From Forbes:
Previously, much has been written about the dangers posed by AI and algorithmic decision-making tools in relation to people with disabilities. ...
[B]rand new tech startup Advocate is not only planning to turn this trend on its head but to achieve this through using a sophisticated mixture of AI and human expertise to address one of the most significant historic pain points afflicting the disability community. Namely, the unbearable delays and complexities involved in navigating America’s burdensome Social Security Disability Insurance system. ...
The fledgling company has just announced $4 million in seed funding to build a technology platform that aims to not only shorten and streamline the process of applying for long-term disability benefits but also provides claimants with a clear view of the strength of their application and chances of success. ...
Advocate CEO Emilie Poteat, an accredited disability benefits advocate herself, has already drafted expert advisors onto the corporate team in the shape of former Commissioner of Social Security, Jo Anne Barnhart and the former Secretary of Veterans Affairs, David Shulkin. ...
At the front end, users of Advocate will encounter an elegant and accessible platform that assists them in optimizing their claim by connecting to, prioritizing and synthesizing electronic health records as well as providing a means of filing the paperwork itself. ...
Why is it that I have a feeling that Jo Anne Barnhart's involvement is a sign that this endeavor is destined to fail? Of course, there's also the fact that this focuses on the application process. Yes, the application process is a barrier but it pales in comparison with the other barriers in the process, such as the long backlogs and high denial rates. A $4 million AI project isn't going to solve those.
Might work if SSA used this technology on their end, to speed up initial decisions.
ReplyDeleteAI the whole process, no reps needed, no judges, no DDS. Fair and impartial.
ReplyDeleteI am 100% for it. You meet you equate or you are denied.
At 11:02, everything sounds great, except when you consider that AI will still be biased. It will just be biased in accordance with whoever programmed it. Worse yet, if an initial bias is programmed into the system (intentionally or not), then through self-learning the system will learn to magnify that bias.
ReplyDelete@10:09 The agency has been using a number of these technologies some for a while, others more recently to acquire and analyze medical records in the disability determination process. To electronically acquire medical records, the agency has had MEGAHIT in place for well over a decade. The major challenge is that Electronic Health Records (EHR) and Health IT Interoperability capabilities are still emerging in the healthcare industry. SSA has partnered with thousands of healthcare providers/facilities using the major national health IT interoperability networks (eHealth Exchange, CareQuality, CommonWell Health Alliance) and needs to be involved in the ONC TEFCA effort. SSA requires complete clinical content in order to avoid manual time-consuming and costly follow-ups. Complete electronic health information is what is holding the agency and the nation back. On the other hand, the agency has been leveraging Natural Language Processing (NLP) to identify and extract key clinical information from medical records, even from scanned images of paper/fax using OCR (image to text). All this information can be used by IMAGEN, which has used AI/ML to model out decisions to assist examiners in rapidly identifying clinical information that could support a disability claim. These capabilities help to decrease case processing time, lower overall operational cost and improve service to the public. When these technologies are used to their fullest, a claimant could even see an approval much sooner (sometimes in 24-48 hours rather than initial decisions taking months). What is really needed is more executive focus, full funding and resources to assist in the further rollout of these capabilities within and external to the agency. Agency executives do not always seem to understand that moving the entire nation forward with healthcare technologies takes time.
ReplyDeleteI don't see how AI will help. The application process is not the problem. It is the processing of the application that is the problem. Not enough people to do the job. smdh
ReplyDeleteMaybe robots can work with the walk in traffic who demand face to face interaction. See how well that goes over with the public.
ReplyDelete@11.02
ReplyDeleteNot sure if you were serious, but most approvals are not based on meeting or equaling a listing, so your solution is not possible under the current statute or regulations.
More generally, this AI start-up is a terrible idea. Anyone with a modicum of experience with dealing with SSA, State Agencies, and our broken medical system, would know its impossible for AI to effectively navigate these issues.
It simply impossible to automate the application process given that SSA does not process all claims in a uniform manner. If SSA's own back-end IT was perfect, maybe this would make sense. There are far too many manual processes that require an SSA employee to complete. I am spending upwards of 15% of my time correcting human errors, which have gotten significantly worse and more frequent the last two years.
Here is prime example--the ERE system at the NY State Agency is currently broken. Records cannot be uploaded directly via the ERE file, and the state agency stopped sending out barcodes to attorneys (and likely many unrepresented claimants). An AI system would not be able to navigate these constant errors and procedure variations between state agencies.
This honestly sounds like a total scam .What AI tool are they using? Because none of the current tools are remotely capable of "connecting to, prioritizing and synthesizing electronic health records as well as providing a means of filing the paperwork itself" And $4 million is a laughable amount to build an AI powered tech platform.
The last time I saw any involvement of JoAnn Barnhart (at least a past if not current member of the GOP) she was lobbying for the American Hospital Association to get access to some SSA data that was off limits and so required explicit permission to see that would help out hospitals under the guise of providing a benefit to SSA. I think staff kept saying no and maybe a COSS said yes? It was a while ago. But there must be a way to make money angle in all this if she's involved.
ReplyDeleteAI has a limited utlity in my opinion. There are endless combinations of impairments and vocational factors that can lead to meeting the disability standard. The medical and vocational information changes over time. Judgment is needed to detect and correct wrong information. No AI could be expected to handle all of that correctly. What an AI can conceivably do is a better job at calling attention to a limited number of the most obvious cases. That could be useful by skimming some of the most clear cut cases off the top so SSA staff can focus on the rest.
