The mainstream advent of Artificial Intelligence (AI) and Artificial General Intelligence (AGI or AutoAI such as AutoGPT) dominates on-line and traditional news feeds with examples and extrapolations of impacts both positive and negative.
Capabilities and Cautions
Educational institutions are struggling with what to do about the sudden availability of this technology in academic settings. I tell participants in the university and professional courses I instruct that AI chatbots such as Google Bard, ChatGPT, and Microsoft Bing Chat are not prohibited within my courses; these tools are helpful in exploring new ideas in prevention, analyzing claim decisions, and generating return-to-work strategies or vocational alternatives for disabled workers. Learning to use them appropriately is an important real-world skill. Simply banning their use in academics makes little sense.
Informal discussions with executives and professional working in workers’ compensation, prevention, and personal injury insurance, reveal widespread if informal use of these tools along with anecdotal accounts of both positive and cautionary experiences. From my own testing and published reports, AI chatbots may make up data, cite non-existent sources, and even attribute direct quotes that were never authored by the attributed source. I asked several chatbots to create a table comparing specific legislated provisions of state and provincial workers’ compensation laws. None provided correct data. When corrected, the chatbots politely apologized and made the necessary corrections in an amended table. When queried in another session, the original errors and omissions returned.
AI applications in sectors and industries
The public-facing chatbots available today are illustrative of AI capabilities (despite current limitations). AI applications just entering public awareness include text-to-image and text-to-video representations that are difficult to differentiate from images captured by any digital or analogue camera. Specialized AI applications are already being marketed to many professional fields including medicine, manufacturing, finance, and legal services. Those AI applications are about to impact many industries and occupations. [For example, see Arianna Johnson, “Which Jobs Will AI Replace? These 4 Industries Will Be Heavily Impacted”, Forbes, Mar 30, 2023 and “16 Industries And Functions That Will Benefit From AI In 2022 And Beyond”, Forbes, Jan 13, 2022].
Workers’ compensation, personal injury insurance and AI
In workers’ compensation and other personal injury insurance organizations, AI-assisted tools are the next generation of resources to facilitate diagnosis, treatment and early, safe return-to-work outcomes. AI’s impacts in these fields are going to be fundamental and disruptive. Here’s what to expect in the near to medium term:
1. Most claims will be automatically adjudicated--quickly, accurately, consistently, efficiently.
AI is already being applied in personal injury insurance and will be the dominant adjudicative and processing modality for claims. Pareto optimization and segmentation will allow fewer but more highly trained, specialized resources to be focused on more complex cases to achieve better outcomes while providing rapid, accurate, consistent service for most other cases [ see Clara Analytics, “Six ways to reduce workers’ compensation costs using AI” at https://claraanalytics.com/blog/six-ways-to-reduce-workers-comp-costs-using-ai] / The human resource challenge of competing for, developing and retaining those resources will be significant.
2. Medical cost-containment will be AI driven.
Ai works well with large data sources and will more effectively track patterns, spot anomalies, and detect issues once the purview of manual audit and review processes. Machine learning and advanced analytics will eliminate the need for manual processes (such as medical bill reviews) and allow for highly focused, real-time control by fewer but more highly trained human agents. AutoAI will handle most but not all results. For the (fewer) cost-containment specialists, the intensity of work will actually increase as a consequence of AI managing much of the routine, lower complexity issues. This “adverse selection” will intensify the complexity of the residual cases.
3. Organizational structures will change significantly
Smaller organizational establishments and facilities for the human resources will be required to support the AI driven systems and highly specialized human adjudicative/claims, prevention, rehabilitation, and administrative processes. In terms of Henry Mintzberg’s organizational structure theory, AI implementations will reduce establishment size (number of staff) of the operating core and middle line components of an organization. Many (but not all) support and oversight functions will be absorbed by AI systems. This does not eliminate the need for all support functions; AI will change the scope of a smaller cohort with additional skills and responsibilities beyond traditional support functions. This will be particularly true in line and technical support areas where knowledge requirements at what was considered support or assistant levels will increase. Professionalization and upskilling in both the use of AI technologies and extra-organizational human interactions that facilitate line functions will be essential for professional and executive assistant roles. While more routine work will flow through AI systems, higher functions including many “middle management” functions such as monitoring, budgeting, contract administration, and deployment tasks will cascade directly from executive levels to AI and human executive assistant and analyst levels. Governance will need to expand or create roles for Chief Technology, Chief Data, and Chief Risk Officers.
