Senators Seek Answers On AI Usage

     From Government Executive:

... Senate Finance Committee Chairman Ron Wyden, D-Ore., and ranking member Mike Crapo, R-Idaho, sent a letter to Social Security Administration Commissioner Martin O’Malley asking for information about how the agency is complying with the Biden administration’s March policy mandating safeguards for AI use. ...

Wyden and Crapo also warned O’Malley against overly relying on artificial intelligence as he seeks to chart a path out of Social Security’s customer service crisis, spurred primarily by a slow disinvestment in the agency by Congress, leading to the agency having its smallest workforce in 30 years amid the greatest number of beneficiaries in history. ...

$81 Million AI Contract

     From a press release:

Accenture Federal Services has won an $81 million artificial intelligence and machine learning (AI / ML) contract from the Social Security Administration (SSA). The company will deliver end-to-end back office intelligent automation services to transform the way 250 million retiree and survivors’ benefits documents are processed every year. The forms are currently being received and managed via email, mail, and fax.

The Social Security Administration has already successfully deployed an enterprise hyperautomation platform that uses advanced deep learning and computer vision techniques to identify data from the agency’s electronic folder, extract text, transcribe data with the highest level of accuracy, and speed processing through downstream business processes. The platform uses internal enterprise data to intelligently and rapidly keep pace with high demand, expedite decision making, and control costs.

Accenture Federal Services will now take this project to the next level providing the licenses necessary to deliver an end-to-end, Infrastructure as a Software (IaaS) roadmap for intelligent automation installation, testing, and training. ...

National Disability Forum On AI Scheduled For May 15

     From Social Security:

Save The Date – National Disability Forum

We will host the next virtual National Disability Forum (NDF) on Wednesday, May 15, 2024. The title is “How Artificial Intelligence May Affect the Landscape of Social Security - Part 2.” The event will take place through Microsoft Teams from 1:00 p.m. – 3:00 p.m. ET.

For more information about the NDF, we encourage you and your clients to visit our What’s New page at www.ssa.gov/news/advocates.html. You and your clients can email questions to OEA.Net.Post@ssa.gov.

Social Security Ready To Start Moving To The Use Of Generative AI

     From a Request for Information published by the Social Security Administration:

SSA is looking for a Generative Artificial Intelligence (GenAI) solution capable of: 

• assisting SSA developers in developing code more expeditiously, and/or
• transforming Legacy Code to modern languages for the purpose of refactoring Legacy Systems to leverage modern technologies and platforms. ...

Report On AI At Social Security

     Here's a piece by four researchers touting Social Security's progress in Artificial Intelligence. It's somewhat odd. Apparently, it hasn't been published unless you consider releasing it online as publication. It has no date on it. I don't see any source cited from after 2021 so this may be a few years old. I may not the one who should say this but I'm not sure if what they're calling "Artificial Intelligence" would generally be called "Artificial Intelligence" today.

What's At Stake For SSA With Generative AI

     The Acting Commissioner of Social Security recently made the decision that the agency will, for now, have no involvement with Generative Artificial Intelligence (AI). While I have a healthy skepticism that AI will ever have much role in taking or adjudicating Social Security claims, there's one area where Generative AI may be of considerable assistance and that's dealing with the COBOL problem. Social Security's most important computer programs are written in the ancient COBOL programming language. Many other agencies and private companies are similarly dependent upon COBOL. Schools are no longer teaching COBOL. Computer programmers don't want to work with it. Few programmers are available to work on COBOL programs and those programmers have long since gone gray.

    IBM, and probably others, have decided that Generative AI may be the cure for COBOL problems. There would be no need to laboriously rewrite all the old programs in newer programming languages. AI would quickly rewrite the old COBOL programs in modern computer programming languages. Programmers would be more widely available for modern computer languages. Corrections and improvements to existing programs would become easier. That would be a godsend to Social Security, if it works.

