UK officials use AI to decide on important issues

[Neutrion]

https://www.theguardian.com/technology/2023/oct/23/uk-officials-use-ai-to-decide-on-issues-from-benefits-to-marriage-licences

 

So regarding the question of taking over, it seems that they are not yet, but we are willingly handing over the decision making powers already.
And again if an AI system is very complex, and one would like to complain or doublecheck things, well than good luck finding out why a decision was made and whether it is correct.
So we are creating a machine which we just have to trust, and that was it, because currently not even the creators understand some of the systems.

Also quiet remarkable, that while for combating the global warming there is a democratic mandate, most of the people and scientist think that goverments need to do something and still hardly anything effective happens, there was I think never a democratic mandate anywhere for handing over complex decisions about human lifes to AI and yet it seems to happen on fast track. We have no clue either what the HR systems of some companies are doing.

Also interesting article:
https://www.theguardian.com/technology/2023/oct/24/ai-firms-must-be-held-responsible-for-harm-they-cause-godfathers-of-technology-say

One of the major request for AI companies:
"AI companies must adopt specific safety measures if dangerous capabilities are found in their models."

Well Deep fake in itself is something which will make us go back to middle age-like circumstances, where no one is going to belive what they see or hear in the news, because it can be a fake.
(Yes we saw even states creating false evidences in the last decades so the trust in official communication could sink to much lower levels than where there is already. )
People will only belive what they saw with their own eyes. Anything else can be a fake.
And we already listed plenty of other such effects which are not hidden or miraculous at all.

And I already see the great ansver from politics: "Well we just have to create a Ministry of Truth to make sure only real things get through to the people"

 

[themadhippy]

not sure which is worse,the old system using muppets or the new using robots.

[schmitt trigger]

George Orwell would be proud.

[ebastler]

What's wrong with using pattern recognition in complex data sets? 

It's not like they are asking ChatGPT for advice there. These are specialized solutions, trained on specific data sets. They flag suspicious patterns for human review which otherwise would go unnoticed.

Come on folks, this is a technology forum. We should be able to go beyond knee-jerk responses to trigger words.

[eTobey]

Today we can send millions of dollars around the globe (if someone has it), so we could make a system and democratically vote for decisions. But this is too democratic, so we replace puppets with robots...

[ebastler]

Today we can send millions of dollars around the globe (if someone has it), so we could make a system and democratically vote for decisions. But this is too democratic, so we replace puppets with robots...

The Guardian article is not about making policy decisions and passing new laws. It's about individual administrative decisions, e.g. which marriage is only pro forma for immigration reasons. And it's about screening even larger data sets for patterns which might indicate problematic constellations. You want to let people vote on this? 

[Neutrion]

What's wrong with using pattern recognition in complex data sets? 

It's not like they are asking ChatGPT for advice there. These are specialized solutions, trained on specific data sets. They flag suspicious patterns for human review which otherwise would go unnoticed.

Come on folks, this is a technology forum. We should be able to go beyond knee-jerk responses to trigger words.

Are you talking about real AI, or some very basic algorythm? In case of the second one, yes you are all right, and it is easy to doublecheck things.
But If it is about AI I think it is going to be very confuse what kind of pattern recognition there is, and on what level humans are intervening, if they are doublechecking things at all, which defies the systems purpose. REAL AI is complex and not just sorting out very easy stuff. And finding out how reliable a self modifiyng algorythm is, and doing that constantly, -as it is changin itself- will be almost impossible.

[Neutrion]

Today we can send millions of dollars around the globe (if someone has it), so we could make a system and democratically vote for decisions. But this is too democratic, so we replace puppets with robots...

The Guardian article is not about making policy decisions and passing new laws. It's about individual administrative decisions, e.g. which marriage is only pro forma for immigration reasons. And it's about screening even larger data sets for patterns which might indicate problematic constellations. You want to let people vote on this? 

Bit off here but,
People ARE voting not just on this but on 100 times as many and complex issues in every 4-5 years.
But there they are doing it well informed I suppose. So only on individual case level is everybody too stupid to decide anything, (and let it be decided for them by experts like Liz Truss or Ms Baerbock with extraordinary knowledge about the world, the universe and just about everything) on the grand scheme of things alltogether everybody understands everything....

And by the way to vote on something we dont't even need electricity so it is not about technical feasibility.

[themadhippy]

People ARE voting not just on this but on 100 times as many and complex issues in every 4-5 years.
But there they are doing it well informed I suppose. So only on individual case level is everybody too stupid to decide anything, (and let it be decided for them by experts like Liz Truss

The vast majority of uk electorate (about 99.7%) didn't get a say in whether dont trust became PM,and  soonsacked  got the job without any ballot  taking place.

[soldar]

I do not trust Artificial Intelligence in government.

I would rather be governed by Natural Stupidity like we have traditionally done for hundreds or thousands of years.

[coppice]

not sure which is worse,the old system using muppets or the new using robots.

A lot of things listed in that article are areas where I wouldn't put much faith in the average civil servant to make a well informed decision. The important questions about using AI instead of a human are what is their percentage of mistakes, and how severe they are? My guess with current technology is less mistakes, but quite a few mistakes being seriously egregious ones. When a pattern matching system gets a match wrong its errors can be hilarious, but if that error affects your life its not so funny.

