Contaminated' AI databases are taken down after UCD work

Calahonda52

https://www.independent.ie/business/irish/contaminated-ai-databases-are-taken-down-after-ucd-work-39365275.html

"..the MIT database contained thousands of images labelled with racist and misogynistic insults and derogatory terms."

Certainly helps to explain a lot of what is happening in the UK with their surveillance state.
.
Had to look up what the word meant.

Definition of physiognomy
1 : the art of discovering temperament and character from outward appearance
2 : the facial features held to show qualities of mind or character by their configuration or expression

Quantum Erasure

Is this like the AI twitter bot that turned Nazi

Ubbquittious

Statesians at it again with their AI overlords and their racism.

Charles Babbage

Here is an example

BIFFO

Aongus Von Bismarck

Life 3.0: Being Human in the Age of Artificial Intelligence by Max Tegmark is a wonderful introduction for the layman into AI, and the impacts it will have on humanity.

Deals with some of the practical matters such as the one being discussed here, as well as the more philosophical and moral questions that surround AI.

Capt'n Midnight

Detroit cops employed facial recognition algos that only misidentifies suspects 96 per cent of the time


On two occasions I have been asked, "Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?" ... I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.
— Charles Babbage, Passages from the Life of a Philosopher 1864

1864. And look how far we haven't come since then :rolleyes:

topper75

Help a lay guy understand here please.
So some humans have maliciously labelled these images at some point, somewhere? And computers ingesting that massive dataset are taking the associated text at face value (no pun intended)? And algorithms are producing mistaken and unjust outcomes as a result?
Talk me through how this might hurt me in an Irish 2020 context so I can understand ...

jimgoose

So Skynet is a Good Ol' Boy. I knew it!! :pac:

riffmongous

topper75 wrote: »

Help a lay guy understand here please.
So some humans have maliciously labelled these images at some point, somewhere? And computers ingesting that massive dataset are taking the associated text at face value (no pun intended)? And algorithms are producing mistaken and unjust outcomes as a result?
Talk me through how this might hurt me in an Irish 2020 context so I can understand ...

Think of it like this, in 2020 there is a child somewhere being taught by its racist parents that Irish people are scum, in 15 years from this child will be the HR assistant looking at your profile when you apply for a job, the cops when you are near where a crime was committed, etc

topper75

riffmongous wrote: »

Think of it like this, in 2020 there is a child somewhere being taught by its racist parents that Irish people are scum, in 15 years from this child will be the HR assistant looking at your profile when you apply for a job, the cops when you are near where a crime was committed, etc

That scenario is as old as time itself. You left out where the AI comes in and why I or indeed anybody ought to worry about that in particular.

Sleepy

I'd be very curious to see some examples of the images and labels deemed problematic. It really wouldn't surprise me to find that a modern academic's judgement of what constitutes "racist and misogynistic insults and derogatory terms" would be entirely at odds with reality. I'm not saying that's the case in this instance, but it wouldn't surprise me one bit if it was.

Deleting the dataset seems very extreme too, surely a cleansing operation would be worthwhile? It could hardly be that difficult to delete any of the problematic labels from the dataset?

is_that_so

Aongus Von Bismarck wrote: »

Life 3.0: Being Human in the Age of Artificial Intelligence by Max Tegmark is a wonderful introduction for the layman into AI, and the impacts it will have on humanity.

Deals with some of the practical matters such as the one being discussed here, as well as the more philosophical and moral questions that surround AI.

They could then also look at Rebooting AI by Gary Marcus & Ernest Davis to see how dumb a lot of current AI is. Not quite a layperson's book but an interesting read.

Dravokivich

Sleepy wrote: »

I'd be very curious to see some examples of the images and labels deemed problematic. It really wouldn't surprise me to find that a modern academic's judgement of what constitutes "racist and misogynistic insults and derogatory terms" would be entirely at odds with reality. I'm not saying that's the case in this instance, but it wouldn't surprise me one bit if it was.

Deleting the dataset seems very extreme too, surely a cleansing operation would be worthwhile? It could hardly be that difficult to delete any of the problematic labels from the dataset?

I would expect the intention would be purely on a cataloging basis, as opposed to systemic racial profiling. Such as how searching for an item on Google would work. Racial based comments, will give racial based results, is Google racist?

is_that_so wrote: »

They could then also look at Rebooting AI by Gary Marcus & Ernest Davis to see how dumb a lot of current AI is. Not quite a layperson's book but an interesting read.

I recon the problem is more so in the name. Lay people put too much emphasise on "Intelligence," when others talk about AI.

4068ac1elhodqr

Capt'n Midnight wrote: »

Detroit cops employed facial recognition algos that only misidentifies suspects 96 per cent of the time

The high error rate relates mainly to people of colour, wonder if this might result in a get out of jail card? Poor whiteys don't have such error rates, so their convictions could stand if challenged based on error rates.

In a similar theme to AI bots looking at faces, Clearview AI has billions of faces (social media scraped) already on it's db.

Chances are if you were to combine this with say GoogleGlasses wearables, then take a walk down any street in the world, everyone's profile (perhaps extensive details) would spring up on screen, just like in de Terminator movie.

The end of privacy:
https://www.nytimes.com/2020/01/18/technology/clearview-privacy-facial-recognition.html

Already in use (by hundreds of agencies), and also under investigation:
https://www.computerweekly.com/news/252485913/Clearview-AI-faces-ICO-investigation-over-facial-recognition
https://www.ft.com/content/851560ab-0078-41ae-ad25-c2d6b154921f

Sleepy

Dravokivich wrote: »

I would expect the intention would be purely on a cataloging basis, as opposed to systemic racial profiling. Such as how searching for an item on Google would work. Racial based comments, will give racial based results, is Google racist?

I was more focused on the "misogynist" part tbh as depending which branch of feminist you're speaking to, recognising a trans woman as being female, being anti-abortion, being in favour of legalisation of prostitution, knowing that the gender pay-gap is nonsense, being sex-positive or simply being a straight white man could all be considered as "misogynist".

riffmongous

topper75 wrote: »

That scenario is as old as time itself. You left out where the AI comes in and why I or indeed anybody ought to worry about that in particular.

I thought it was clear, the AI is what will be used for scanning your CV, for identifying you as a suspect through CTV facial recognition

topper75

riffmongous wrote: »

I thought it was clear, the AI is what will be used for scanning your CV, for identifying you as a suspect through CTV facial recognition

But you referred to a HR assistant and cops. I took these to be humans. What do you mean by scanning the CV? I mean if I understand rightly, recruitment agencies have been putting keywords into databases for almost two decades. I have got several jobs this way (without any photo on my CV as I'm an uggo and aim to continue in this regard)!

I'm still to be convinced that I have anything to fear. I still think I have read a non-story.

riffmongous

topper75 wrote: »

But you referred to a HR assistant and cops. I took these to be humans. What do you mean by scanning the CV? I mean if I understand rightly, recruitment agencies have been putting keywords into databases for almost two decades. I have got several jobs this way (without any photo on my CV as I'm an uggo and aim to continue in this regard)!

I'm still to be convinced that I have anything to fear. I still think I have read a non-story.

Ups, I thought the analogy was clear, considering the thread we are in and the link in the OP, it's even mentioned there about the potential uses for the AI.

Well if you know about keyword databases already can't you see why it would be bad if the database was corrupt? Also a CV without a photo is fine as long it's accepted, which can change. It's pretty much mandatory in Central Europe and companies won't entertain a CV without one (which I really don't agree with either).