AI Out At Social Security, For Now

     From Fedscoop:

The Social Security Administration has banned the use of generative artificial intelligence-based tools on agency devices, FedScoop has confirmed. The block, which is temporary, is meant to ensure the security of data and systems.

The agency’s block of these third-party tools was disclosed earlier this month in a management advisory report for this fiscal year, published by the SSA’s inspector general office. The report noted that the decision was made to protect personally identifiable information, along with health, sensitive and other non-public information, that risked being shared through the use of the technology. 

The Social Security Administration said the block was a precautionary measure and that the agency has yet to use generative AI. When asked if the ban applies to agency laptops and mobile devices, a spokesperson said the block “is designed to be agency-wide.”  ...

AI And White Collar Social Security Jobs: Part III

     Here's the third and final installment in my series on what Chat GPT can do in a Social Security context.

Request: Write a 10 line poem in blank verse on the subject of Social Security

Chat GPT Response:

Amidst life's journey, a safety net unfurls,

Social Security, a promise held firm.

A shelter for the elders, time's weathered crew,

Their toil remembered, their dignity due.

A woven tapestry of care and grace,

Support in twilight years, a steadfast embrace.

Contributions sown as seeds for the age,

Harvested in retirement, a just wage.

Through cycles and years, its shield endures,

Social Security, a nation's pledge secures

    I'd call this impressive. Of course, if you read anything about Chat GPT you’ll quickly learn about the app’s ability to write poetry. For me, I couldn't write even one couplet in blank verse. One of the questions I ask my clients is "Do you have any hobbies?" It never ceases to amaze me that a not insignificant number of people answer "Writing poetry" and some of those people lack even a high school education. I think there's a poetry gene that I didn't inherit.

 

AI And White Collar Social Security Jobs: Part II

     Continuing on with my series on what Chat GPT can do in the Social Security context, here's another question and answer:

Question: I am an attorney representing a Social Security disability claimant. She is 56 years old, has a 10th grade education and has worked for the last 20 years as a CNA. She has two bad knees and is a diabetic with peripheral neuropathy. What arguments can I use to prove disability?

Chat GPT Answer: Given your client's circumstances, you could use the following arguments to prove disability:

1. **Medical Evidence**: Gather detailed medical records, test results, and doctor's assessments that clearly document her knee problems, diabetes, and peripheral neuropathy. Highlight any treatments, medications, surgeries, or therapies she has undergone to manage these conditions.

2. **Functional Limitations**: Describe how her medical conditions impact her daily life and ability to work. Highlight how her knee pain and neuropathy limit her mobility, standing, and ability to perform physical tasks required in her previous job as a CNA.

3. **Vocational History**: Emphasize her 20 years of work as a CNA, and how her conditions now prevent her from performing the physical demands of that job. Discuss any attempts she may have made to continue working despite her disabilities.

4. **Transferable Skills**: Evaluate whether her skills from being a CNA can be transferred to a less physically demanding job, given her current limitations. If not, this strengthens the argument that she is unable to work.

5. **Age and Education**: Her age and education level can be factors in proving disability. Explain how her limited education might restrict her ability to transition to a new job that requires less physical activity.

6. **Residual Functional Capacity (RFC)**: Obtain a detailed RFC assessment from a medical professional, outlining her ability to perform work-related activities. This can provide clear evidence of her limitations.

7. **Consistency of Treatment**: Highlight any inconsistencies in her medical records or treatment history that might suggest her conditions have not been managed consistently.

8. **Expert Opinions**: If necessary, seek expert opinions from medical professionals, vocational experts, or disability specialists who can testify to her inability to work based on her medical conditions and limitations.

9. **Impact on Daily Living**: Describe how her disabilities affect her ability to perform everyday tasks like cooking, cleaning, shopping, and personal care. This can provide additional evidence of her limitations.

10. **Social Security Blue Book**: Refer to the Social Security Administration's "Blue Book" that outlines the specific medical criteria for qualifying as disabled. Match her medical conditions to the relevant listings.

