Monday, November 27, 2023

How do I get started using AI in Disability Management?

 

“Ai won’t replace people… People who use AI will replace people who don’t.”  [* IBM, “Augmented work for an automated, AI-driven world”, https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/augmented-workforce  accessed Oct 5, 2023]


IBM’s assessment applies to Disability Management professionals, Rehabilitation Consultants and Personal Injury Advisors as well as clinicians, administrators, and even CEOs; those who use artificial intelligence (AI) will replace those who don’t. 


When that realization sinks in, students and participants in my workers’ compensation courses and disability management seminars often ask, “How do I start using AI?” 


Here’s what I tell them.



Step 1:  Get access to an AI chat application.

While ChatGPT is the highest profile, other popular AI chat applications like Bard, Bing’s Chat and Perplexity are available for free access in most countries (or through VPN access) subject to registration.  Subscription-based and more specialized AI applications are also available.  If your organization already subscribes to one, you may want to get started with that.


Step 2:  Create a prompt.

Chatbots respond to you.  Unlike an internet search engine, AI chat applications apply Large Language Models to what you “prompt” (the question you write or speak as an input to initiate a chat session.  One person described their first experiences with AI chats this way: “It’s like texting a detailed question to a really knowledgeable expert and getting a really complete response… but with no time lag.”


Step 3:  Interact with the application.

In an internet query, you typically ask one question then investigate by following supplied links from the response.  With AI, you can carry on a conversation; most chat applications allow you multiple (five to thirty or so) follow-up prompts, questions, or comments to which the AI chat will respond  Subsequent prompts may ask for more detail or a justification of a particular response.  You can probe, disagree, or ask for a response to be put another way.  The AI chat applications are pretty good at following the chain of conversation without restatement information already appearing in the session.


Step 4:  Review responses critically and with caution

If you are already knowledgeable about something, start with a prompt that will allow you to evaluate the responses.  AI chat applications are not authorities; they just present information in a way that sounds definitive.  In their present state of development, they make errors, hallucinate, and miss key points while emphasizing sometimes trivial matters.  What you bring to the interaction is your knowledge and judgement.  The experience of detecting and exploring errors, omissions, and overstatements will help you evaluate responses when you are not expert.


Step 5:  Verify

Most chat applications will now provide sources to support their responses either directly or if prompted.  This is important for any work you do at least at this stage in AI development. Check the references provided very carefully.  Read the papers cited to see if they actually support the statements made in the AI response. 


Practice with AI applications will help you develop your prompts to be more effective.  Here are some examples of how AI can inform and assist disability management professionals, stakeholder and advocates in their practice.


Use AI to understand appeal case decisions.

The Appeal Commission for Alberta Workers’ Compensation publishes “Significant Decisions”  on commonly appealed issues in workers’ compensation [see https://www.appealscommission.ab.ca/resources/significant-decisions] .  The selected cases are particularly clear in their organization, analysis and application of principles.  The decisions are listed by topic but not summarized or annotated.  For example, the Appeal Commission’s website lists two decisions by number under the heading “Claim Acceptance” [see Decision No. 2014-1090 , Decision No. 2014-0657].

In this example, I took the text of the two listed decisions and uploaded them to ChatGPT with the following result: 

Prompt: 

Based on the two exhibits, prepare a short summary of about 250 words for each case then provide several paragraphs outlining the principles, analysis and approach that advocates can use to inform their submissions on the acceptance of workers' compensation claims.

ChatGPT response:

Case Summary 1: Left Shoulder Labral Tear

 

In this case, a worker filed a compensation claim for a left shoulder labral tear. After examination, the first medical consultant indicated that the worker's job duties, which included frequent work in an "impingement position," likely contributed to the injury. The third orthopaedic surgeon, however, deemed the examination invalid and could not correlate the symptoms with the described injury mechanisms. The Appeals Commission ultimately sided with the first medical consultant, citing his detailed review of the worker's job demands and physical analyses, which the third orthopaedic surgeon failed to do. They also noted that the third surgeon's opinion lacked a clear rationale.

