>LOL, so an AI isn't "allowed" to tell if a person is black, white or Asian?
Yes. I don't see what you find so funny about this.
Yes. I don't see what you find so funny about this.
Does it mean “The AI should not have any bias towards any race”
Or does it mean “The AI should not be able to distinguish whatsoever between two races”
To me, the phrasing above “AI should not be able to tell” sounds like the latter, which is drastically wrong and I suspect difficult to prove.
I think most people would agree it should not have bias, but it’s important that in the wording of the law we don’t eliminate functional stuff that’s important - ie being able to determine race - for example there are certain medical conditions that affect different races at different rates, so the ability to distinguish is an important factor in analysis.
A bio-metrics detector, detects biometrics! And the law basically bans bio metrics detectors from detecting biometrics.
It would be like if there was a law that said that restaurants can't sell food.
Or if hammer regulations prevented hammers from being used on nails.
Or that water could be used for anything but making things wet or drinking.
It is almost definitionally contradictory.
That is indeed funny.
Person A won't be identified because of being Chinese, he will be identified because of being a second generation immigrant that has a Chinese diet, interest in STEM and has a Chinese name.
The reason there are racial stereotypes is because there are trends among different races and skin color is a very evident trait for humans to identify. AI will draw the trends from everything else instead, which will draw the same picture, but skipping one specific parameter because it feels icky for humans.
Assume an increased chance of someone being interested into the arts and liberal because of dyed hair? Sure! Assume an increased chance of knowing Spanish because of being Hispanic? Ew no that's prejudiced.
It's so very stupid.
That is funny.
Your example? Not funny at all.
1) The overall point is to not let AI use racial bias as a factor in its decision-making.
2) Actually, there are a lot of people whose "race" is vague. My buddy has been asked if he's <random, different ethnicity/nationality> at least 5 times that I've noticed.
3) "Race" is a social construct anyways. In the US, in modern times, we delineate "races" in a specific manner, but these delineations have varied across history and cultures. See the one-drop rule (https://en.wikipedia.org/wiki/One-drop_rule).
Sex, race, height, build, hair, last known clothing
And if it is legal to collect and use that information then it seems nonsensical to ban the use of algorithms specifically...
And a HN entrepreneur will say that they do not collect the race, a neuron somewhere in the AI will turn on for a specific race but it is not their fault, the input data had no race the AI deduced it.
The laws regulate high risk stuff, it will prevent that HN entrepreneur to sell medical AIs , crime predicting AIs, car driving AIs etc just because they put a disclaimer that the AI might be wrong.
That's why drafting of laws is so tricky and important. You need to find the right way to make a general principle apply without having to spell every single possible case, which is impossible.
But on those things I am convinced that there is no need to single out AI if the law is drafted so as to spell put general principles.
On the other hand, AI can bring major benefits and, while we should be careful, we should allow "tech entrepreneurs" to bring those benefits to society.
When I read about this last time , the limits were n place only for risky domains. So AI for coding will not be affected.
Also the laws are not specific that the AI means neural networks, in my opinion from my experience tech is fulled with "fake it" people, with AI it is much easier for many more people to "fake it", so we can end up with this people selling crap to people and not taking responsibility for their actions.
I seen someone was working on a therapist AI, if he sells it as a service do you think that a simple ToS that says"the AI might be wrong, I take no responsibility" is enough? It is similar like Tesla AI, they also do not want to claim any responability and push the fault to the customer.
Even if it is legal to collect that information, it does not imply that it's legal (or desirable to begin with) to feed that information to a blackbox algorithm.
The GDPR use "process it", which is generic enough to cover AI.
Nope, it's not. Because, again, sending user data to a magic black box may be illegal. While processing it in other ways could be legal.
My point is that if collection and processing is regulated then that covers AI. And thus there is no "sending user data to a magic black box may be illegal" but my whole comment was that this wasn't needed. AI is just a tool, it's just automated processing.
Btw, 'nope' passes off as rather dismissive and rude.
No. I'm showing that your conclusions don't logically follow from the premise.
> My point is that if collection and processing is regulated then that covers AI.
No. No it doesn't.
> but my whole comment was that this wasn't needed. AI is just a tool, it's just automated processing.
And that, too, isn't entirely true, precisely because AI is a tool, and a different one. We always adapt or pass new laws and regulations for new, significantly different, tools.
For example, just because you can process data doesn't necessarily mean that you can process it in any way by any tools available to you. "GDPR allows you process data" does not immediately imply "this data can be processed by AIs". There is a reason why I was "adding additional constraints". Because these constraints are why additional laws and/or regulations around AI are needed.
For example, one of the stipulations of the AI Act is [1]: "AI systems shall be developed and used in a way that allows appropriate traceability and explainability, while making humans aware that they communicate or interact with an AI system as well as duly informing users of the capabilities and limitations of that AI system and affected persons about their rights". The reason being stories like this: https://www.politico.eu/article/dutch-scandal-serves-as-a-wa...
That stipulation alone isn't covered by any existing laws. There are bits and pieces here and there about various processes in various contexts, but this act brings them all together in one coherent form applicable to AI systems (which are appropriately broadly defined in the act [2]).
Or, there's an entire Article 5 explicitly prohibiting the use of AI in sensitive and outright harmful areas. Including things like banning them for large-scale suveillance except in very narrow scopes. Note that, again, there's undoubtedly pieces of this scattered around various regulations, but this one makes them clear and make them specifically applicable to AIs.
That's why I keep repeating: no, the fact that someone allows to collect and process data does not immediately imply that this collection and processing can be done with AIs. Where applicable, this law references existing ones.
[1] Quoting from this excellent article https://softwarecrisis.dev/letters/the-truth-about-the-eu-ac...
[2] Annex I: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52...
You’re right that the borders are not well defined though - it is a probably distribution - just like the weights in a neural network are also a probability distribution
You maybe could sort of define a group of people coming from the same locality and having the same history , and they would have maybe some degree of biological similarity as well, but there would be a high number of those groups and even then it would be quite useless and with number of exceptions.
You implied that in order for you to be proved wrong, the test must have 100% accuracy. I was saying that 100% accuracy is an impossible standard.
No tests in the world are 100% accurate, let alone in biology, the fact that you asked for this shows you are numerically illiterate
Here is DNA clustering that shows race: https://4.bp.blogspot.com/-t9Va5ai9-A4/Xd0cmvRBL4I/AAAAAAAAu...
Here is a link to the Kahn Academy course on statistics: https://www.khanacademy.org/math/statistics-probability If you take this course, you will understand probability better, and then you wont commit the logical fallacy of "false dichotemy" where you try to reduce complexity to binary classes which is a view that is too simplistic to accuratly represent the distributions we are talking about
That image does not contain 'race', and it comes from a paper that does not even mention the word 'race'. You have provided no arguments that race exists. The paper writes:
> This observation ... will make it a challenge to trace the ancestry of African Americans to specific ethnic groups in Africa, unless considerably more markers are used.
The more markers you use, the more different groups you get. You can of course choose to 'cluster' them, but then some precision will be lost, and what exactly will be gained?
The reason I mentioned cats and dogs is you can tell them apart by their DNA with 100% accuracy.
Yes, you can see what someone's DNA is, if you look at their DNA.