CNBC has posted a recent interview with Social Security Commissioner nominee Frank Bisignano. It sounds to me as if he's having trouble shifting gears mentally to the idea of working in the public sector. Fiserv handles humongous quantities of extremely simple financial transactions. Social Security handles huge quantities of complicated transactions. It's a different ballgame. Fiserv has the money it needs to hire the staff it needs and the technology it needs. Social Security works in an arena where its administrative budget is controlled by people who are often indifferent to the proper functioning of the agency and in some cases eager to see it fail.
19 comments:
Hopefully he is open to listening to people who know what they are talking about and not just the Doggy Doo Doo goons
Comes across as thoroughly unknowledgeable. All he does is reference his time in the private sector and drops some memorized SSA stats. Public service is a different beast. I give him a year to last in the role.
Very true, Charles. But the extent to which SSA’s software is outdated, and the positive impact good software could have, can’t be overstated. That applies double to the use of cutting edge AI techniques in the near future (a four year term as Commissioner years will be enough for a good start). SSA has always been at least a decade behind the curve in adopting new technology, but if this guy comes in committed to maximizing AI’s use it will revolutionize SSA processes just like it’s going to revolutionize every other area of society. An AI model/agent that is customized for SSA‘s purposes is going to be much more effective than simply hiring even 10,000 new employees who need to be trained and then might stay or not stay more than a year or two. Every big company is clamoring to figure out how to adapt AI for their purposes because they know if they don’t have it their you competitors will put them out of business. What possible reason would there be for government, and specifically SSA, to fail to use it to serve the public better? It would be better to just all get on board with its inevitability and really try to provide world class service instead of clinging to how we’ve always done things. It’s not as if we don’t all constantly complain about how things are done and being overworked.
Like it or not, we are truly on the cusp a brave new world. If you disagree you simply have not read up enough on the current state of AI and where it is heading. It’s like disagreeing with the advent horseless carriages - a new age is coming. It has long-term terrifying implications for our society as a whole as more and more jobs are replaced, but for SSA’s service to the public it’s good news. Waits will be much shorter and our work product much more accurate. It seems to me there should be no reason for anyone that currently has a job with SSA to lose it due to AI, but there is very good reason to believe SSA won’t need to hire as many people to carry out its mission 5 to 10 years from now.
@8:57 AI is not some magic bullet. The systems SSA relies on for core IT infrastructure would not even be interoperable with AI (at least to a degree that would be meaningful). That is what needs to be fixed first. There are many software automations and efficiency improvements that could be accomplished without AI. POM rule changes are also a potential goldmine for identifying non-statutory mandated roadblocks/wasteful procedures. For example, attorney assignment could be automated.
Magical thinking. He's there to dismantle, cannibalize, privatize. He'll last as long as GOP holds the White House.
I agree with 8:57AM above BUT I think the SS and SSI programs would have to be simplified for AI to work. For example:
1) Get rid of offsetting SS and SSI benefits. Just pay benefits owed under both programs.
2) Get SSA out of the business of being a third party for payment between an attorney and their client (SSA claimant). If an attorney takes a case, they should be the ones to deal with getting what is owed them from their client NOT relying on SSA to pay them. If they don't get what is owed, take them to court just like any other business has to do when owed money.
3) Get rid of SSI couples' cases. The amount of work and time involved to separate a couple or to make two SSI individuals a couple is an absolute nightmare. And, as soon as the members of the couple realize they will get less money, they are right back into the office stating they are now separated so they can get the higher amount of SSI.
4) SSI needs a living arrangement where if you live alone, you get the full FBR minus countable income/resources and if you live with others, you get a reduced FBR minus countable income/resources. LA-D would continue.
The issue of separate purchase of food, room rental, rental subsidy because a parent charges less rent than a non-relative - get rid of all of it - way too much to document and verify.
I have my doubts that if AI were brought in, the above scenarios would be able to be done based on current rules. They are just too complex.
Make both programs simpler and not take so long to make changes. Anyone (congress/senate/president/vice president) who has NOT worked at SSA and experienced working in a FO should just not comment as they have no idea what is involved with dealing with the public (now feeling more entitled than ever).
Customer service is not the same as it used to be either. I used to take care of the public with everything needed when they came in or called on the phone. I didn't piece-meal everything so they didn't have to make multiple trips into the office to take care of multiple things. Some workers would just do the bare minimum (change the address on the SSI record, but not deal with a diary or RZ that was also on the record). Made for frustrated customers and employees that cared about doing what was needed.
Just my opinion as someone who has previously worked in multiple FOs, a TSC, and HQ for 40+ years and is now retired.
Looks like they already pushed release of an AI tool that is in no way ready. It doesn’t even seem to be trained on the publicly available policy manual. Even the training video for the tool had an example “answer” that was between misleading and just wrong. Maybe the AI bubble bursting and people realizing it’s all a smoke and mirrors scam will end this tech bro nonsense.
