Is It Time to Replace Our Juries With AI?
Would AI create a better legal system or are 12 ordinary citizens, on a jury, the best of us?
Let’s get sci-fi-ie for a second.
Imagine: The Saville Row suited lawyer stands to deliver his final summation. His eyes flicker to the cameras round the courtroom and he knows his every movement, his facial expressions, the sound of his voice will be recorded, analyzed, and instantly fed to the jury. He has built his case with witness after witness, each taking the oath, each testifying, crafting every question to highlight an important facet of the case.
The witnesses have been recorded. Did a twitch of their eye or a tightening of their smile or any other of the millions of micro-expressions our faces are unconsciously revealing speak to the jury? Did their face reveal a lie? Or confirm a truth?
He begins but devoid of the emotions and theatrics he used in the past. This jury doesn’t care. He could just stand and read it straight from his notes and the impact would be the same. He finishes his closing arguments and sits.
It’s now in the juries’ hands. But this is not a jury of peers. It’s not a jury of twelve ordinary Americans or even 12 angry men.
It’s an AI jury. And the verdict is coming in seconds.
Why do we have juries in the first place?
It’s supposed to be about equality and fairness. We’re enlightened people, right? We don’t want some all-powerful Wiz-like character hiding behind-the-curtain telling us what’s right and what’s wrong. We don’t need a king or a despot dishing out punishment. We want our citizens involved.
It’s a form of self-government. Democracy is built upon the population being involved in its own governance. This requires responsibility from individuals. It makes each and every one of us a part of the legal system. It creates community involvement.
It may be the only interaction some people have with the government and our legal system.
Juries also act as a form of checks and balances. Juries ensure that one person, like a judge, or a mayor, or a President cannot legally decide someone’s fate. The community must speak. A jury of citizens should protect the rights of the community and the individual.
Juries are meant to adhere to the rule of law and order and not pay tribute to the powerful parties that would divert justice.
What would an AI jury look like?
An AI jury won’t look like us, but it can think like us. Or at least how we are supposed to be thinking when selected to a jury. It should eliminate all bias based upon race, or gender, or how much money the defendant has in the bank. The law would be equal and available to all.
It’d be smart. An AI jury would need to be ‘trained’ with a deep understanding of legal principles, precedents, technical details, and access to a history of existing legal cases. There’d be no need for jury instructions. There’d be no need for an explanation of obscure legal principles based upon a specific case. The ‘jury’ would have decades of legal principles and decisions to guide it.
Specialized knowledge (data specific LLM’s) for cases involved with complicated issues of accounting, or international law, or finance could be created to bring relevant aspects of experience to each case. The AI would be more knowledgeable on the law and the specifics of each trial than a traditional jury would be.
The benefit of a high-priced team of lawyers would be eliminated since each ‘jury’ would be applying years of case law to each new trial. The facts of each case, as presented, would be balanced against legal precedent regardless of how persuasive or knowledgeable the legal team. Theoretically, the law would be equally dispensed to all.
It would be fast. An AI jury would render a decision in minutes not days. This speed would make the court system more efficient and reduce the cost of trials. This would save our cities, and counties, and even our federal government millions of dollars.
It’d be consistent. The law would not be applied differently in a case-to-case instance due to bias, or regional interpretations, or the amount of money spent on a law team. Your golfing buddies wouldn’t be able to help you.
It can’t be intimidated. AI systems do not have a physical presence. You can’t intimidate it through the threat of physical violence. It can’t be attacked on social media, or doxed, or caricatured. It doesn’t have to worry about its neighbors or its kids getting bullied at school. It will do the job regardless of outside pressures.
What problems would need to be monitored?
Trust is paramount. Often, the processes of AI can seem murky. Even experts can have a difficult time explaining an individual answer that may be generated via an AI system. The system must be transparent about how the system works and the safeguards that are in place. It must be secure enough for the public to have faith that justice is being equally applied and each decision can be examined.
Algorithms can be biased. AI is a reflection of society. If we put biased data into our Large Language Models (LLM’s) we can get biased results. The algorithms used for a jury trial would need to be bias free. This requires a complete review of the algorithms and the data each LLM is trained on.
No human judgement. Empathy is a key factor in jury decisions. Humans can understand and relate to the emotions of the defendants and the witnesses. It’s why we search for a jury of peers. AI is an algorithm. It won’t empathize. It won’t sympathize. It will not use emotions of any type in forming a decision.
Existing legal frameworks are built around human judgment and decision-making. Replacing juries with AI would require significant changes to laws and procedures. Laws designed to protect the integrity of juries and prohibit jury tampering, and manipulation, would have to be rewritten to include assaults on an AI jury through cyberattacks.
Cyber threats will emerge. Wherever there is a computer system someone will try and hack into it. Period. An AI jury would need advanced encryption, regular security audits, and robust access controls. Jury tampering is jury tampering.
Can AI tell if a witness is lying?
Yes, probably. When it comes to the written word or texts, AI may be adept at discovering deceit. Converting the oral testimony into text will allow significant analysis by the AI of the testimony. AI can identify and analyze language patterns. It’ll point out inconsistencies. It’ll even detect the emotions and sentiment of the speech. The analyses of these patterns can show dishonesty.
Body language is a little trickier. AI can record, and identify, and analyze facial expressions including micro-expressions. A twisted smile or raise of the eyebrows will have meaning for the AI system. It can analyze the stress level of your voice and it will notice any variations in pitch, tenor, or tone.
