Stupid AI, stupid policy or stupid world?

[CMCanavessi]

I just saw this: https://www.dailymail.co.uk/sciencetech ... ategy.html

I have mixed feelings of what the world (and society in general) is becoming (or has already become)...

[towforce]

It was clearly a mistake. On the plus side, good publicity for Agadmator's channel!

[purechess]

Stupid Google bots

[mvanthoor]

That's nothing. In the Netherlands, the entire government has fallen because our tax department has gone after thousands of people that where displaying fraudulent behavior... and one of the main reasons for that decision was that "the computer said so."

In short: we have a system AI system, which tries to determine if people are fraudulent (tax evasion, getting child support while not having children, that sort of thing.) If this system says that you're fraudulent ("the computer says so"), it's not yet proven... but our tax department thought it to be enough to just cut people off from certain allowances and start investigating. If you want to know more about it, search for "Toeslagenaffaire". It landed thousands of people in financial problems.

The government has fallen over this, the system has been put on hold for now (as far as I know) but we don't seem to learn... there are already plans for "Version 2.0" that would need to "fix" the problems of the old system by including even more extensive data and deeper analysis of this data.

People, companies and entire governments are losing grip on the situation because AI's are trusted almost blindly. If the computer says so, it must be right.

However, many of those systems have simple flaws, and they find correlations where there are none. For example (I am making up the numbers for illustrative purposes here):

- Our crime statistics say that 52% of people in jail are from Moroccan descent.
- 65% of people who are in prison, have fraudulent behavior in their past.
- People who are in jail are criminals.
- Criminals get into jail by prosecuting and convicting them.
- Person X living at Amsterdam Street 10 is of Moroccan descent.

There you go.

Because Person X is a Moroccan guy, there is a high chance he'll end up in prison at some point (because the statistics say so), and thus, there's a large certainty that he will have something fraudulent in his past. Because of this, he is probably a criminal, and thus should thus be prosecuted (according to the computer). Therefore, lets cancel all of his allowances (child support, rent, etc), and start an investigation, assuming he is indeed fraudulent, a criminal, should be prosecuted, and be in jail in jail ...

... only to find out that Person X is not fraudulent, not a criminal, and he's just a regular dude who happens to be of Moroccan descent, working a minimum wage job in the factory, and thus entitled to the allowances he received.

Oops. Err... sorry dude.

But now Person X is in trouble, has a 20K debt, lives like half a fugitive and his marriage is ****, where before, he led a simple, but respectable life.

That's AI for you.

As said: Search for "Toeslagenaffaire", that goes _somewhat_ beyond "Youtube blocked my video." There are AI's in this world, right now, that are not only annoying if something goes wrong, but that are able to wreck your entire life because people in power trust the things the AI says and act on it, without actually knowing how or why (or worse yet, maybe not even caring how or why) the AI came to the conclusions it did.

IMHO, AI's are fine for playing games. They're fine for finding the perfectly optimal, contorted an twisted antenna for receiving signals, or for testing and finding the best combination of chemicals to make a powerful, but non-aggressive cleaning component that tastes like lemonade and can be drunken by toddlers without ill effects... but they shouldn't to be used to make come to conclusions about people's lives and/or supposed behaviors of which you have no proof that they actually happened.

[carldaman]

Unleashing the monstrous AI on people like that is something that not even Orwell could have dreamed up in his worst nightmares. I thought AI was going to be used to maybe solve some of the world's problems.

[hgm]

Well, the problem in the 'toeslagenaffaire' was not just the use of AI. It was also criminal abuse of power of the people in charge (who at some point decided, when pressured from above to clean up the administrative mess they were in before a deadline, decided that instead of actually figuring out who of the people having a tax debt had a legitimate reason, to just declare everyone with a debt of over 3000 euro guilty of tax fraud was a great time saver). Plus total failure of the appeal courts for people-vs-the-state affairs, which almost by default ruled in favor of the tax service, motivated by the argument that the tax service was a mighty government organization, and the person appealing their decision just a single civilian, so that the tax service must be right.

[towforce]

Unleashing the monstrous AI on people like that is something that not even Orwell could have dreamed up in his worst nightmares. I thought AI was going to be used to maybe solve some of the world's problems.

Here's the simple version: artificial intelligence is still intelligence - and most intelligence is guesswork.

Here's the longer version:

There are three ways to reason:

1. deductive reasoning (Clyde is an elephant, all elephants are grey, so Clyde is grey)

2. abductive reasoning (Clyde is grey, elephants are grey, so Clyde is an elephant)

3. inductive reasoning (Clyde is an elephant and Clyde is grey, Marcia's an elephant and Marcia's grey, so all elephants are grey)

The overwhelming majority of reasoning that people do on a daily basis is abductive reasoning, which is not guaranteed to give the correct answer even if the premises are correct.

[MikeGL]

In one football match at Europe, I remember that blooper when an A.I. camera focused on bald head of a referee instead of the ball.
Became one of the trending youtube vid. Not sure if I picked up the news here on talkchess or other news sites online.

[Guenther]

In one football match at Europe, I remember that blooper when an A.I. camera focused on bald head of a referee instead of the ball.
Became one of the trending youtube vid. Not sure if I picked up the news here on talkchess or other news sites online.

Also seen this sometimes when dart players suddenly choose an unusual different path and the camera could not find the dart hitting the board ;-)
(not sure though if this is ai driven too, probably yes, because it needs very quick reactions?)