 

The Commission emphasized the absence of evidence supporting an alternate cause for the diagnosed labral tear, ruling out other potential factors. Consequently, the worker's left shoulder labral tear was deemed compensable due to the consistent medical evidence and job-related factors.

 

Case Summary 2: Left Medial Epicondylitis

 

The worker in this case was diagnosed with left medial epicondylitis. Although medical evidence confirmed the condition, it did not sufficiently establish a causal relationship with the employment hazard. The medical consultant suggested that the worker's significant exposure to force, repetition, and abnormal postures in forearm-related activities could reasonably relate to his job activities. However, the Commission found this insufficient to establish causation, as the consultant also noted that the worker reported elbow symptoms after ceasing work.

 

Additionally, the third orthopedic surgeon's examination did not support a causal connection between the worker's duties and the medial epicondylitis. The Commission emphasized that absence of evidence for an alternate cause did not automatically establish a causal link. Therefore, the claim for left medial epicondylitis was not accepted.

Principles, Analysis, and Approach for Advocates

Detailed Job Analysis: Advocates should emphasize the importance of conducting a comprehensive review of job demands and physical analyses when assessing the causal relationship between a worker's duties and their injury.

 

Clear Rationale for Opinions: Medical consultants and experts should provide a clear and detailed rationale for their opinions, ensuring that they explain the basis for their conclusions.

 

Consideration of Timing: Advocates should be aware of the temporal relationship between the reported injury and the cessation of work. This can be a critical factor in establishing causation.

 

Evaluation of Alternate Causes: While absence of evidence for an alternate cause is relevant, it alone is not sufficient to establish a causal link. Advocates should focus on gathering evidence that directly links the injury to the employment hazard.

 

Specialized Expertise: Advocates should consider the qualifications and expertise of medical consultants, especially in comparison to other examining professionals, to determine the weight given to their opinions.

 

These principles, rooted in the Appeals Commission's decisions, can guide advocates in constructing strong submissions for workers' compensation claims, ensuring they are well-supported and likely to be accepted.

 

Note that the result is highly dependent on the prompt and the AI model used.  Even replicating the above example in the same AI application will likely result in a similar but not identical response.  Changing or omitting the word count will also change the detail in the response.  If you have a decision letter on a similar claim, you can test if the principles identified in the first response are properly applied in a subsequent case.

 

Develop a job modifications and adaptations to accommodate a return-to-work  objective

In the following example, the prompt includes the instruction to assume a particular role in developing a GRTW plan.  This is a hypothetical example and for confidentiality and privacy reasons, actual names and circumstances should be avoided as you experiment with AI.

In this example, note that the prompt prescribes a point of view, an assumption of a professional position.  Note also the details contained in the prompt.


Prompt:

Assume you are a vocational rehabilitation consultant assisting a 40-year-old bank employee return to work. The employee was pursuing a hobby involving woodworking. Having removed the guard on the table saw to make a close cut, the employee amputated the ring and middle fingers on both hands below the medial pharyngeal joint. The employee is recovering well enough (after three months away) to return to work. The job duties involve a lot of sedentary work with extensive keyboarding using a standard desktop computer, keyboard, and mouse. Phone calls are made and received through a standard desktop phone with a headset. The employee has full use of thumbs, third and little finger on both hands. What modifications, devices, and adaptations to the workstation would you recommend to accommodate this employee's

ChatGPT Response:

Given the employee's condition, I would recommend the following modifications and accommodations:

 

Ergonomic Workspace: Ensure that the workstation is ergonomically designed to reduce strain on the remaining fingers and thumbs. This includes proper chair height, monitor placement, and keyboard/mouse positioning.

 

Specialized Keyboard and Mouse: Consider providing a keyboard and mouse designed for individuals with limited dexterity. There are options available with larger keys or alternative input methods like touchpads.

 

Voice Recognition Software: Implement voice recognition software to reduce reliance on extensive keyboarding. This would allow the employee to dictate text and commands.

