Watch Benjamin Alarie, co-founder and CEO of Blue J Legal, discuss the expanding role of artificial intelligence in tax with contributing editor at Tax Notes Federal Benjamin Willis.
Here are some highlights…
On machine learning and tax law
When we talk about machine learning and artificial intelligence of the law, what we’re doing is talking about collecting the raw materials, the rulings, the cases, the legislation, the regs, all that information, and bringing it to bear on a particular problem. We’re synthesizing all of those materials to make a prediction about how a new situation would likely be decided by the courts. . . . Law should be predictable. We have lots of data out there in the form of rulings, in the form of judgments that we can collect as good examples of how the courts have decided these matters in the past. And we can reverse engineer using machine learning methods how the courts are mapping the facts of different situations into outcomes. And we can do this in a really elegant way that leads to high quality prediction. So predictions of 90 percent or better accuracy about how the courts are going to make those decisions in the future, which is incredibly valuable to taxpayers to tax administration and to anyone who’s looking for certainty, predictability and fairness, in the application of law.
On the availability of artificial intelligence
We’re doing a lot to make this technology available throughout industry. Law firms are increasingly seeing this as one of the tools that they need to have in order to practice tax as effectively as possible. Academic programs … see using this kind of technology [as] a huge boost for their graduates who are going to go into practice being familiar already with the leading tools for how to leverage machine learning and artificial intelligence. Accounting firms are also quite interested in this approach too because it has huge implications in terms of speeding up research [and] doing quality assurance . . .
On the moldability of results
You can play with different dimensions. You can swap out that assumption of fact, swap in a different assumption of fact, and see how that’s likely to influence the results. So, then you can do scenario testing to really get comfortable with how much risk there is in a particular situation as the one providing a new opinion or providing advice to a client. That’s really reassuring. You might say, “Okay, I need to get this to 80 percent probability. I’m not willing to bite off more than that” . . . Or you might be like, “Well, I have a really risk-loving client. I just need to get to 51 percent” . . . [machine learning] allows you to really calibrate the amount of risk that you’re taking on, depending on the risk appetite of the client and your comfort as the practitioner.
On artificial intelligence and the courts
[Machine learning] is… a great tool to encourage settlement between the parties, and so I think we’re increasingly seeing that phenomenon where the party with the really strong position is using this to support their argument. They say, “Don’t take our word for it. We ran it on this independent system. . . Here’s the report from the system saying that we have a 95 percent or better chance of winning this case. Are you still sure you don’t want to enter into terms of settlement?” That’s often very convincing to the other side, who then run their analysis through the same system and they say, “Okay . . . It’s not nearly as strong as we thought it might be. Maybe we should talk about settling this” and that saves judges from having to contend with cases that really aren’t the best use of their time because it’s pretty clear how those cases should get decided.
On artificial intelligence and low-income taxpayers
There are early adopters at these low income taxpayer clinics across the country who are interested in using technology to allow them to give faster advice to the low income taxpayers . . . Folks understand increasingly how to use the software and how it can materially assist their clientele and so the goal is to learn from those early adopters and to figure out how to position the software to help as much as possible in other clinics where maybe we don’t have early adopters present, but who could still genuinely, really benefit from this.