Seeing Trojan Horses in Lightening with the Eyes Wide Shut: A Reflection on Prof. Catherine Rogers Post

Kluwer Arbitration Blog
December 10, 2020

Please refer to this post as: , ‘Seeing Trojan Horses in Lightening with the Eyes Wide Shut: A Reflection on Prof. Catherine Rogers Post’, Kluwer Arbitration Blog, December 10 2020,

Let us be clear, the lightening’s spirit is out of the bottle and here to stay. It is neither possible nor desirable to prevent party counsel from using tools that increase the efficiency of party representation. Prof. Rogers and her co-authors provide a correct general description of AI-based information systems on decision makers. My firm uses such a platform for analyzing possible outcomes in proceedings before the EPO opposition divisions and boards of appeal. The proceedings are always subject to the same procedural and substantive rules and the issue categories are stable. A relatively small group of persons forms the decision makers. In addition to somehow unsuspicious data, such as duration of the proceedings or win/loss rates, the system provides deep insight into the types of arguments that are more likely to persuade a specific decision maker. This statistical information is certainly helpful for preparing written and oral arguments. Therefore, I have no difficulty to imagine that in the field of treaty based investment arbitration such systems provide the same level of insight and statistical predictability.

In international commercial arbitration, the panorama is different. Substantive and procedural rules vary from case to case. The group of decision makers is much more varied and instable. There are more and varying cultural factors that affect the case, such as inter alia language and professional training. Transporting an AI platform trained in a more stable environment such as investment arbitration or the US legal system into this arena for crunching information may provide problematic statistical information, which in the worst case can be biased and mislead without any user being able to apprehend or understand. There will be a confirmation bias that a system trusted by others can be always trusted.

Of course, in a scenario where everybody may choose between many AI platforms offering essentially identical services one could expect an auto correction effect over time, because users could compare their experiences and results. However, the substantial resources required for the development, deployment and constant improvement of such AI platforms, the somehow limited number of potential customers (restricted market) and a comparison with other core platform services in the Internet seem to indicate with a high degree of likelihood, that one may end up with not more than a handful of such AI platforms, which among them cover the bigger part of the market. This could lead to interesting situations, if all parties use the same platform for case outcome prediction and the preparation of their submissions.

Considering the ever increasing ability of AI to crunch ubiquitous unstructured data as long as it is machine readable (these include digital photographs and voice recordings, not only digital documents) and the advancement in automated semantic analysis, it is easy to foresee that AI based predictive analysis may extend its scope beyond the obvious (directly case related information) to other more human factors driving the decision makers. Think of attitudes and inclinations of a more personal character. The AI systems may provide users with statistical information about decision makers, which these – sometimes for good reasons – ignore.

It is interesting to note that the most likely users of these AI-platform services, the law firms, have not yet voiced a strong demand for being offered similar predictive services covering their potential opponent counsel. Prof. Rogers and her co-authors stress the high degree of control by counsel, i.e. their determinative role, in arbitration proceedings. Intuition would prompt the question, whether this does not result in a bigger influence on duration, cost, and result, than arbitrator behavior. At least, predictive counsel AI could also help to make crucial choices concerning party representation, at least from a party perspective. Perhaps because they see no market, the information aggregators have apparently not yet dedicated resources to this market segment, because their potential clients are the very same law firms and not the clients, who mostly have an arbitration only occasionally.

Presently not much is asked and little is known about the inner workings of the information gatherers’ AI platforms that are probably still in their infant stage. This is worrying.

I think that:

  • Any person who is an arbitrator and for whom information or information processing results are stored should know this and have a right to inspect such information. This right would be hollow if this person were not allowed to gain access to generated reports for users insofar as she is concerned and have the right to correction of false data. Whilst these rights are statutory under the EU GDPR, they may not be taken for granted on a global level.
  • Any AI platform owner should be transparent about ownership, control structures, affiliations, and used sub-contractors. They should also be transparent about the algorithms underlying their AI and the measures taken to monitor and avoid unreliable or biased results. Admittedly, transparency needs to be balanced against the legitimate need of protecting proprietary know-how.
  • There should be ethical standards and accountability for the AI platform owners and/or service providers, because after all they provide services in a field that – even if privately administered – is a public good: Justice and the Law.

Platforms for arbitrator intelligence also claim that these platforms will provide more arbitrator market transparence, more opportunities for newcomers or underrepresented groups, and more objective information than the present opaque arbitrator selection approaches. This may well be the case. However, if these platforms enable better arbitrator selection using AI algorithms, one may suspect that those candidates on which more machine readable and accessible information is available, will have a systemic advantage, if this is the right information. This may not privilege newcomers and is critical, if the algorithm is inadvertently biased. Here new business opportunities loom for consultants who help prospective arbitrators getting their accessible information profile right in the same way consultants help companies to move up the Google display lists.

Those among us, with more philosophical or social science inclinations, may take their analysis of AI driven arbitrator intelligence a step further. They may ask what it means for our traditional concept of law, if those who have to find and apply it, are inherently conceived as rickety machines that cannot be repaired but must be reined in by taking preemptive measures based on historical information and statistical methodology applied by unaccountable self-learning algorithms. We can also ask whether we like to see ourselves as rickety machines – probably not a very attractive content of our self-narrative, but true.

Finally, corporations and other clients may ask whether automated outcome prediction is not sufficient to meet their needs. After all, business is used on a daily basis to take decisions based on outcome probabilities. To them the resources required to gain the ultimate 20% of outcome certainty may be too big, given that these are often more expensive to obtain than the other 80%. Why not use the 80% for seeking a deal that sometimes can yield more than is covered by legal claims or is preferable from a financial perspective, because less time is consumed in the process?

Too many questions? Wrong questions? Good! Don’t walk the path with your eyes wide shut. Ask your own questions.