AI and Workers' Comp - Summary of ARPA presentation

The recent Australian Association of Rehabilitation Providers (ARPA) Roundtable meeting in Perth provided an opportunity to speak on how AI is and will impact workers’ compensation.  

In my presentation, I noted that AI has been applied in research and proprietary settings for years but that the advent of low or no-barrier access to Large Language Models such as ChatGPT, Bard, Bing’s chat, Perplexity,  and others have quickened the pace of AI application and adoption. 

AI will not change the functions of any disability insurance or workers’ compensation systems.  Using Henry Mintzberg’s approach, I illustrated how AI can be applied to functions in the technostructure, support, and strategic apex of organizations, as well as through the operating core.  The impact at the interface between the organization and those it serves (stakeholders, injured workers, employers, advocates, etc.) is dynamic and bi-directional.  AI enables organizational functions and changes how others outside the structure use, analyze and react to information and decisions.    

AON, one of the world’s largest reinsurers, [see https://insights-north-america.aon.com/total-cost-of-risk/how-artificial-intelligence-is-changing-workers-compensation ]  and  Clara Analytics [see https://claraanalytics.com/blog/six-ways-to-reduce-workers-comp-costs-using-ai/ ].have reported current application of AI in the insurance and workers’ compensation sector.  Applications that include:

• Analyzing workers’ compensation wealth of structured and unstructured data
• Sorting out the impact of events like COVID-19
• Loss control benefits for employers
• Detect anomalies in procedures and drug costs
• Evaluate, decide, and pay 60-70% of straightforward claims
• Identifying high risk and complex cases early and direct to specialist staff
• Enable early intervention.

I demonstrated how a complex 43-page appeal-level decision could be uploaded to a chatbot with a prompt to produce a 500 word, plain-language summary along with the strengths and weaknesses of the decision.  The near instantaneous output matched the quality of my own summary and analysis that took more than half a day to prepare. 

In another example, I showed how AI had been trained to analyze and code a few thousand workers’ compensation cases, apply that learning to a dataset of 1.2 million cases and produce actionable analytics for prevention of injuries in specific sectors. 

A recent IBM report [see https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/augmented-workforce noted “AI will not replace people – but people who use AI will replace people who don’t.”  Putting a finer point on this, I noted:

• AI won’t replace clinicians but clinicians who use AI will replace clinicians who don’t.
• AI won’t replace rehabilitation and return to work specialists but rehabilitation specialists  who use AI will replace rehab professionals who don’t.
• AI won’t replace CEOs but CEOs who use AI will replace CEOs who don’t.

My final points reiterated that AI is here, it is not going away, and it is going to increase in power.  Ignoring and fearing it are not really options.  Understanding AI, how to use it, and its limitations is essential.

Workers’ compensation or disability insurance are not islands.  As clients, injured workers, regulators, advocates, and stakeholders adopt AI and experience how sectors and services using this technology to improve quality and timeliness, they will demand improvements in these sectors as well.   Even if you are meeting or exceeding your service, timeliness, and quality standards now, by the measure of the AI-enabled world, you are falling behind.  Your reputation and social capital will suffer as a result. 

The discussion following the presentation included insightful questions and observations.  One key insight was the importance of crafting the prompts used with large language models.  The richness and accuracy of the prompts will be somewhat determinant of the product of an AI interaction. 

Also covered in the discussion was the importance of the professionalism, knowledge, and experience users bring to AI interactions.  At this point in AI implementation, those human insights and judgements are essential.  In one of the presentation examples, the AI chatbot was prompted to read a case study from the ARPA website and develop a specific multi-faceted rehabilitation plan for a complex case.  The output accurately reflected what the current body of disability management and rehabilitation management knowledge would ideally recommend.  The realities and constraints of implementing the plan, adapting the ideal to the specific context, and “doing” the work required professional human judgement to achiever the real-world objective of a safe, durable return-to-work outcome.

Another comment built on the cautions raised in the presentation regarding privacy, confidentiality, and regulatory issues. While all the examples used in my presentation were from publicly available sources, professional judgement and rigorous compliance with regulatory and professional standards will complicate implementation of AI in organizations.

The call to action for this presentation was engagement.  It is time to educate subject matter experts in all functional areas of your organization about AI and begin the process of improving and safeguarding functions across the organization. 

AI will improve workers’ compensation, disability management, personal injury insurance, and prevention. Developing your AI strategy is not just something to put on your to do list, it is a current imperative.