Sean Lynch of Ricoh on the mutual dependence of AI and human input


At Legaltech in New York I spoke to Sean Lynch, who is is Director, Review Services, at Ricoh eDiscovery in Canada. One of the most discussed topics at Legaltech was artificial intelligence, and I asked Sean Lynch what was happening with AI and what was useful.

Sean Lynch said that AI had perhaps been overhyped and given more importance than it currently had in practical terms. The term “AI” covers a lot of sophisticated software, used by sophisticated people, which produces data models of enormous value to lawyers.

The software itself doesn’t know anything, Sean Lynch said. This kind of software is good at learning that this type of document is good and that one is not, and can amplify that conclusion across very large datasets. It will, however, never replace lawyers. It can make their lives easier and less complex, and enable them to take on more diverse matters. This makes it of particular value to smaller firms.

Sean Lynch spoke in particular of the active learning component of Relativity. The software can take any element from a document, including its words and its metadata, and find others which look similar. The more data you give it the better it gets.

This gives a significant advantage to lawyers, who start seeing important documents in the first few hours of the review not in weeks as used to be the case.

I observed that we see new technologies come and go all the time – they don’t, perhaps, “go” but just become accepted as the norm. I asked Sean Lynch if it could be the same with AI.

It will, he said, take a long time for this happen. He gave as an example home computers for which great promises were made. It took 15 years before those promises developed and home computers became part of society.

The same will be true of AI, he said. We can get excited about it, but it is a long way from becoming what it will eventually become. People come across a lot of similar technology in their everyday lives, with Netflix, Spotify and Pandora as examples of applications which “learn” from user interaction and makes recommendations based on past input.

The proper conclusion from this is that function matters more than labels. One of the consistent themes in this blog, going back well before “AI” became a thing, has been that potential users should go and look at what new technology does and work out what benefits that brings to their business. Its purpose, as Sean Lynch says, is to make lawyers’ lives easier and less complex, and enable them to take on more diverse matters. It is not to replace them.

There is a kind of mutual dependence here. The software depends on human instructions. Armed with those, it can make choices, and do so very much faster than humans can. It then depends on humans to check and validate those choices and the result may be a refining of the instructions. Humans and machines are dependent on each other, whether the aim is to choose a film from Netflix or a category of documents for disclosure.

More from Ricoh