4. Policy development and legal analysis cycles will shorten.
As noted in the introduction, AI tools are already preparing legal documents, synthesizing decisions, analyzing rulings, and drafting policies. For workers’ compensation and other insurance, AI will facilitate more rapid development and allow greater attention to communication. AI will also facilitate the “use case” analysis and financial impact modeling needed for legislative and policy consultations. I recently took the proposed WorkSafeBC guideline for G7.19(5) Exposure to non-ionizing radiation – Ultraviolet radiation [see https://www.worksafebc.com/en/resources/law-policy/discussion-papers/guidelines-preliminary-posting/g7-19-5-exposure-to-non-ionizing-radiation-ultraviolet-radiation] and asked a chatbot for a review including strengths and weaknesses. The response lauded its structure, informative nature, and up-to-date information but suggested several possible improvements in the form of expansions, using UV germicidal lights that have become more widely used for disinfection purposes in hospitals and hotels.
5. AI will detect risk and injury patterns creating opportunities for injury and disability prevention.
AI works with data. Workers’ compensation and personal injury insurance work with people, specifically their hearts, minds, attitudes, and beliefs. While AI can detect patterns of cause and suggest opportunities for better prevention of injury and disability, interventions in populations will require emotional intelligence, operational agility, and creativity mediated by capable human agents. AI will help prevention (loss prevention departments) more effectively target their primary prevention activities and initiatives. For inspectorates (regulatory agencies), AI will allow for faster action and more effective targeted compliance interventions. Disability management (DM) and vocational rehabilitation (VR) specialists will be aided by AI in their secondary and tertiary prevention roles. Upskilling prevention staff, workplace inspectors and DM/VR professionals to use AI effectively will be a challenge.
6. Appeals against decisions will become fewer but more complex.
Better initial decision-making by AI or AI-assisted adjudication will result in fewer errors in law or application of policy. This will adversely select for more complex cases involving the weighing of evidence and the exercise of discretion. I recently took the summary of a publicly available noteworthy decision from the Workers’ Compensation Appeal Tribunal of British Columbia (WCAT) and asked an AI chatbot to determine weaknesses in the decision. The analysis was instant and provided several weaknesses. While this test was based on an appellate level decision, it illustrates how AI allows initial, review, and appeal decisions to be examined to find weaknesses and facilitate grounds for further appellate action. It also suggests how AI may be applied in the decision drafting stage to identify weaknesses that may be addressed before the final decision is released (weaknesses addressed or decision reconsidered). If this improves decision making, the result will be improved reasoning in the ultimate decision and potentially fewer appeals.
7. AI supported fraud and cyber attacks will increase and become more difficult to detect.
AI can craft stories that are pure fiction from a few facts and suggestions. It can model scenarios that are indistinguishable from those insurers receive to insure and compensate employers, workers, and others. With graphical AI applications, even photographic and video evidence may be misleading. AI can find vulnerabilities in systems and codes and create scenarios that may open workers’ compensation and personal injury insurers to fraud. This issue is not limited to claims or supplier attempts to exploit systems. This threat includes internal fraud and external exploits on financial and other systems.
8. AI will be visible… until it is not
At the moment, the public awareness of AI and its potentials is heightened. Insurers in workers’ comp and personal injury will be under scrutiny for how they implement AI technology. Public filings and disclosure rules may even be imposed. This demand for transparency will be significant, persistent, but will fade over time. Just as the public-facing ChatBots will fade away as a novelty and “AI enhanced search” functions will simply be considered integral to the “search” function, AI will be seamless incorporated across all components of the insurance value chain. No one will demand transparency over the AI that controls the traffic flow on our streets or optimizes the energy use in our homes; the same will be true for most personal injury and workers’ compensation claims decisions, it will become just part of environment in which we live and operate.
9. AI will become trusted despite making (a few) spectacular errors
AI is improving at astounding rates. Already, AI has reportedly passed business, law and medical exams. That does not mean to say I’m ready to trust my medical care or the drafting of my contracts to a chatbot. On the other hand, I am reaching the point where I may be less assured by a physician or lawyer who refuses to use AI in their work. AI will increasingly be trusted despite errors. As those errors decline, that trust will become implicit… mostly. Widely reported errors in any AI application in any industry will impact trust in all sectors using the technology.
10. AI will not eliminate bias
AI works from existing data. Most language-based AI systems will reflect the bias inherent the language and data they access. There is a risk that AI implementations will be bias towards the weight of data available historically, making it more difficult for newer, more accurate data to be reported and presented in AI enabled systems with the appropriate weight or prominence. For example, AI can’t correct the inherent lack of female representation in past medical studies or correct gender bias that may lead to underestimation of impacts on women. I warn my students to not consider AI chatbot outputs as authorities or their responses as definitive. While the AI chatbots I’ve tested will admit errors, omissions, they only do so when challenged. While factual errors may be reduced, AI will continue to reflect bias—and that will be hard to identify.
Closing thoughts
AI has changed the operating and threat environment for workers’ compensation and personal injury insurance. Its impacts can be perilous for the sector. Workers’ compensation and personal injury insurance organizations need explicit and comprehensive strategies in place now to manage the threats and capitalize on the opportunities created by the present and rapidly evolving AI technology landscape. Yes, the impacts will be disruptive and varied. Failing to act now may hurt jeopardize those these systems were designed to protect.
No comments:
Post a Comment