ReplyDeleteIf you already have trained staff helping people fill out forms then you don't need an AI. An AI might help trained staff save time and provide greater accuracy if it was very well done. Some ideas to help with clients who often don't remember things or provide sufficient detail: 1. AI scans available medical records for reported symptoms and signs. When filling out function report, AI prompts for whether to include them. It could also prompt for whether to include the known major symptoms and signs that could impair function for the claimant's diagnosed medical conditions. 2. When filling out the work history report, AI searches for published job descriptions on how the job is usually perofrmed and prompts for whether to include those in the reported job duties. Claimants almost always under-report past work job requirements on that form.
Joanne Barnhart of HPI fame? Enough said as to her contributions to process improvement.
ReplyDelete@421 we are already dealing with robots when we come to the office, they cant do any worse than the drones there.
ReplyDeleteThis blog lives in the dark ages.
ReplyDeleteAI is already creating treatment plans in hospitals and private healthcare providers based on the information in the medical record, testing, notes, all of it. If it can treat then it can decide a dib claim.
AI can be used to automate the drafting of legal documents such as contracts, agreements, and briefs. This can save significant time and reduce the need for junior lawyers to spend time on routine tasks, allowing them to focus on more complex legal work. Bye bye paralegals and decision writers!
ReplyDelete@11:43
ReplyDeleteI would not trust AI to create treatment plans based on the medical records I review on a daily basis. Yesterday I had to spend time reaching out to my client's oncologist office, because the treatment notes were silent on his cancer stage. If AI reviewed those records, they would be mislead to think he only had stage II cancer, not stage IV. (the records did not include any raw diagnostic images, because they never do).
AI might be better at interpreting raw lab and imaging data, and at diagnosing acute symptoms that do not require a review of the longitudinal records, but the failure point is still human error. The treatment notes are still inputted by humans. AI needs quality data to make the correct decision.
Same problem exists in the disability context, only more so. AI will be great tool for making parts of the disability process more efficient. But anyone that thinks AI can replace state agency disability analyst or Judges in the short-term, simply does not understand the complexity of the disability evaluation process, or the logistical challenge of record development. Claimant's are still entitled to due process and statutory protections.
Also, people seem to be underestimating how slow the administrative state adapts to new technology. Even if SSA wanted to adopt AI tools, it will take 10 years for it to roll out -- and by that time they will be locked into outdated AI tech. There needs to be serious government reforms in policy and operations to ensure AI is effective and compliant with the law. Do you really think lawyers will just bow down to an AI single decision maker? SSA would have to contend with years of federal court litigation just on statutory compliance issues.
@2:36 and 11:43
ReplyDeleteAs 11:43 states the garbage in/garbage out problem prevents an AI from having the capacity for accurate disability decision making. Most medical conditions are capable of producing a broad and varying degree of symptom-related functional limitations, both singly and in combination. With very few exceptions medical records are neither designed nor employed to record that kind of symptom detail. A human needs to make the judgment call on how they would limit a disability claimant's function in a full-time work environment.
Isn’t a huge chunk of this blog dedicated to discussing how poorly humans are using their judgment to make these decisions?
DeleteYou can supplement records with information akin to testimony, teach AI about possible symptoms/limitations arising from various conditions, etc. Current AI likely couldn’t handle it, but AI is rapidly developing and is far from stagnant.
@8:52
ReplyDeleteExactly, AI is the perfect case study for the "garbage in/garbage out problem". In the short-term, AI could be used to improve the quality of the records themselves, but it wont help interpret an existing garbage record. Ideally, AI will be employed first as a tool to prevent/identify human error. I could also see AI being useful in the near-term for vocational issues like transferability of skills -- but not until there is replacement for the DOT (so I guess not near-term).
The DOT was last updated in 1991, when the internet was barely more than a few AOL chat rooms. Since then advances in internet and computer technology have changed almost every job; yet the DOT (with VE support) still talks about jobs that supposedly exists in the hundreds of thousands that claimants can perform the same is in 1991 unaffected by the passage of time. Now the AI will ultimately make many more jobs obsolete. Look at the writer's strike which has been going on for weeks as writers fear that their jobs will be replaced by AI. Nevertheless, in the world of SSA hearings a claimant that has circa 1991 skills will still be considered fully-employable throughout the entire economy. Jobs such as hand packager, bakery assembly line inspector, and all the other wonderful non-existent jobs that VE's love to toss out, like confetti over a parade, will continue to be trotted out with their supposed existence by the millions, despite technology making all these jobs either non-existent or completely different than how they are described in the DOT.
ReplyDelete8:30 AM. How "poorly humans are using their judgment..." This suggests they just are not "understanding" the records... It would be more accurate to say "they are IGNORING the evidence" that would require them to approve, while downplaying limitations...while accepting VE testimony that is outdated and using hypotheticals that simply aren't valid of the claimant's real capacity of sustainable work. Not merely a "mistake" but, an intentional decision to deny people who should have been approved.
ReplyDelete@ 8:30
ReplyDeleteYes, this is the real utility of AI. It much harder to train AI to make subjective legal decisions, but there is growing evidence that existing AI is already better at interpreting raw medical data (images specifically). If AI is trained with mountains of academic medical research papers, it could also be a great tool for educating decision makers about the significance of particular objective findings and projected treatment outcomes -- without the need for state agency non-examining physicians opinions. The state agency doctors simply do not have enough time to fairly review the record, and recruitment of these doctors has been a sore point for many state agencies.
You would be surprised by how many ALJs attempt to interpret raw medical evidence with no support from medical opinion sources. This often results in very sloppy denial decisions that can be easily picked apart in Federal Court. I am currently ligating a case on this exact issue.