How long before we find out it always flags certain makes and models as criminals?
it will just hand out random DUIs to RAM owners based on statistics and reckless driving tickets to black altimas with tinted windows and bedazzled plate frames.
That would be great.
Imagine an AI that’s morally blind, pulling over expensive cars because the owners are more likely to break traffic laws knowing they can afford the fine.
Or an AI that's pulling over cheap and old cars because the owners are more likely to get ticketed due to living in over-policed neighbourhoods.
do 'the poor' all drive really bad?
We've already got numerous examples of how these ai models and face recognition models tend to have biases or are fed data that accidentally has a racial bias. Its not a stretch of the imagination to see how this can go wrong.
Yep, the age old "garbage in garbage out". If we had a perfect track record we could just send in all the cop data, but we know for a fact the poor and PoC are stopped more than others. You send that into AI it will learn those same biases
No, but any automated system can be used to punish people who cannot afford to fight it.
Well to be fair, this is because of the stupid justice system in the US.
Just the term "afford to fight for it" is something that never should exist in a civilized society.
I agree, but can we not with the r word?
Yes I can edit it, didn't mean to offend.
No, but they're disproportionately affected by fines. For rich people a fine is just the cost of a privilege.
Exception being something like Finnish speeding fines which are income dependent.
Not that I’m aware of.
But we know the criminal justice system currently has biases. If the data the “AI” is trained on was affected by these biases, or others that we don’t realize, then it will produce biased results.
One of the worst parts about all this to me is that the AI and the dataset used to trained it are kept secret as proprietary information, and the police and governments buy it anyway despite that nobody can even try to check the code or dataset to see what biases or errors it might have (and definitely does).
Great! Automated discrimination against the poor
How long before we find out it always flags certain makes and models as criminals?
it will just hand out random DUIs to RAM owners based on statistics and reckless driving tickets to black altimas with tinted windows and bedazzled plate frames.
That would be great.
Imagine an AI that’s morally blind, pulling over expensive cars because the owners are more likely to break traffic laws knowing they can afford the fine.
Or an AI that's pulling over cheap and old cars because the owners are more likely to get ticketed due to living in over-policed neighbourhoods.
do 'the poor' all drive really bad?
We've already got numerous examples of how these ai models and face recognition models tend to have biases or are fed data that accidentally has a racial bias. Its not a stretch of the imagination to see how this can go wrong.
Yep, the age old "garbage in garbage out". If we had a perfect track record we could just send in all the cop data, but we know for a fact the poor and PoC are stopped more than others. You send that into AI it will learn those same biases
No, but any automated system can be used to punish people who cannot afford to fight it.
Well to be fair, this is because of the stupid justice system in the US.
Just the term "afford to fight for it" is something that never should exist in a civilized society.
I agree, but can we not with the r word?
Yes I can edit it, didn't mean to offend.
No, but they're disproportionately affected by fines. For rich people a fine is just the cost of a privilege.
Exception being something like Finnish speeding fines which are income dependent.
Not that I’m aware of.
But we know the criminal justice system currently has biases. If the data the “AI” is trained on was affected by these biases, or others that we don’t realize, then it will produce biased results.
One of the worst parts about all this to me is that the AI and the dataset used to trained it are kept secret as proprietary information, and the police and governments buy it anyway despite that nobody can even try to check the code or dataset to see what biases or errors it might have (and definitely does).