What Happened When California's State Government Examined the Risks and Benefits of AI? (msn.com) 80
An anonymous reader shared this report from the Los Angeles Times:
AI that can generate text, images and other content could help improve state programs but also poses risks, according to a report released by the governor's office on Tuesday. Generative AI could help quickly translate government materials into multiple languages, analyze tax claims to detect fraud, summarize public comments and answer questions about state services. Still, deploying the technology, the analysis warned, also comes with concerns around data privacy, misinformation, equity and bias. "When used ethically and transparently, GenAI has the potential to dramatically improve service delivery outcomes and increase access to and utilization of government programs," the report stated...
AI advancements could benefit California's economy. The state is home to 35 of the world's 50 top AI companies and data from Pitchfork says the GenAI market could reach $42.6 billion in 2023, the report said. Some of the risks outlined in the report include spreading false information, giving consumers dangerous medical advice and enabling the creation of harmful chemicals and nuclear weapons. Data breaches, privacy and bias are also top concerns along with whether AI will take away jobs. "Given these risks, the use of GenAI technology should always be evaluated to determine if this tool is necessary and beneficial to solve a problem compared to the status quo," the report said.
AI advancements could benefit California's economy. The state is home to 35 of the world's 50 top AI companies and data from Pitchfork says the GenAI market could reach $42.6 billion in 2023, the report said. Some of the risks outlined in the report include spreading false information, giving consumers dangerous medical advice and enabling the creation of harmful chemicals and nuclear weapons. Data breaches, privacy and bias are also top concerns along with whether AI will take away jobs. "Given these risks, the use of GenAI technology should always be evaluated to determine if this tool is necessary and beneficial to solve a problem compared to the status quo," the report said.
Re: (Score:3, Insightful)
If it's used a tool to point out oddities for further investigation, I see no problem with it. It's only when you rely on it as your sole quality analyst that the shan hits the fit.
Re: (Score:2)
Re:They seem to be missing a few risks (Score:5, Insightful)
If your time is valuable, why do you want to spend it having to pour through every single possible return rather than just those that look the most suspicious?
Re: (Score:2)
Re: (Score:2)
You're arguing that they're no better than random. Which is nonsense.
Today, tax returns are sampled randomly and/or through simple heuristics. Why not have much smarter heuristics so the time spent analyzing them will catch a higher percentage of cheats, and higher value cheats?
Re: (Score:3)
Re: (Score:2)
Then find the budget to hire orders of magnitude more humans. AI is dirt-cheap to deploy.
Re: (Score:2)
Re: (Score:2)
The difference in costs between the two is meaninglessly small compared to the cost of the human reviewers. Increasing the efficiency of the human reviewers by presenting them with a higher rate of bad returns is by far the driving cost factor.
Re: (Score:2)
California only looks at about 1% of returns. So if you want a human professional to look at every return to determine if it is suspicious or not, they'll need a hundred-fold increase in their budget.
So what do you do?
1. You could continue the current system of selecting that 1% pretty much at random.
2. You could use an AI that correctly identifies suspicious returns 90% of the time.
Re: (Score:2)
On the other hand, if the budget is more important than people, then maybe they should start by looking at the top 5% of income earners, they can increase the budget 5-fold and have a good chance to recoup their costs. If they look at the bottom 5% instead, they'll never make their money back.
Re: (Score:2)
I don't see a problem with increasing the budget on something that can have a huge impact on people's lives if done incorrectly. With great power comes great responsibility, etc.
So you want to increase the budget for, Police, military, health, education, housing, social security, etc ... basically everything?
Where is this money going to come from?
Re: (Score:2)
I don't see a problem with increasing the budget
Every dollar spent on tax audits is a dollar not spent on something useful.
maybe they should start by looking at the top 5% of income earners
Golly, why did no one ever think of that before?
Re: (Score:2)
Every dollar spent on tax audits is a dollar not spent on something useful.
Dollars spent on tax audits return more money than you spend.
Re: (Score:2)
Dollars spent on tax audits return more money than you spend.
That is only true if you're currently spending too little.
Perhaps auditing 2% instead of 1% makes sense, but diminishing returns will happen very soon.
