Select Page

When the American Arbitration Association (AAA) recently announced that it would be launching an AI-powered arbitrator in November, it raised the question of the future role of AI in litigation. Indeed, it could suggest a possible future that many litigators still insist will never arrive.

I often give presentations on the use of AI in litigation and the impact it could and will have. I frequently hear from older litigators that they aren’t all that concerned about what AI could do to their practices. After all, they reason, litigators have to effectively persuade other humans. They need to have empathy, sympathy, and assess body language and subtleties in others. And they have to have the proverbial gut instinct. None of these things does AI have. Yet.

That may be true, I say. But have you considered the possibility that, in the future, the decision maker is, itself, an AI tool? How necessary will litigators be when all the relevant information is fed into a bot which then makes a decision? What will the litigator’s job be? How realistic is this?

The AAA Announcement

And lest we think that AI decision making is far-fetched, eBay has been using an AI bot to resolve disputes between buyers and sellers for some time. Then came the AAA announcement that it would be launching its AI-powered arbitrator in November. The AI arbitrator will, for now, be deciding documents-only construction defect cases, although in the future, according to AAA, it may be used for insurance cases and specifically high-volume but low-dollar-amount payer provider disputes. 

In an interview on Bob Ambrogi’s podcast, Bridget McCormack, AAA’s president and CEO, claimed that use of the tool would reduce the cost of construction cases by some 30-50% and the time required to litigate and resolve a case by 25-35%. She expects improvement over time.

It’s All About Cost v. Exposure

It’s those metrics that stand out. Particularly for arbitration but for all litigation, cost and time are critical. Lots of disputes go unresolved because of these two factors. And businesses and insurance companies would tell you that the transactional costs of litigation are substantial. 

In thinking about whether AI decision making in litigation is realistic, think about the following: I was talking to a general counsel recently about AI and its impact. I asked her if she were given the option of having an AI tool decide a case without so much cost would she agree? Her answer, even a few months ago, was “Absolutely. If I could refer any case where the amount at stake was less than, say, $50k, I would do it in a heartbeat.”

Why? It’s because she was spending more in legal fees and transactional costs for those low exposure cases than what they were worth. So even if the AI bot might get a few cases wrong or achieve a result worse than what a human lawyer might achieve, it doesn’t matter all that much in the long run. It’s why insurance companies are willing to pay lawyers with low hourly rates: the difference between an A job and a C job doesn’t affect the overall result that much. So why pay any more in legal fees than you have to?

Cases Ripe for AI Decision Making

If that’s the case, there are certain kinds of cases that might be ideal for this kind of decision making.

I talked recently with Sarannah McMurtry, Executive Vice President and General Counsel of First Acceptance Insurance Company. First Acceptance Insurance Company provides nonstandard auto insurance and specializes in coverage for high-risk drivers who may not qualify for traditional policies. 

First Acceptance is in the business of claims that are often lower exposure, the kinds of cases previously mentioned by the GC that could be ripe for AI decision making. These are cases where the cost of litigating the cases could easily outweigh the exposure. Perhaps not surprisingly then, McMurtry told me that AI is “going to revolutionize the insurance business from rate, claims, intake.” 

McMurtry agreed that there are certain types of claims that would be better candidates for some portion of AI review and decision making. Claims with estimates and photos and other documentation of property damage that could be examined by AI for initial decision, for example. AI could also help determine the claims that could go straight through for payment, saving time and cost.

And another key area that might be ripe for AI decision making is insurance subrogation. For those unfamiliar, subrogation claims occur when one carrier pays a claim and then seeks recovery from some other entity, often, in the automobile context, another insurance carrier. 

For those claims, AI decision making may make sense. According to McMurtry “where you have a defined submission process, having those claims decided by AI makes sense. For one thing it’s cost effective. It allows your people to do other things. And you’re not impacting claimants. It’s just simply a transaction between the two insurance companies to allocate that risk appropriately.”

Some Road Blocks

But there are roadblocks. For insurance companies like First Acceptance, the biggest roadblock is the specter of bad faith. Insurance companies have a duty to deal with policy holders in good faith. A breach of that duty can turn a minor claim into one that may result in a catastrophic nuclear verdict since the damages far exceed the policy policy limits. McMurtry explains: “We’re very very cautious about where we want to use something like AI or insert a tool that would not be human reviewed. A tool that helps with the initial evaluation is valuable but there still must be a significant human touch in the process.”