Remember, the key is to build a comprehensive case that demonstrates how her medical conditions substantially prevent her from engaging in substantial gainful activity. Consulting with legal experts experienced in Social Security disability claims can further enhance your client's chances of success.

    Duh, I'm looking to prove this lady is limited to sedentary work and I may have a good shot at it given the facts I've mentioned. So, why couldn't Chat GPT give more than an extremely generic answer which doesn't speak to the facts I've given? I'd have to give this answer a failing grade -- my job seems secure -- but stay tuned. Part III shows Chat GPT doing something that's impressive in its own way.

AI And White Collar Social Security Jobs: Part I

     There's been talk lately of Artificial Intelligence (AI) taking over white collar jobs. I thought I would take a look at what Chat GPT, the widely available free AI app could do in the Social Security context. In Part I of a three part series this is a question I asked and the Chat GPT response:

Question:  What kinds of disability benefits does the Social Security Administration offer?

Answer: The Social Security Administration offers several disability benefits programs, including Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI). SSDI provides benefits to disabled individuals who have worked and paid Social Security taxes, while SSI offers assistance to disabled individuals with limited income and resources. Additionally, there are programs like Disabled Adult Child (DAC) benefits for adults with disabilities who were dependent on a parent's Social Security earnings, and Compassionate Allowances for faster processing of severe medical conditions. Keep in mind that eligibility criteria and benefits can vary, so it's best to check the SSA's official website for the most up-to-date information.

    It missed Disabled Widow and Widower benefits which is not good. On a more obscure note, it missed that disability benefits for blindness are a separate category in both Titles II and XVI of the Social Security Act.

Could This Sort Of Thing Happen Here? It Already Is!

     From Government Executive:

...  In Australia, the government set off on a radical plan to reduce overpayment of government benefits in 2016. ...

The Australian government had been manually searching for overpayments in programs for retirees, people with disabilities and students, among others. The 2016 program used algorithms to search out overpayments and send the bills. Christened “Robodebt,” the algorithm checked each individual’s payment against the average income of people in similar circumstances. If the algorithm determined that the person was likely overpaid by the government, it generated a bill.

Robodebt allowed the government to review 20,000 cases per week, instead of the 20,000 cases per year in the manual system it replaced. Government officials no longer had to contact employers to obtain data on employment history and payroll amounts, and the government no longer had to prove an individual had been overpaid. Instead, individuals had to prove that they had received the correct amount. If individuals didn’t pay quickly, debt collectors went to work.

The government launched Robodebt fast and claimed credit for catching recipients who had benefited from mistakes in the system. But many people receiving the notices were distraught. They often had to come up with big payments in just a few weeks. Some people had to sell their cars or take out loans, which was a huge burden on some of the country’s neediest residents. Others drained their meager savings. At least three people committed suicide, a Royal Commission found in a devastating 1,000-page report.

An investigation revealed that some repayment notices were incorrect. Some simply were false. Moreover, a 2019 court challenge found that Robodebt had violated important provisions of Australian law.

In July 2023, the Royal Commission pointed to “Robodebt’s unfairness, probable illegality, and cruelty.” When problems surfaced along the way, the commission concluded, “the path taken was to double down, to go on the attack in the media against those who complained and to maintain the falsehood that in fact the system had not changed at all” from the previous system. ...

    Could this happen here? To a great extent, it already is happening. Republicans, abetted by Social Security's Office of Inspector General, have long implied that all overpayments are the result of fraud and that the agency must be merciless in collecting these debts. Recently they have been blaming the agency for not creating overpayments automatically based upon data from payroll companies. Already, there is no statute of limitations, those informed of alleged overpayments are given no information about how the alleged overpayments occurred, and all benefit payments are seized until the desperate claimant asks for a repayment schedule. It's harsh by design despite the fact that the agency often has no basis in fact for asserting an overpayment and many overpayments are due to mistakes made by Social Security. The current attitude is that if the computer says there's an overpayment, there must be an overpayment. It could all get worse with artificial intelligence.