Auditing 100%, as you suggest, makes no sense.
Google says the audit cost is $2,500 per return. California has 18 million tax returns, so the cost would be $45B out of a total budget of $225B.
Re: (Score:2)
Where did I say audit 100%?
It would be like me claiming you wanted to audit 0% because you think it's too expensive.
Re: (Score:1)
Re: (Score:2)
And in a perfect world, it would only be used as one tool. However, the corporate structure of companies will push to squeeze every ounce of value they can out of it. And some of those ounces of value will be people who merely use it as a tool instead of having the tool replace them. And who could stand up and say no when it becomes company policy, either openly so or clandestinely so?
Then there is the issue with local and national governments, especially those that see government workers as an enemy. They'
Re: (Score:2)
Care to bet which way they'll use it?
Re: (Score:2)
Cops, including tax cops, are not smart enough to understand that. They think the AI thingie is an oracle of truth, and act accordingly.
Re: (Score:2)
Cops, including tax cops, are not smart enough to understand that. They think the AI thingie is an oracle of truth, and act accordingly.
How would they even know?
They used to get a pile of returns to audit. And instead now they get a pile of returns to audit.
The only difference is they are much more likely to find stuff now, since the returns are now filtered for more likely errors and fraud.
They'd still have to find and show the errors and fraud. Same as before.
Re: (Score:2)
That's a risk when folks who aren't inclined toward critical thinking (I'm inclined to include anyone who reflexively says nonsense like "taxation is theft!") naively trust whatever Facebook I mean the computer tells them, for sure. Fortunately, the party of "government is incompetent, let me show you how incompetent it can be!" seems to be self-destructing and we're increased participation in citizen government.
Back to square one (Score:1)
So, it can do wonderful things and it can do horrible things. But nobody knows the actual future mix.
Re: (Score:2)
I pretty mich doubt it can actually do things at this time. Well, let's say I doubt it can do more than a small number of things, that do not matter that much, with the accuracy and reliability needed. And hallucinations cannot be fixed with LLMs, or statistical models in general. Oh, and model collapse will pretty much make the current generation the last one trained on public data from the Internet. Oh, and model poisoning is a thing.
My take is some people want to get filthy rich and hence hype this tech
Re: (Score:2)
Actually the report is brutal: 7 pages of marginal benfits and 13 pages of substantial
risks.
Potential benefits:
- "Conduct sentiment analysis of public feedback on state policies"
- "Summarize meetings"
- "identify specific groups or subsets of participants that may benefit from additional outreach"
- "convert educational materials into formats like audio books"
- "Translate COBOL into [Ruby on Rails]. This can slash timelines, reduce bugs, and democratize development." (Good luck with that.)
Among the many risks
Would need a lot of caution in using it unguarded (Score:2)
It may have not been ai but the Australian government used automated analysis of benefit payments and resulted in many people being hounded incorrectly and a number of persons removing themselves from among the living. It would need to be backed up with multiple levels of actual people checking before action should be taken based on an ai analysis. Use it as a prompt but donâ(TM)t treat it as gospel
False positives vs false negatives (Score:2)
Human review of the positives doesn't resolve the issue of people getting their hands on the AI and using it to tune their tax filings to avoid detection of their own tax fraud.
And then getting away with it even if caught on the grounds that "the government's own tax audit AI was used to review this to ensure compliance!"
Re: (Score:2)
Human review of the positives doesn't resolve the issue of people getting their hands on the AI and using it to tune their tax filings to avoid detection of their own tax fraud.
"AI" is a black box that doesn't show its workings. You won't know what to hide, because nobody will know what the criteria actually is.
And then getting away with it even if caught on the grounds that "the government's own tax audit AI was used to review this to ensure compliance!"
The AI isn't making the final decisions. It's just a tool to help the people who are making the decisions.
Re: (Score:2)
There is tax software that follows rules to fill out your tax forms for you. Now imagine you change the rules so they produce options for deliberate misreporting, and you feed those into the AI to see what gets flagged. You don't have to understand the black box to utilize it.
And your second point is not a rebuttal of mine.
