Probably most readers of this blog have tried using AI at least to some extent. Some may be using it a lot. How much use can the Social Security Administration get out of AI? I’m generally skeptical if not hostile to the idea but I’m old. What do you think? Are you already using AI for Social Security purposes? How well does it work for you? For that matter, to what extent is AI even available to agency employees? Is ChatGPT blocked on Social Security computers?
30 comments:
Nothing wrong with being a dinosaur! Senior citizens have the money and power to dictate policy. I only use my credit card for big item purchases. AI is getting out of hand and it’s very dangerous on how much individuals rely on it. Ask any college graduate how the job
AI is not perfect but neither are ALJs and decision writers (I know that will come as a shock to a few of them). The thing that would make AI superior to them is the lack of bias. Of course, the agency would like program it in such a way, as they do with ALJs, that decisions would continue to lack support by any evidence based or science based research such as the current policy of putting anyone with problems using their hands, no matter the severity, at frequent handling and fingering. At least an AI would have the excuse of the way it is programmed. What excuse do ALJ's have?
AI will eventually eliminate most decision writers but no time soon. Our “tech” is terrible from laptops to case processing to time and attendance. Eventually, though, you will have judges using AI to draft most cases.
Will AI, in writing unfavorable decisions, cite to hallucinated evidence?
The DDS and Review components use IMAGEN. It is an AI program, but infantile. Eventually, it and other AI programs will be increasingly used in disability cases.
SSA rolled out its own internal chatbot called the Agency Support Companion (ASC) in late 2024 which is based on some iteration of ChatGPT. Its training data was general purpose and did not include specific SSA information like the POMS. I used it to generate my 5 bullet points for "What did you do last week?" back when DOGE was destroying the agency.
Everything is blocked on agency computers except for the applications that we use for our jobs, and any approved websites that are needed for our jobs. Leland decided we needed zero access to anything at all, to the point that they blocked access to websites we actually need for work, and had to walk some of that back.
Not sure how available AI is for us honestly.
AI tends to reinforce and magnify the biases of the infallible humans whose work is used to train it. Endless papers and books have been written about this very well documented darkside of the technology.
ChatGPT and all other external AI is blocked on our laptops. We do have Agency Support Companion (ASC) which is like an enterprise version of an older version of ChatGPT. We also should, if not now in the very near future, have the Policy Analysis Tool (PAT) which is like a standalone ASC but they fed it POMS. It's hit or miss with accuracy.
from the Social Services side of things, I do use AI some, letters, in specific, lowering the reading level so clients understand. Its also okay for some basic things. Our agency cannot afford the AIs that are designed to work with SSA disability cases, so for now its just an advanced spreadsheet and word document for me. I expect SSA to use it more and more to replace the depleted staff. I will take it, any intelligence at the agency would be welcomed. You should see some of the OP letters, denial letters and requests for additional info letters we are getting. AI could do those better, certainly the same level of inaccuracy.
I think the term AI gets generalized with bots.
The agency is saying AI is processing voluntary tax withholding forms. But it really is not AI. It’s looking for what box is checked and comparing the SSN and the person’s name. It’s not thinking for itself. Just what it was programmed to detect and what action to take.
I’d like more of the automated help. For example - auto sending f/up notices when the tickle matures. Comparing what is in WorkTrack first.
We can’t even get the appointment scheduling system to recognize when a person files online. That would be an awesome automation right there.
There are not and never will be any computers or AI tools that can think. I wish people would realize this.
At the NOSSCR Conference, the use of Claude and ChatGPT were demonstrated as a tool to quickly generate comprehensive summaries of the evidence. A version this with no sharing of PII beyond SS files could enable both ALJs to see what is in the file in detail and, for decision writers, that write decisions that mainly just regurgitate the evidence, it would expedite decision writing. But decisions still require a rationale that goes beyond the summary. No doubt they would use it to write decisions but more than mere summaries make a competent decision.
There are people that can't think either - even some who seem very intelligent.
One of AI's many problems is a human nature problem. Workers under pressure to produce more work are greatly tempted to use every every short cut and efficiency they can get. You can save a lot of time with AI if you don't bother to check its results. You don't save much time with AI if you do responsibly check the results. What do you think will happen in that scenario, given an agency obsessed with having its employees produce more work in less time?
Who is bothering to check the results of the current processes? Might be an improvement.
Do not be surprised here. AI is making great strides at SSA in MER summarization and decision support. Beta deployments are in use now. Headed towards widespread use in this calendar year. Newer examiners will adopt it quickly as they do not know any better. Experienced older examiners are a dying breed as have not adopted tools like IMAGEN. Workload shifts are coming soon.
Between Health IT to acquire MER quickly, IMAGEN to extract supporting information that matches disability impairments, AI to summarize medical records and ML to write write decisions. The tools are in place to make dramatic shifts in processing claims. We could see auto-adjudication sooner than you think. Frank and the Chiefs will push it. Most staff and claimants will not know the difference. ALJs are next. Fix it in CDRs with all the cost savings. Automate that part too.Frank’s goal is cut SSA and DDS staff down to a couple of thousand folks. The future is coming soon.