She explained that if an AI tool approved a pay out quickly, great but if it denied a claim, that would be much tougher.

And Then There Is That Bias Thing

I also discussed the bias problem with McMurtry. The problem, she says, is that the data going into AI models often comes from humans with their own bias. So, the models will always have some bias. She agreed the trick will be getting the AI decision maker to a level of acceptable bias, keeping in mind human decision makers also have bias. 

Indeed, many of our procedural and evidentiary safeguards in litigation are designed to minimize human bias. We will have to figure out what kinds of guardrails need to be in place to reduce bias to that acceptable level if AI decision making is to be used, and in what contexts.

Other Open Questions

As with the use of any AI tool, particularly in dispute resolution. there remain open questions:

• How do we correct errors and allow for appeal?

• What about transparency and explainability?

• What should the regulatory and ethical frameworks be?

• Who bears liability for AI mistakes?

Where Are We?

Going back to the AAA announcement, it’s important to remember that particularly with businesses, arbitration is an agreed to dispute resolution technique. Indeed, I recently wrote about a tool from Arbitrus.ai. The tool is essentially an AI decision maker: where the parties agree, Arbitrus.ai can be used to resolve any disputes arising out of the contract.

And that’s the key issue at least for now. Where the parties agree that a dispute or disputes can be resolved by AI, great. It makes sense from a cost and time perspective. But where they don’t, there’s no way we can use an AI decision maker. 

It’s much like the right to a jury trial: the parties can agree to waive their right to trial by jury but can’t be forced to. The danger is that AI decision making might be forced by contract to those that don’t want it but have little bargaining power. We have seen this often where large companies attempt to force arbitration by contract terms.

It Depends

So, yes, AI dispute resolution may hold promise in litigation, whether it will, depends. It can’t be forced on unwilling parties. It makes the most sense for low-exposure disputes, particularly between businesses with equal bargaining power. 

But like everything with AI, we need guardrails. For now, consent must remain the cornerstone. We must ensure that consent is truly voluntary, not coerced through adhesion contracts that leave consumers with no real choice.


Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.

The post The AI Arbitrator Is Here: What’s Next? appeared first on Above the Law.

When the American Arbitration Association (AAA) recently announced that it would be launching an AI-powered arbitrator in November, it raised the question of the future role of AI in litigation. Indeed, it could suggest a possible future that many litigators still insist will never arrive.

I often give presentations on the use of AI in litigation and the impact it could and will have. I frequently hear from older litigators that they aren’t all that concerned about what AI could do to their practices. After all, they reason, litigators have to effectively persuade other humans. They need to have empathy, sympathy, and assess body language and subtleties in others. And they have to have the proverbial gut instinct. None of these things does AI have. Yet.

That may be true, I say. But have you considered the possibility that, in the future, the decision maker is, itself, an AI tool? How necessary will litigators be when all the relevant information is fed into a bot which then makes a decision? What will the litigator’s job be? How realistic is this?

The AAA Announcement

And lest we think that AI decision making is far-fetched, eBay has been using an AI bot to resolve disputes between buyers and sellers for some time. Then came the AAA announcement that it would be launching its AI-powered arbitrator in November. The AI arbitrator will, for now, be deciding documents-only construction defect cases, although in the future, according to AAA, it may be used for insurance cases and specifically high-volume but low-dollar-amount payer provider disputes. 

In an interview on Bob Ambrogi’s podcast, Bridget McCormack, AAA’s president and CEO, claimed that use of the tool would reduce the cost of construction cases by some 30-50% and the time required to litigate and resolve a case by 25-35%. She expects improvement over time.

It’s All About Cost v. Exposure

It’s those metrics that stand out. Particularly for arbitration but for all litigation, cost and time are critical. Lots of disputes go unresolved because of these two factors. And businesses and insurance companies would tell you that the transactional costs of litigation are substantial. 

In thinking about whether AI decision making in litigation is realistic, think about the following: I was talking to a general counsel recently about AI and its impact. I asked her if she were given the option of having an AI tool decide a case without so much cost would she agree? Her answer, even a few months ago, was “Absolutely. If I could refer any case where the amount at stake was less than, say, $50k, I would do it in a heartbeat.”

