Law Will Be Unrecognizable in Five Years Due to AI Disruption
Human + AI > AI
By Troy Lowry
In this first post of the new year, I make a bold prediction: because law is both clearly bounded by written agreements and limited by human knowledge, it is ripe for huge disruption by the quickness and completeness AI brings to issues. This trend has already started and will accelerate rapidly. This will upend law as we know it as it becomes almost trivially easy to correlate people’s actions against the law.
Over the end of year holiday break, I decided to do some deep AI work with models. This meant brushing up on my vector calculus1 and my Python coding. I was not successful in making a ChatGPT-like model that would help guide people through the admissions process; at first it looked very promising. Sometimes it was amazing, but unfortunately, it was too unreliable with too many hallucinations and too little consistency. What I did accomplish was reminding me of the strengths of AI: the ability to sift huge amounts of data and make associations.
That led me to think about the areas in which AI is superhuman: chess, the game of Go, protein folding, and, most importantly, recognizing cats in images. All of these have in common that the rules are finite and stable. Once the AI understands the boundaries, it can iterate and teach itself the intricacies.2
While my mention of recognizing cat images was tongue-in-cheek, it’s also important. AI is better than people at recognizing cats in images and live video and does so far faster. Even AI running directly on your laptop can sort hundreds of thousands of images correctly by whether they have a cat or not in the time you can sort one. Speed and accuracy are a powerful combination!
Google’s AlphaGo taught itself to play the game of Go. First it looked at many games of Go played by humans to understand the game. Then it played against itself millions of times using the rules it deduced. This gave it an understanding of the game that surpassed what it had learned from people. The key was first understanding the rules by looking at past behavior.
Our legal code is vast. U.S. Congress alone puts four to six million words of law into effect every year. It would take the average human reader over 300 hours just to read it through once. State and local governments, plus federal agencies such as SEC, FDA, IRS, etc., add many, many more. In short, the sheer volume of laws makes it very difficult for even a dedicated expert to fully understand what all of the pertaining laws are to any given topic.
Now, along comes an AI that reads, and can keep track of, over 5 million words a minute.3 Suddenly, the potential to be able to ask a chatbot to make a legal opinion with all of the relevant citations seems inevitable. Combine that with the ability to comb through case law to see past results and predictive analytics to tell you how likely a case is to succeed.
On the plus side, this will greatly reduce the cost of many legal services and make things like writing and maintaining a will or writing a simple contract obtainable to most people, many of whom can’t afford these services now. However, just as AI writing is much improved by human oversight and interaction, so too will legal AI benefit from careful oversight by lawyers. In short, it will not eliminate lawyers but will make them far more efficient.
On the minus side, AI will be weaponized for use against certain industries and businesses. Those businesses large enough and savvy enough to use AI will gain further advantages over their smaller competitors. As an example, a company launching a new product and using AI to search all patent filings quickly and effectively will be able to better proactively avoid any indication of infringing on another’s patent while at the same time more quickly and easily flag any competitor product that may infringe on these patents. While patent fights are nothing new in the business world, they are about to enter a new era of potency and frequency.
Another inevitability is an increase in court filings due to AI being able to quickly assemble a strong sounding case, predicting the outcome of any legal action taken and determining how and where to best pursue it. Predicting outcomes and jurisdiction shopping are nothing new, but AI’s combination of quickness and accuracy will be unprecedented and will encourage this behavior.
My bold prediction is that in 2024 there will be an AI service where people put their potential case into an AI chatbot. The chatbot collects all relevant information, as well as consent to automatically look at police reports, criminal records, credit reports, and other personal information. It will then assess the case’s chances of success and will forward the most promising cases to lawyers who pay to receive leads from the service. Just like how currently many loans are approved automatically online,4 this will prove so effective that in 2025 many cases will be accepted automatically by the AI, pending final review from a human, of course.
There are many things that might derail this vision. I may be overestimating AI capabilities. I have seen enough hallucinations and bad information from AI to have some skepticism. Also, I might be underestimating the complexities of legal reasoning and how judges and juries are persuaded. AI works best with very large datasets and in many types of cases there may not be enough data to create good predictions. Last, but not least, privacy and regulatory issues may make this model unfeasible.
To close, I want to say that while this is my prediction, I am not in favor of it happening. I think the ability to quickly put a dollar amount on any injury or grievance would be a net bad for the world as it would tilt the axis of law further away from justice and closer to a business only out for profits. I hope that the legal system will proactively address these challenges to prevent such a shift. My aim is not just to predict but to encourage a balance between technological progress and the preservation of justice in our legal systems.
- I have a love-hate relationship with vector calculus. Imagining things in 2084 dimensions is both mind stretching and too difficult to fully grasp.
- Apparently, I’m not the only one thinking about this lately. One of my favorite bloggers goes into a lot of depth on whether AI becoming superhuman at some tasks means that eventually it will be superhuman at many tasks.
- This is a vast underestimation and is changing all the time. Most laptops can read 5 million words a minute. A supercomputer designed to read and understand laws could read many orders of magnitude more.
- In truth, larger loans, such as mortgages, almost always require human oversight before final approval. Online is just pre-approval.