You are hallucinating so much that you must be AI yourself. Health IT has been around for over a decade. IMAGEN is a clumsy 1995 Web 1.0-looking tool which can pull pages but can't think or simulate conversation. There is not yet AI to summarize MER at the hearing level. SSA simply can't afford cutting edge AI tools. Exactly none of this is going to happen. New ALJ hiring announcement went out today in fact.
Says someone who clearly has never done the jobs he thinks these s**tty software programs facilitate.
I don’t believe Frank and the deputy commissioners will be allowed to cut services in SSA. A portion of the public will never go online or need help.
Maybe, but also consider hallucinated C.F.R. and SSR citations.
You think they’re bad now? Wait until October, when everyone has to work even faster. You’re going to love it. And remember, this crap is what’s going to be used to train the (literally) brainless machine you think is going to do an even better job!
Greater than human level intelligence will soon be a general commodity. You will pay for it like electricity to help you solve problems and accomplish work of all kinds. It’s going to obsolete not only occupations, , but entire fields. We are on the cusp of mind-blowing advances in every field of human endeavor. Currently it is solving mathematical paradoxes that have bedeviled humans for a century. Similar progress is being made pharmaceutically and with respect to biology and robotics. Almost all intelligent, informed people who have skin in the game, meaning they want to stay competitive in the real world, now agree on this general path, although they may disagree on when certain milestones will be reached or the rate of implementation. But however you look at it, it will be soon.
There is zero reason why AI can’t revolutionize SSA’s work (or the work of SSA–adjacent professions like representing claimants). I personally believe that not only should a human remain in loop, but the ultimate (disability) decision should always reside with a human, not AI. But short of that, AI can and should help us with every single task we perform. AI can help with anything you use your brain for. Agentic AI will be able to do a much better job than humans with respect to carrying out online tasks (for example, reviewing productivity and workload data and taking effective online actions to make things more efficient). That’s particularly true when you compare what it will be able to provide to the kind of service many people get right now from humans - not infrequently inaccurate and leaving members of the public feeling disrespected or rushed along.
SSA has been woefully incompetent for decades in terms of utilizing software for the nitty-gritty tasks its employees perform. That’s the only thing protecting the SSA workforce from further reductions right now. Frank came into office talking a big game about utilizing software, but so far software improvements are proceeding at the usual snail’s pace. The AI tools SSA employees currently utilize are less powerful than what anyone can get for a $20 subscription to OpenAI or Anthropic. But sooner or later the AI tools will overcome even SSA’s legendary hidebound software inertia. The difference between today’s AI and the original ChatGPT released less than five years ago is enormous, and that rate of change is continuing to increase. The frontier labs will soon reach the point where AI itself creates the next generation of AI models, a recursive loop of self-improvement. Sooner or later those advances will filter down to all organizations in society, Including SSA.
If I had the power to determine the future I would not necessarily choose this future. But it is inevitable. There is not much to debate about what will happen in that regard, only how fast. The real question is whether society as a whole will be able to adjust to that rate of change and upheaval and come up with solutions to not only survive, but thrive. That is absolutely possible, though hard to imagine in light of the ongoing dysfunction in our politics and government. We need to focus on that or dystopia won’t begin to describe what society will look like.
I could totally see an AI program citing to rescinded SSRs. In an area I practice in, there is an Acquiescence Ruling that the ALJs would frequently cite. Eventually, because the underlying Decision was later overturned, the AR was rescinded. However, I still get ALJs/writers citing to it. If humans are doing that, I easily see an AI program doing that
I'm a bit of a Luddite, so I very sparingly use AI. Generally, I use it to write one-off letters that we don't have a template for. I also have recently started using it to give insights on data in a spreadsheet (e.g. is there any correlation between cases aging and Hearing Office assigned). One of my concerns with using it for records, summaries, etc. is: when it "reads" the records, is private client information being stored on some server somewhere?
A podcast addressing the use of AI in the legal field. It is co-hosted by a former Chief Justice of the Michigan Supreme Court. Very illuminating.
https://youtube.com/@aiandthefutureoflawpodcast?si=539plyIRsUttgO2X
Zero reason AI can’t do the work? How about the FACT that AI isn’t really good at doing much more than generating amalgamations of saved data, dryly summarizing large volumes of text (and only if the text is relatively repetitive and limited in scope), or solving problems with relatively few variables and very limited and well-defined solutions (i.e., solving math problems or programming snippets of computer code for discrete applications).
I believe Frank is trying harder to update our technology than the last few commissioners.
That being said, SSA is in the worst condition it’s been in decades from a workload and morale standpoint. My office is being held together by OT only at this point. If that goes away, things are going to snowball quickly.
@2:00 You are uninformed. Read up on the technical news that is pouring in about AI benefits in virtually every field. Pretty unlikely that private industry would be investing billions in it if it was as anemic and fault-ridden as you portray. And it is getting better by orders of magnitude every year.
Is it doing these things in SSA yet? No. But within a couple of years nobody in SSA will be arguing about this. It will be quite clear. It just seems foolish to pretend like it’s not happening. And it has little to do with Frank – it’s way bigger than him or the president.
Post a Comment