 

Keyboard Shortcuts and Macros: Set up customized keyboard shortcuts and macros to streamline tasks and minimize the need for complex typing.

 

Adjustable Desk: Provide a desk with adjustable height to accommodate any specific needs related to the employee's comfort and posture.

 

Phone with Speaker Functionality: Use a desktop phone with speaker functionality to allow the employee to use it without needing to hold the receiver.

 

Headset with One-Sided Earpiece: Opt for a headset with a one-sided earpiece to allow the employee to use the phone while still having a free hand for typing.

 

Document Holder: Provide a document holder to keep necessary documents at eye level and reduce strain when referencing them while typing.

 

Wrist Rests and Palm Supports: Include wrist rests and palm supports to provide additional comfort and stability during keyboarding tasks.

 

Monitor Mounts: Use adjustable monitor mounts to position screens at an optimal viewing angle for the employee.

 

Training and Support: Offer training on any new assistive technologies or workstation adaptations to ensure the employee is comfortable and proficient in their use.

 

Regular Check-Ins: Schedule regular follow-ups to assess the effectiveness of the accommodations and make any necessary adjustments.

 

Remember to consult with the employee throughout this process to understand their specific needs and preferences. Additionally, consider involving an occupational therapist to provide specialized guidance if needed.

 

This is a detailed outline but a bit short on details.  Follow-up questions can ask for additional information.  You could, for example, expand on the alternative keyboards items.  In my follow up, the idea of a hand-held keyboard about the size of a smartphone allowing for typing with the thumbs similar to texting was suggested. 

 

Draft a return-to-work plan

For this example, I gave the employee in the hypothetical case a name and relationships and asked the AI chat to prepare a return-to-work plan with the following prompt. 

Prompt:

Assume the employee, Arlene W., is cleared to gradually return to work in two weeks.  The employer has agreed to the points above.  Develop a formal return-to-work plan outlining what happens each week for up to 8 weeks with the employee back at full duties at the end of the plan.  The plan will be overseen by the employee’s supervisor, Jay Q., and the RTW coordinator, Peter L.  Any adjustments to the plan will be agreed upon by all three parties.  The plan should be formatted as an agreement using the calendar weeks and dates that would apply.


I’ve posted the ChatGPT response as an image and a PDF (some formatting added). 



The product is well organized but sparse in detail.  That is understandable given the brevity of the prompt.  Adding more details could result in a more detailed. In practice, the output could be used as a starting point.


As an additional exercise, paste an anonymized version of on of our own RTW plans into an AI chat with a prompt as follows:  Review the following return to work plan commenting on its completeness, pointing out any deficiencies and make recommendations for amendments, deletions, or other changes. 


This exercise is framed as a self-check, something you might do in your own practice, but that is too narrow a perspective.  You can be certain that many of those receiving your RTW plan proposal will run a similar exercise as part of their own due diligence before signing off on your proposal.


Recommendations

I’ve stressed the need to be knowledgeable about AI but to be cautious in its implementation.  The only way you can develop your understanding of AI is by using it! 


And, keep in mind these  four recommendations, one hint, and a final caution:

  •  Start interacting with AI tools now to become familiar with their evolving utility.
  • Be sure to adhere to all ethical, confidential, privacy and academic integrity guidelines.
  •  Practice being critical of AI responses. (Hint:  Practice by creating AI prompts that will generate responses in your area of expertise and examine the response for errors, omissions, and misinterpretations or outdated perspectives).
  • Verify information before using AI information (AI is NOT an authority even if its responses are framed as authoritative assertions).

 

Remember, AI chat applications are known to hallucinate, make up quotes, and attribute them to articles that may not exist.  If you catch AI making an error or flawed reasoning, follow up with the AI application.  It is very common for AI chatbots to apologize, acknowledge their errors, and amend their responses—but only if you pursue the point. 


AI is here to stay, being integrated into existing technologies,  and continuing to evolve.  Use it to make a difference in your profession.

No comments: