Software developers have been automating low-level document review decision-making for years. That automation helps lawyers zero-in on relevant language faster. The end game? Faster review speeds can result in lower costs for clients and improved competitiveness for lawyers.
The promise of document analysis software
Anne McNulty said modern document analysis systems boast significant time savings and accuracy. “It saves anywhere from 20 to 60 per cent of your time,” said the director of legal knowledge engineering for contract analysis software developer Kira Inc., adding that computer systems are less likely to make mistakes when they work late at night.
Mitigating issues with document analysis
Lawyers who successfully implement analysis systems into their review work have dealt with issues like the ones listed here.
Unrealistic user expectations
At present, artificial intelligence (AI) amounts to increasingly sophisticated algorithms that do low-level intellectual tasks. Skynet (from the Terminator movies) it’s not.
Machine learning, for example, is also the result of advanced algorithms. But machines learn only what lawyers teach them.
Different algorithms lead to different results
Two separate tools that boast artificial intelligence may not deliver the same results, “which I find problematic,” said Dominic Jaar. The KPMG Canadian advisory leader for clients and markets likened the issue to running an identical search using both Google and Bing. They run different algorithms, so they return different results.
To avoid this issue, “all parties (in a matter) should use the same platform,” Jaar said.
File support
Few document analysis tools handle all types of computer files. This can mean they don’t render a given file type in the same way other tools do. In some cases (such as AutoCAD) they can’t open the files at all.
Spreadsheet files, for example, can prove tricky, since not all tools expose the functions within each cell. Similar issues arise with structured databases. Viewing such files in their native format may be the only option.
When lawyers don’t have the software (few have licences for AutoCAD, for instance), they may review files remotely using a shared licence.
User experience (UX)
Once widely referred to using the term “user friendliness,” the UX of certain tools reflects the needs of knowledgeable users only. Reports may be equally unfriendly. “Sometimes they provide too much information, or not enough to make an informed decision,” said Chuck Rothman, director of data engineering and analytics for information management firm MT>3. “It’s hard to figure out what the system tells you.”
This problem is for software developers to solve. For example, Aird & Berlis recently seconded M&A lawyer Aaron Baer to contract review software developer Diligen “to work alongside our developers and designers to create new AI use cases,” said Konrad Pola, Diligen’s CEO. “Real-world insights from practising lawyers are key to addressing emerging legal tech needs.”
Putting software expertise before people
Pola described a common misallocation of resources: “In some areas of law, like mergers and acquisitions, the stakes are high; missed clauses can have material impacts on the outcome of a transaction — and often this task falls to the most junior members of the team.”
Paying a higher rate for experienced reviewers who understand the nature of a case and applicable law may well lead to lower costs and higher quality review than a “brute-force” approach.
Assistants, not replacements
Fear of technology replacing lawyers is overblown. Software can organize information far more quickly than humans can, but it still can’t make decisions based on nuances in language. To prove the point, try this experiment: copy this paragraph, paste it into Google Translate, translate it to another language, and then translate the result back to English. The resulting gibberish ought to reassure any lawyer who reads it.
When document analysis tools eliminate drudgework, lawyers can use the time saved to learn, to delve more deeply into relevant language, to make better decisions.
“It doesn’t take learning away from junior people,” McNulty said. “I think they learn more. Instead of doing menial tasks, you get to the meatier part faster. You have the mental energy to think about the language.”
Future developments
Artificial intelligence appears to be a long way from replicating human intelligence. However, software developers are recognizing ever more ways to automate basic tasks. MT>3’s Rothman noted that more powerful machine learning means “systems will require less human input to provide results.” He cited his predictive coding experience: several years ago, it took him 15,000 to 20,000 records to train a system. Now he gets comparable results using about 4,000 records.
Document custody could benefit from the introduction of new blockchain technology. Creating a workspace could trigger the creation of all required fields, perhaps customized for the type of matter that workspace serves.
“Whenever document reviewers do something they think is low value, they ought to ask for it to be automated,” KPMG’s Jaar said.
This article originally appeared on The Lawyer’s Daily website, published by LexisNexis Canada Inc.