual value. It could make him feel hopeless or, just as bad, create a false impression of what technology can do.
She decided to be direct but careful. She walked Michael through her own valuation first, breaking down the numbers for his medical bills, lost income, and his pain and suffering. Then, she explained she used a variety of tools to test her numbers, including some advanced software that gave a different, lower perspective. She made it crystal clear that these were just tools to help her think, not decision-makers. Her professional judgment, built on years of fighting insurance companies in Fulton County Superior Court, was what mattered. The conversation worked. Michael felt included in the strategy without feeling like his future was being decided by a machine.
Digging into the Data’s Hidden Biases
Sarah’s gut told her LexPredict Pro’s low number for Michael’s case was likely a bias problem. She started thinking about the data. Michael was a 45-year-old African American man with a high school diploma. What if the AI’s training data was full of settlements for plaintiffs who didn’t look like him? If historical data showed lower awards for certain groups, a sad reality of our justice system, the AI would learn to undervalue their claims. The State Bar of Georgia’s 2025 guidance on AI specifically warned about this, telling lawyers to be on high alert for algorithmic bias that creates unfair outcomes.
For example, if the AI was trained mostly on cases from wealthy suburbs and not from more diverse areas like South Fulton, the values it produced wouldn’t fit Michael’s reality. This is a real-world problem, not a hypothetical. Researchers at the Georgia Institute of Technology have published papers on how neutral-looking algorithms can actually magnify societal biases when you feed them raw historical data. Sarah knew her job was to be a zealous advocate, and that meant she couldn’t let a biased algorithm shortchange her client.
Why the Lawyer’s Judgment Still Wins
In the end, Sarah’s experience with LexPredict Pro just proved what good lawyers already know: our judgment is still the most valuable thing we bring to a case. The predictive analytics gave her a data point, a way to pressure-test her own assumptions, but it was no substitute for her understanding of Michael as a person. It couldn’t gauge his credibility on the witness stand, the personalities of the defense lawyers, or the specific legal arguments that would resonate with a jury in the Fulton County Courthouse. Her deep knowledge of Georgia law on damages, like O.C.G.A. Section 51-1-6, gave her an edge that no algorithm could match.
She used that lowball AI prediction as a strategic tool. It forced her to ask: why is the AI so low? Am I missing something, or is it? Her answer was to hire an independent vocational rehabilitation expert to give a detailed report on Michael’s future earning capacity. The expert’s report was damning, spelling out the severe limits on Michael’s ability to work and the costs of retraining him for a new career. It strengthened her higher valuation and exposed a huge blind spot in the AI: it could count up medical bills, but it had no way to grasp the human cost of an injury.
The case of Vance v. Global Logistics, Inc. settled in mediation. With her own solid valuation, the new expert report, and a clear-eyed view of the AI’s flaws, Sarah negotiated a $310,000 settlement for Michael. That was more than $100,000 above the AI’s highest prediction and right in her own target range. It was concrete proof that while AI is a powerful assistant, it can’t replace an experienced lawyer.
The lesson from Sarah’s experience is that AI is coming, and it can help, but it also puts a bigger ethical burden on us. We have to approach these tools with a healthy dose of skepticism, pop the hood to see how they work, question their outputs like a hostile witness, and be completely transparent with our clients. The Georgia Rules of Professional Conduct already give us the framework for using this tech responsibly. The real future of law is a lawyer using AI to test theories and run numbers, but with the human lawyer always calling the final shots and taking the heat if it goes wrong.
How does an AI come up with a settlement number?
AI platforms digest huge amounts of data from old cases, court filings, settlement numbers, medical records, you name it. By churning through all that information, their algorithms spot patterns. They look at things like the injury type, the defendant, and the jurisdiction, and based on what happened in thousands of similar past cases, they generate a probable settlement range for a new one.
What are my ethical duties when using an AI for settlement math?
Your core duties don’t change. You still have to be competent (Rule 1.1), diligent (Rule 1.3), and communicate clearly with your client (Rule 1.4). In practice, that means you have to kick the tires on the AI’s output, understand its potential to be biased, and make sure you’re using it to help your client, not just to close files faster.
Can an AI’s prediction really be biased?
Absolutely. And it’s a huge risk. If the historical data fed into the AI reflects past biases, like certain groups of people getting lower settlements for the same injury, the AI will learn that bias as a fact. It will then spit out unfairly low predictions for future clients from those same groups. It’s your job to catch and correct for that.
Do I have to tell my client I’m using AI on their case?
Yes, you should. While there isn’t a specific rule that says “disclose all software,” the spirit of Georgia Rule 1.4 (Communication) is all about transparency. When you’re using a tool that directly impacts something as big as case valuation, you need to tell the client what it is, why you’re using it, and that you, the human lawyer, are still making the final call.
How can I make sure my firm uses these AI tools correctly?
First, create an internal policy for AI use. Second, demand training for your lawyers on legal tech. Third, do your homework on any AI vendor before you buy. Most importantly, you must insist on human oversight for every algorithmic recommendation. The lawyer’s judgment, informed by things like Georgia’s discovery rules (O.C.G.A. Section 9-11-26), is what protects the client.