Re: (Score:2)
There is tax software that follows rules to fill out your tax forms for you. Now imagine you change the rules so they produce options for deliberate misreporting, and you feed those into the AI to see what gets flagged. You don't have to understand the black box to utilize it.
You're assuming the AI never changes. Next year when the AI notices a pattern of deliberate misreporting, it will be trivial to go back X number of years and catch everyone who did this all at once.
X is whatever the statutory limits on audits are.
And your second point is not a rebuttal of mine.
How are they getting away with anything? You just said they were caught.
And then getting away with it even if caught on the grounds that "the government's own tax audit AI was used to review this to ensure compliance!"
Another silver bullet. (Score:2)
We seem to find one every few years.
Re: (Score:2)
This internet thing is a fad. Paper is faster.
Re: (Score:2)
Oh, the old Musk shill is still about. Yeah, paper is a lot better than the Internet, but why compare apples and steel pipes? Be smart, compare paper to color e-ink, FFS.
Re: (Score:2)
Oh man, you totally nailed me [slashdot.org]. Love me some Musk [slashdot.org]. What a wonderful person [slashdot.org]. Definitely not a deeply problematic conspiracy theorist. [slashdot.org] I'm his biggest fan. [slashdot.org] Think he personally invented everything and is a lovable guy. [slashdot.org]
Learn to tell the difference between a company and its CEO.
I'm sorry, I didn't get your message. Could you write it down and then fax it to me? Thanks!
Re: (Score:2)
I don't fax, darling, I send real letters written with a quill on a parchment. Now be so kind to bury yourself under whatever comes out of your nearest Katla.
Re: (Score:2)
There is a sucker born every minute...
No idea why with AI, perpetrating this ages-old scam seems to be even simpler. All demonstrations of AI so far left me very unimpressed. Simplistic, has no ability to fact-check, hallucinates, cannot even do beginner-level stuff in many areas (students of mine tried it on an utterly simple firewall exercise with zero success), etc. Especially hilarious was a presentation were a supposed senior pedagogics expert demonstrated chat AI to create teaching material. I mean, m
Re: (Score:2)
> It did create the crap it created very fast though and it looked good, even if it was not.
Re: (Score:2)
You imply that is enough to impress many people, including many people with money? Makes some sense, in a very unfortunate way. Well, I have concluded a while ago that most people are idiots, and this hype is just one more inciator for that.
Re: (Score:2)
Re: (Score:2)
Ridiculous, really, sort of similar in my field (physics) where people would feed data to AI, receive some "results" and when asked to explain them in terms of theory will grind to a halt.
The few times I've helped dig through "results", they turn out to be statistical artifacts from hidden or poorly understood correlations in detectors or other experiment setup defects. Sad, really.
I mean, it is useful to have a helper to sift through a lot of data, but the researcher should know why and how it works, not j
Re: (Score:2)
the researcher should know why and how it works, not just feed "AI" garbage and label the output as "results".
Yep. Otherwise they have no business calling themselves "researchers".
analyze tax claims to detect fraud.. (Score:2)
Can't have that, the lobbyists will prevent it.
Re: (Score:1)
s/lobbyists/donors/
That's why a certain US political party wants to defund the IRS.
Re: (Score:2)
Yep. So utterly convinced they have truth and honor and know how to build a bright future, that they are convinced _they_ do not need any checks and balances. That is how you build fascism and have everything reliably go to shit.
Re: (Score:2)
Naa, it will hallucinate so many problems with small-time tax payers that the large ones will be even safer!
translations (Score:3, Insightful)
It'll be used to do... (Score:2)
Re: (Score:2)
Re: (Score:2)
If you'd like to leave a review of your online shopping experiences, please find the appr
Re: (Score:2)
Do you buy things that are shipped by some entity other than Amazon? If so, that's your problem. Don't trust "third-party" fulfillment. (Yes, I know that Amazon is the third party from a certain perspective. I don't think that's a helpful perspective.)
