The summary written here by the AI is slightly wrong in some respects, but hilarious. In some ways the rewrite is very insightful, maybe better. This second-grader version is certainly much more fun to read than the original overly-adult Tar-Class it summarizes. (Note: I have a grandchild in second grade.) I love it and hope you do too. It might even cause you to read my grown-up version. By the way, the images here are all by OpenAI’s Dall-E 2.
A long time ago, in 2009, a company called Recommind made a computer program called Axcelerate to help lawyers do their job. The program used a special way of learning called “machine learning” to help sort papers into two piles, important and not important, based on just looking at some of the papers. Lawyers used this program to help them make decisions about what papers were important in a big case. They worked with the computer program to teach it what was important, then the computer program did the rest. Later, another company bought Recommind and the Axcelerate program isn’t used much anymore.
A lawyer named Losey didn’t like how the program worked because it kept some parts secret from him, but he wasn’t sure how to argue against it. He tried talking to the people who made the program, but they didn’t understand how lawyers work. So, he just made the program work the way he wanted it to. The people who made the program didn’t understand that lawyers are important and have a hard job to do, so they were making things difficult for the lawyers. But some other people, like Gordon Cormack and Maura Grossman, were nicer and helped the lawyers understand how to use the program.
Four Generations of Predictive Coding
Predictive coding is a type of software used by lawyers to help them find important papers. The first version of this software had two steps: teaching the software how to find important papers and then actually looking at the papers the software found. But later, people started to use a new version where you keep teaching the software as you look at the papers. This is called “Continuous Training” or “Continuous Active Learning.”
A man named John Tredennick started using this new way of using the software and he called it “Predictive Coding 2.0.” He said it was better to keep teaching the software instead of just doing it in two steps because then you can keep finding more and more important papers. Some other people also started using this way of doing things, but they still used a secret set of papers to help teach the software. But Losey did some experiments and found that using random papers was not very helpful. Now, Losey uses a different way to teach the software, which is called “hybrid multimodal” predictive coding 4.0.
Predictive Coding 3.0 is a way to search for information that is different from older ways. Some people used to use something called secret control sets, but now they don’t use them anymore. This is because control sets don’t work well in legal search. Legal search is when lawyers look for information to help them with their work. The control sets would give false information because what is considered important or not important can change over time.
Scientists made the old versions of predictive coding, but they didn’t understand how lawyers work or how they search for information. They only knew about statistics and didn’t understand the way relevance can change. Nowadays, things are better and many companies make better predictive coding software, but some still make the old version.
At the beginning of a search, it can be hard to know what is important and what isn’t. As the search goes on, the understanding of what is important can change. The control sets would be based on what people thought was important at the start, but by the end, that might not be the case. This is why the control sets don’t work well in legal search.
Some experts in information and technology, like William Webber and Douglas Oard, learned about these problems and started to help make better ways of searching. They learned from people like Jason Baron who is an expert in legal search. They believe that only at the end of a project can we really know what is important and have a true measure of what we found.
Control sets are a way some lawyers use to help them find important papers, but it’s not a good way. It can cause their work to not work right and make them waste time. Lawyers may not like it because it’s too complicated and it seems like fake magic.
Some lawyers don’t want to spend a lot of time going through papers that aren’t important, so they don’t want to use control sets. They would rather do other things with their time.
People who make the software want to stop using control sets because it causes problems and doesn’t work well. They made a new way called Predictive Coding 4.0 to help make it easier and work better.
Using control sets doesn’t work well because the papers that are important are usually rare, so it’s hard to find them. And if they did find them, they would have to look at a lot of papers that aren’t important, which would take a lot of time.
In the end, control sets can cause more problems than they solve, so it’s best to not use them and try Predictive Coding 4.0 instead.
Predictive Coding 4.0 is a way to help computers understand and sort information. It’s still the best way to do it in 2023. This method is like putting two steps into one and it doesn’t keep secrets. Now, there are many ways to do it. But Losey only knows one way that he made and changed in 2016. He doesn’t own the idea, but two people named Gordon Cormack and Maura Grossman do. They also teach this stuff and they’re married now. They made software that uses this method but they don’t sell it.
Gordon and Maura wrote about problems with other ways of doing this and said their way is better. They said the old ways were too hard to use and the computer wasn’t doing a good job. But their way is better and easier. They also said that the old ways didn’t work well when you have a lot of information and only a little bit is important.
Their way doesn’t use secret control sets like the old ways. They wrote about this in a special paper called a patent. They said they made their way better by making it easier to use and not having to use secret sets. They also made their own software that does this, but it’s different than Losey’s.
Predictive Coding is a way to help people find important information in a big pile of papers or documents. There have been many different ways to do this over the years.
The first type of software was really hard to use and needed a lot of help from experts. That’s why we don’t use it anymore.
The second type was a little better because it helped you learn as you went along, but it still had some problems.
The third type was a big change because it got rid of the secret rules and made it easier for the experts to do their job. They didn’t have to spend as much time on it and the information they found was more accurate.
The fourth and latest type is called Intelligently Space Training. It’s like a smart way to find the important information, and it’s what we’ll be learning about in this class. It’s a lot easier to use and helps the experts find the information they need even better.
The way we learn about finding important papers is called Predictive Coding 4.0. We don’t use secret ways to pick papers anymore. Instead, we use random selection to help us pick the right papers and make sure we find all the important ones. We do this with the help of contract lawyers and without much help from SMEs. After we’ve gone through all the papers, we check again to make sure we found everything important.
In Predictive Coding 4.0, we start with a big or small set of papers and begin our training right away. There are no rules for how many papers we start with or what kind they are. We will learn more about this in our TAR Training class.