[ebastler]

Are you talking about real AI, or some very basic algorythm? In case of the second one, yes you are all right, and it is easy to doublecheck things.
But If it is about AI I think it is going to be very confuse what kind of pattern recognition there is, and on what level humans are intervening, if they are doublechecking things at all, which defies the systems purpose. REAL AI is complex and not just sorting out very easy stuff. And finding out how reliable a self modifiyng algorythm is, and doing that constantly, -as it is changin itself- will be almost impossible.

Not sure how familiar you are with neural networks used for pattern recognition? They would certainly be considered "real AI" in my book, since neural networks are underlying the vast majority of today's AI solutions. 

And no, using them does not automatically imply that the algorithm keeps changing. It is very common to train a network, then freeze that state, test thoroughly and deploy.

Look, guys -- I am not by any means an expert on neural networks, but have worked with a team which has developed them for various targeted applications. I have not read a single post here which indicates familiarity with the technology; just gut responses and knee-jerk reactions. I don't think there is value to this discussion, so I'm out.

[TimFox]

The old data-processing term "garbage-in, garbage-out" still applies to pattern recognition on an ill-chosen but huge database.
I would like to see a detailed analysis of what went wrong in the well-publicized "hallucination" of non-existent and false legal cases cited (in proper format) in the infamous court case (May, 2023), which were somehow generated by ChatGPT but not verified by the naïve lawyers who used it as a search engine.

[coppice]

The old data-processing term "garbage-in, garbage-out" still applies to pattern recognition on an ill-chosen but huge database.
I would like to see a detailed analysis of what went wrong in the well-publicized "hallucination" of non-existent and false legal cases cited (in proper format) in the infamous court case (May, 2023), which were somehow generated by ChatGPT but not verified by the naïve lawyers who used it as a search engine.

I don't think ChatGPT is like the AIs currently being used by governments. Government interest today is mostly in sophisticated pattern matching, which is what the things in the Guardian article need. ChatGPT starts to get into trouble when it acts in a generative manner. Generative AI tries to take baby steps beyond pattern matching, and towards something we would consider intelligent behaviour. Like any infant it stumbles a lot. Current pure pattern matching is much more effective.

[TimFox]

I can easily see turning a competent AI system loose on a corporation's full data archive to generate a good annual report.
Similarly, turning it loose on the complete published works of English authors between Marlowe and Donne (including Shakespeare and the King James translation) could reveal useful results.
Recently, there was a semi-comic suggestion on this site that we should all post claptrap here in order to trick the omniverous data-skimming AI systems into producing nonsense.
(As can be seen in other threads, some posters apparently think this is a good idea.)

[jonovid]

the use of AI is definitely a massive fail for the North American goverment.
the decision making in the last week has been atrocious , in making enemies all over the world.
 

[TimFox]

North America has more than one government.
With the current chaos in the US Congress, there has been little, if any, decision making.
Luckily, the other governments in the world have made up for it during the current crises.

[SiliconWizard]

Luckily, the other governments in the world have made up for it during the current crises.

Is that so?

[mendip_discovery]

The news articles are trying to stir up stuff, no change there then.

Using AI to help summarise data might be handy. Do you really think that officials read all 501 pages of a report or do they ask a intern to read it and give them a summary. How does a intern differ from AI in accuracy? Even management don't read stuff they pay others to do it an report it back to them.

[TimFox]

Luckily, the other governments in the world have made up for it during the current crises.

Is that so?

No.

[TimFox]

The news articles are trying to stir up stuff, no change there then.

Using AI to help summarise data might be handy. Do you really think that officials read all 501 pages of a report or do they ask a intern to read it and give them a summary. How does a intern differ from AI in accuracy? Even management don't read stuff they pay others to do it an report it back to them.

I always assumed that's why the opening paragraphs are called "executive summary".

[coppice]

The news articles are trying to stir up stuff, no change there then.

Using AI to help summarise data might be handy. Do you really think that officials read all 501 pages of a report or do they ask a intern to read it and give them a summary. How does a intern differ from AI in accuracy? Even management don't read stuff they pay others to do it an report it back to them.

I always assumed that's why the opening paragraphs are called "executive summary".

Its a sad reflection of the times that executives apparently aren't smart enough to skip to the last page for a summary, so their one has to be placed at the front.

[TimFox]

I was taught that rather than end a report with "conclusions", that information should be moved to an "executive summary" at the beginning, since that's where the big wheels stopped reading.
(I learned that in an actual refresher training course on report writing.)

[Haenk]

Instead of AI, I'd rather like a simple coin toss.
It' *way' cheaper, faster and you only have a 50% chance of messing things up. That seems like a much better outcome than using standard "political decisions".

[mendip_discovery]

Just wondered if the uses could also be,

"If I make this change what are the potential reactions to this news from the press?"

"After this change what are the potential conflicts with other rules?"

"What are the knock on effects of these changes? Will things get better?"

"Take this dataset of facts and provide me talking points that are positive?"

"Using the data provided please come up with a ideas that could solve the problems we are having."

I know this might be s bit of a stretch for the current systems but I grew up watching Star Trek.