Re: (Score:2)
To quote XKCD about any new technology:
Will [ ] make us all geniuses? No
Will [ ] make us all morons? No
Will [ ] destroy whole industries? Yes
Will [ ] make us more empathetic? No
Will [ ] make us less caring? No
Will teens use [ ] for sex? Yes
Were they going to have sex anyway? Yes
Will [ ] destroy music? No
Will [ ] destroy art? No
But can't we go back to a time when- No
Will [ ] bring about world peace? No
Will [ ] cause widespread alienation by creating a world of empty experienc
Re: (Score:2)
Re: "We were already alienated." - Not around here, we aren't.
Also: "Will [ ] further concentrate wealth & power into the hands of an unelected, unaccountable few? - Yes."
Re: (Score:2)
If you had spent time in the open source LLM community, you might not be so confident of that statement.
It's like thinking that operating systems will stay in the domain of wealthy powerful corporations because it takes such a large amount of resources to develop an OS.
Re: (Score:2)
Re: (Score:2)
They did?
Who did? I don't remember such a timeperiod. I don't even remember the term "social media" being "a thing" until Facebook. Like, there were "blogs", but these weren't generally self-hosted. People running their own websites was seen as nerdy and esoteric.
Re: (Score:2)
There were many such claims about decentralised networks & many attempts continue to this day, e.g. Mastodon. Still, the space is dominated by a handful of large corporations & even for specialised, closed systems for corporations & institutions, those social networking platform
California? (Score:2)
Re: (Score:2)
Genius. Nobody else knows it but Big Brother from California does.
Risks of AIs ... (Score:2)
Articles about the risks of AIs make me think of the several Star Trek: Lower Decks episodes and the Daystrom Institute Self-Aware Megalomanical Computer Storage Facility housing the various evil AIs like "Lord Tyrannikillicus" ...
Sad that California is now a land of intrigue. (Score:2)
Re: (Score:2)
When was that, exactly - 1518?
Re: (Score:2)
Re: (Score:2)
Let me ask Rodney King if he remembers the California of several decades ago in as idyllic of a manner as you do.
Re: (Score:2)
Re: (Score:2)
I'd rather be the curator of anecdotes than the architect of delusions.
Re: (Score:2)
Re: (Score:2)
I spent far more years than I would have cared to living in that country, thanks.
Including, I should add, three in California, plus numerous visits.
Re: (Score:2)
Re: (Score:2)
Wow, amazing, Rodney King was a freak one-off incident, and not just emblematic of decade after decade of seething racial conflict and inequality, amazing!
Next you'll inform me that there's weren't any conflicts related to migrants in California! No housing crises! No massive problems with wildfires from overdevelopment! No aging infrastructure, esp. the power grid! No problems with a government constantly struggling with a debt crisis! No mess of a proposition system where the public has voted away right
Re: (Score:2)
The whole rest of the country used to stereotype it that way for most of living memory, so I'm gonna say your question is dumb.
I used to listen to the Beach Boys and watch surfer movies starring Annette Funicello.
Then I grew up, moved to California, and saw reality.
The Beach Boys weren't surfers, it was all an act.
The beaches in California are cold, rocky, foggy, and the waves are usually less than a meter. It's one of the worst places in the world to surf but one of the best for hypothermia.
The "California Dream" was all false advertising.
Re: (Score:2)
morons (Score:2)
New headline: government ignoramuses produce study about something they don't understand.
creation of harmful chemicals and nuclear weapons (Score:5, Insightful)
Do not like this new trend of security by scientific illiteracy. The knowledge of how to make nuclear weapons is nothing, you get dangerous when you have mined the uranium, and have the hundreds of millions of dollars of centrifuges turning to enrich it. Same with dangerous chemicals. A person making TNT without a license is dangerous, a person who knows how to make it is any chemist. The war against dangerous knowledge is a war against a scientifically informed public.
Re: (Score:2)
Let me guess (Score:2)
"AI" is just new data structures (Score:1)
It blows my mind that a site like this has such ignorant discussion of this topic. It's not artificial intelligence or anything close to it. It's a collection of data structures and algorithms. AI processors are just like normal processors except they lack most of the scalar instructions you'd expect in a CPU and replaces them with vector instructions that play well with the data structures required/utilized by neural nets.
Which is exactly why the state of California made their recommendation. None of thi