AI is advancing rapidly in every intellectual endeavor. Diagnosing conditions. Writing essays. Programming. Read up on it. Listen to podcasts about it, from reputable sources. There is no reason it can’t help with our mostly straightforward application of regulatory policies. It’s not rocket science, but even if it was, AI could help. Discard 1990’s-style thinking. It won’t replace SSA employees for quite a while, but having an AI digital assistant will dramatically improve productivity for any CR or SR or decision writer or ALJ, as long as the SSA culture embraces it instead of opposing it or work-grouping it to death before rolling it out. Our processes are unbelievably antiquated. If you care about our mission and actually helping the people we serve, you should embrace it.
Yeah, we want some AI getting deranged while doing mission critical work (payments, dib decisions, mission critical).
SSA cannot train people properly, they cannot even remotely think they can properly train an AI. Garbage in, garbage out. Just look at the state of some of the notices coming out of the agency.
There will still be humans in the loop. They just won’t have to do everything from scratch. It’s not an all-or-nothing proposition or global implementation. And are you seriously arguing that there is no human error currently? Like getting inaccurate info during 1-800 calls or faulty income and resource determinations or inaccurate overpayment calculations or erroneous PRW findings, etc? It happens constantly. We create entire divisions or branches to catch the errors. Why wouldn’t we try to make that more efficient? Hate to tell you but AI is probably being used in medical contexts already in the “real world,” possibly the next time you go to a hospital or doctors appt.
It’s easy enough to limit AI to contexts in which hallucinations can’t affect service to the public. And the problem of hallucinations will lessen over time.
Anyone who thinks this guy cares about SSA programs is kidding themselves. He's only here to fire as many people as he can and replace with AI that doesn't work. He's being appointed to kill the agency.
To quote Henry Ford, you cannot build a reputation on what you *plan* to do. AI has yet to provide any meaningful help to the hearing office. IMAGEN is still not available for use, the de-duplicate program we were promised 8 years ago never materalized, and INSIGHT A) still makes laughable mistakes, B) hasn't been uupdated with recent reg. changes, and C) with the Appeals Council now issuing substitution-of-judgment remands, the whole point of giving INSIGHT to the HOs is moot as it still doesn't prevent remands.
WHile i agree that the things you mentioned could (should?) be changed to simplify SSA program, they are EXACTLY the type of things that AI can handle easily.
Good luck cramming that RZ via CCE while answering GI line - nobody else around to answer the rest of the calls. Sounds nice but not practical in 2025.
When Medicare Part D was rolled out, I was hired as a new fresh SSI CR (yeah we were CRs back then) We had live video training that was interactive. It took 16 weeks. Then two years of monitored work before we were totally turned loose with our workload. It took all that time to become proficient and professional due to the widely varied claims work that needs to be done. Training an AI properly is going to take a great deal of time to understand the complexity and intricacy of the program. AI also cannot interpret that the numberholder is confused, low intellectual functioning and a poor historian. A trained and seasoned worker understands that the answer given by the nh may not be complete or factual and rephrase and develop the actual truth of a situation. Sure AI might be able to handle things like a straight forward single, no past marriage, no children retirement. But there are a lot of complicated cases. The only way to make AI work with SSA is to simplify the programs that have drifted from the original purpose and been patched and added onto for over a century by lawyers and politicians that have no idea how the programs work. Simplified means people will get dumped and benefits will cease. 68 million people get a check every month. AI is a direct threat to all of them.
Completely agree. Which is why it is worth reserving judgment on seeing what a Commissioner with experience running an organization built on tech can do. Our last few Commissioners have been: 1) a politician; 2) an academic; 3) a businessman/crony; 4) a long time SSA employee steeped in doing things the old way. All had their strengths, but none was well-positioned to implement tech initiatives.
1) many CRs are great, some aren’t. There is plenty of human error occurring every day
2) even for great CRs AI can help them do their jobs faster/better. Who wouldn’t want a digital assistant that knows its stuff?
3) ok, use AI on the easy stuff. That frees up time for humans to do tasks requiring judgment and discretion
4) the tax code is complex too, but plain old traditional programming has been implemented to help that for decades
5) AI is already performing many tasks much more complicated than applying our regs.
It is a matter of will and competence and accountability at the highest levels. Not saying that will happen. Just saying it could and should.
At the very least, AI could be used to help answer calls. For example, if implemented correctly, AI-directed calls would be able to print and mail a BEVE, 1099, or appeal form/packet to someone. I agree that the programs are too complex for a quick-AI-fix, and the public is not always straightforward with their answers for AI to handle it (e.g., were you married 10 years? Yes/No/Maybe/Separated). Technology that could scan, read and transcribe forms would also go a long way to assist the current employees who spend too much time on data entry on hard-to-read handwriting.
Post a Comment