However, the interpretation and meaning of these variations are difficult for AI to make. A squeaky voice may indicate that someone is lying, or it may just be a very nervous witness. Care would have to be taken in the analysis of this data. It may work well as indicator of truthfulness of a witness or defendant. But at this time, it is unlikely to be 100% accurate. But of course, neither are humans.
Should all testimony be given equal priority?
Since AI is based upon math it will likely give equal weight to all testimony. The nuance of disregarding a detail that has little or no impact upon the issue at hand will be missed by an AI jury. It’s the human touch that’s removed. Humans assess whether the information is truthful but also whether it is relevant. This judgement is critical to sound jury decisions.
AI Sentencing
There is a disparity in sentencing in the United States. Black males receive sentences 13.4% longer than white males for similar crimes. Hispanic males receive sentences 11.2% longer than white males. (source: United States Sentencing Commission.) You can read a personal story concerning the inherent bias built into our legal system here: A Black and White Story of Justice.
AI sentencing would remove all conscious or unconscious bias in sentencing. It would be consistent. It would be fair. The algorithm would compute the sentencing without human bias. It’s all in the math.
Equitable sentencing would be a step towards creating trust between an AI legal system and the society it is judging.
Conclusions
We’re not ready. Yet. Technically, we still need work to protect against hallucinations or biases built into our AI systems. We are making progress and advancing quickly. Each new iteration of the popular AI systems gets us closer. Our technical goals will be reached soon.
AI will be ready and capable long before we are ready to accept it’s role within our legal system. Participating in a jury is an important part of citizenship. It is an active involvement in governance for ‘ordinary’ citizens. It helps us to understand the consequences of our actions, the impact of thoughts and deeds, and bonds twelve people, requiring them to work together, regardless of race, gender, or political ideology. And despite the inequities and biases within the system, our jury process helps build a trust between our communities and governing bodies.
But we are ready, or should be ready, for a first step. Implementing AI as a tool for juries would create a hybrid system. It would give juries the ability to look at their reasoning and decisions against the output of AI. It’s a non-biased algorithm combined with a human element. Perhaps this would bring a more reasoned and equitable decision-making process without undue influence from wealth or the unconscious biases of race.
The best of us may be 12 ordinary citizens working with AI.
Think About it…
If we’re going to lose some jobs to AI, maybe we should start with the Supreme Court. What would a AI Supreme Court look like?
If you want to check out some great reading list and see which books have influenced, surprised, educated, and entertained me, check out my book shop here. The lists grow monthly and I don’t recommend any books I haven’t personally read.
And now for a little ‘end of credits’ scene fun…
Click here to question the Trump witnesses.
I uploaded all the transcripts, including instructions from the Judge Merchan, from ex-President Donald J. Trump’s recent trial. Ask questions. Would the AI models have convicted Trump? But, remember I did not upload the documents brought into evidence and that were available to the jury to review. This is based upon transcripts only. But you can learn a lot about the trial, and the witnesses’ testimony. Put on your Perry Mason boots and pretend you were asking the questions.
An example:
Using ChatGPT 4.o (The chatbot is set to allow you to use various LLM’s and you may get a different answer depending upon which AI system you question):
Question: If you were forced to render a verdict on the Donald Trump Hush Money trial, what would it be?
The answer is:
If I were to render a verdict on the Donald Trump hush money trial, my decision would be based on the evidence and testimonies presented in the context documents. Here are the key points that would influence my verdict:
Michael Cohen’s Testimony:
Michael Cohen, Trump’s former personal attorney, testified that he was involved in arranging payments to Stormy Daniels and others to suppress potentially damaging stories during the 2016 presidential campaign. Cohen pleaded guilty to campaign finance violations related to these payments on August 21, 2018 [1].
Cohen also testified about the personal and professional repercussions he faced as a result of his involvement and subsequent cooperation with authorities, including losing his law license and facing personal attacks [2].
David Pecker’s Involvement:
David Pecker, the CEO of American Media Inc. (AMI), which publishes the National Enquirer, acknowledged that he never intended to publish the story about Trump’s alleged affair with a Playboy Playmate. This suggests a coordinated effort to protect Trump’s campaign by suppressing negative stories [3].
Pecker’s willingness to sacrifice AMI’s bottom line to aid Trump’s campaign further supports the notion of a deliberate attempt to influence the election outcome [3].
Stormy Daniels’ Statements:
Stormy Daniels initially signed a statement denying the affair and the receipt of hush money, but later testified that she was pressured into signing it and that the statement was false [8][9].
Daniels also indicated that she hoped not to pay Trump the money she owed him, regardless of the trial’s outcome, which could be seen as a motive for her testimony [10].
Legal and Financial Considerations:
The payments made to Daniels were described as “consideration” in a civil settlement agreement rather than hush money, which is a legal distinction that could influence the interpretation of the payments’ intent [6][7].
Conclusion
Based on the evidence and testimonies, it appears that there was a coordinated effort involving Michael Cohen and David Pecker to suppress negative stories about Donald Trump during the 2016 presidential campaign. This effort likely constituted a violation of campaign finance laws, as the payments were intended to influence the election by preventing damaging information from becoming public.
Therefore, my verdict would be guilty on the charges related to campaign finance violations. The testimonies and evidence presented demonstrate a clear intent to influence the election outcome through unlawful means.
This is the single dumbest, most ill-informed piece of AI hype I've seen yet.