Why? It’s because she was spending more in legal fees and transactional costs for those low exposure cases than what they were worth. So even if the AI bot might get a few cases wrong or achieve a result worse than what a human lawyer might achieve, it doesn’t matter all that much in the long run. It’s why insurance companies are willing to pay lawyers with low hourly rates: the difference between an A job and a C job doesn’t affect the overall result that much. So why pay any more in legal fees than you have to?

Cases Ripe for AI Decision Making

If that’s the case, there are certain kinds of cases that might be ideal for this kind of decision making.

I talked recently with Sarannah McMurtry, Executive Vice President and General Counsel of First Acceptance Insurance Company. First Acceptance Insurance Company provides nonstandard auto insurance and specializes in coverage for high-risk drivers who may not qualify for traditional policies. 

First Acceptance is in the business of claims that are often lower exposure, the kinds of cases previously mentioned by the GC that could be ripe for AI decision making. These are cases where the cost of litigating the cases could easily outweigh the exposure. Perhaps not surprisingly then, McMurtry told me that AI is “going to revolutionize the insurance business from rate, claims, intake.” 

McMurtry agreed that there are certain types of claims that would be better candidates for some portion of AI review and decision making. Claims with estimates and photos and other documentation of property damage that could be examined by AI for initial decision, for example. AI could also help determine the claims that could go straight through for payment, saving time and cost.

And another key area that might be ripe for AI decision making is insurance subrogation. For those unfamiliar, subrogation claims occur when one carrier pays a claim and then seeks recovery from some other entity, often, in the automobile context, another insurance carrier. 

For those claims, AI decision making may make sense. According to McMurtry “where you have a defined submission process, having those claims decided by AI makes sense. For one thing it’s cost effective. It allows your people to do other things. And you’re not impacting claimants. It’s just simply a transaction between the two insurance companies to allocate that risk appropriately.”

Some Road Blocks

But there are roadblocks. For insurance companies like First Acceptance, the biggest roadblock is the specter of bad faith. Insurance companies have a duty to deal with policy holders in good faith. A breach of that duty can turn a minor claim into one that may result in a catastrophic nuclear verdict since the damages far exceed the policy policy limits. McMurtry explains: “We’re very very cautious about where we want to use something like AI or insert a tool that would not be human reviewed. A tool that helps with the initial evaluation is valuable but there still must be a significant human touch in the process.”

She explained that if an AI tool approved a pay out quickly, great but if it denied a claim, that would be much tougher.

And Then There Is That Bias Thing

I also discussed the bias problem with McMurtry. The problem, she says, is that the data going into AI models often comes from humans with their own bias. So, the models will always have some bias. She agreed the trick will be getting the AI decision maker to a level of acceptable bias, keeping in mind human decision makers also have bias. 

Indeed, many of our procedural and evidentiary safeguards in litigation are designed to minimize human bias. We will have to figure out what kinds of guardrails need to be in place to reduce bias to that acceptable level if AI decision making is to be used, and in what contexts.

Other Open Questions

As with the use of any AI tool, particularly in dispute resolution. there remain open questions:

• How do we correct errors and allow for appeal?

• What about transparency and explainability?

• What should the regulatory and ethical frameworks be?

• Who bears liability for AI mistakes?

Where Are We?

Going back to the AAA announcement, it’s important to remember that particularly with businesses, arbitration is an agreed to dispute resolution technique. Indeed, I recently wrote about a tool from Arbitrus.ai. The tool is essentially an AI decision maker: where the parties agree, Arbitrus.ai can be used to resolve any disputes arising out of the contract.

And that’s the key issue at least for now. Where the parties agree that a dispute or disputes can be resolved by AI, great. It makes sense from a cost and time perspective. But where they don’t, there’s no way we can use an AI decision maker. 

It’s much like the right to a jury trial: the parties can agree to waive their right to trial by jury but can’t be forced to. The danger is that AI decision making might be forced by contract to those that don’t want it but have little bargaining power. We have seen this often where large companies attempt to force arbitration by contract terms.

It Depends

So, yes, AI dispute resolution may hold promise in litigation, whether it will, depends. It can’t be forced on unwilling parties. It makes the most sense for low-exposure disputes, particularly between businesses with equal bargaining power. 

But like everything with AI, we need guardrails. For now, consent must remain the cornerstone. We must ensure that consent is truly voluntary, not coerced through adhesion contracts that leave consumers with no real choice.


Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.

The post The AI Arbitrator Is Here: What’s Next? appeared first on Above the Law.