Slashdot Mirror
CEO of Facial Recognition Company Kairos Argues that the Technology's Bias and Capacity For Abuse Make It Too Dangerous For Use By Law Enforcement (techcrunch.com)
Brian Brackeen, chief executive officer of the facial recognition software developer Kairos, writes in an op-ed: Recent news of Amazon's engagement with law enforcement to provide facial recognition surveillance (branded "Rekognition"), along with the almost unbelievable news of China's use of the technology, means that the technology industry needs to address the darker, more offensive side of some of its more spectacular advancements. Facial recognition technologies, used in the identification of suspects, negatively affects people of color. To deny this fact would be a lie. And clearly, facial recognition-powered government surveillance is an extraordinary invasion of the privacy of all citizens -- and a slippery slope to losing control of our identities altogether.
There's really no "nice" way to acknowledge these things. I've been pretty clear about the potential dangers associated with current racial biases in face recognition, and open in my opposition to the use of the technology in law enforcement. [...] To be truly effective, the algorithms powering facial recognition software require a massive amount of information. The more images of people of color it sees, the more likely it is to properly identify them. The problem is, existing software has not been exposed to enough images of people of color to be confidently relied upon to identify them.
There's really no "nice" way to acknowledge these things. I've been pretty clear about the potential dangers associated with current racial biases in face recognition, and open in my opposition to the use of the technology in law enforcement. [...] To be truly effective, the algorithms powering facial recognition software require a massive amount of information. The more images of people of color it sees, the more likely it is to properly identify them. The problem is, existing software has not been exposed to enough images of people of color to be confidently relied upon to identify them.
made it will be and those in power will use it to expand and protect their power
Facial recognition technologies, used in the identification of suspects, negatively affects people of color.
Surely only if the suspect is a person of color.
Facial recognition technologies, used in the identification of suspects, negatively affects people of color
This statement is outright saying that black people are mostly criminals, the only case in which facial recognition identifying suspects "negatively affects people of color".
I think it's time we put an end to subtle racism like this, just because someone is black does not mean they are a criminal.
He's saying that the ML training dataset for people of color is too small. The machine needs to see more black people to identify them properly. This has nothing to do with black crime rate.
https://en.wikipedia.org/wiki/Race_and_crime_in_the_United_States#Crime_rate_statistics
Ugh, this racism has me SO TRIGGERED right now.
Well, the fact is that most people in jail are from the lower socio economic class and in the USA that means blacks. Therefore, there should be an enormous database of mug shots to scan.
...is a lie. Welcome to 2018.
I think his comment is more narrow - that the training sets used in facial recognition are smaller among ethnic minorities and therefore may have a higher rate of false positives ("they all look alike"). This could mean that minorities would be falsely targeted at a higher rate than the majority culture.
This, of course, is fixable - add larger, more diverse data sets and eventually the AIs will be just as good (or bad) at their job regardless of vagaries of skin tone, face shape and the like. That leads to the second problem - a pervasive surveillance state. There are upsides to such a thing (crime becomes a lot easier to solve) but the downsides are huge and obvious too.
Time for the devil's advocate point of view: This technology can save lives, so having this, as well as the Chinese detection of emotions on faces.
Scenario 1: Some student deciding to grab their dad's assault weapon, and go rack up some points. Camera detects their oddball emotions, resource officer does questioning, and they are stopped from doing this. Lives saved: Tens to hundreds.
Scenario 2: Some person looking for a weak target. The recognition cameras detect them, police nab them. Lives saved: Lots.
Scenario 3: Someone evicted from an apartment for a domestic comes back. Cameras detect them, police nab them. Lives saved: A mother and her family.
We -need- this technology NOW. It will do more to save lives than anything else we have.
but does it give more false positives ? It seems like a %accuracy calculation would be most appropriate with any computer identification system, so the humans on the ground can behave differently with a 85% likelihood vs a 99% likelihood vs a 20% are these systems not able to generate that kind of data?
âoeTolerance applies only to persons, but never to truth. Intolerance applies only to truth, but never to persons.
You guys created it to help sell things, obviously. That tech, once created, can also be used for law enforcement, in the "good" countries, and suppression and oppression, in the "bad" countries.
Maybe next time think before you tech.
Hire a Linux system administrator, systems engineer,
it doesn't work.
Mostly random stuff.
Seems to work well over there. As it is, it is a tool and any defense lawyer worth his salt will recognize that they better have more than that.
I'm afraid you are going to have to show your work here.
The problem is, existing software has not been exposed to enough images of people of color to be confidently relied upon to identify them.
Are you sure? And if so, why hasn't it?
This isn't the 1960s. Who exactly is biasing facial image databases, in 2018? Noted hotbeds of racism like universities and tech companies? How are they doing so?
This kind of bias can happen if the recognition software has trouble with darker skin tones (lower light information, less contrast? Who knows), or if there is some other external bias that causes poverty and a lack of social mobility to affect certain ethnic groups disproportionately and thus creates more crime among those groups (a societal issue).
It can also happen if your software has been tuned to recognize features of one ethnic group but not the variations in another, and then trained without further tuning, thus flattening the group out into broad categories instead of individuals.
Support my political activism on Patreon.
Well, why not put a few million faces of each race or whatever it's called now into your training dataset? I'm sure there are underground datasets that exist.
But are the mug shots of sufficient quality for this system? IIUC they will all be from the same angle (or pair of angles) and against the same background.
There's also the question of whether it is used in learning mode, and many AI programs do learning and use in two separate modes, and they don't learn from data encountered in use mode.
It still seems to me as if he's understating the problems with the system...but I only read the summary.
That said, even with reliable systems the police have a history of jimmying them to come up with the answers they desire, even if it means counterfeiting evidence. It's not clear to me how often they are complicit in labs producing false evidence. (That often seems to be a combination of laziness and carelessness, combined with no repercussions for fraud.)
I think we've pushed this "anyone can grow up to be president" thing too far.
simply so he could benefit personally. This fucker should be hanged.
The technology doesn't routinely make judgement calls that are inaccurate in a specific direction. It is however much less accurate but lack of accuracy does not mean bias.
Second, it is the policies around how it is used that negatively affect non-white people. This is a policy problem not a technology problem. I'm really not keen on being tracked and scanned by facial recognition or any of the other ways organizations track me but please don't exaggerate and play the racism card just to get clicks. In the end it numbs us to real abuse.
Facial recognition technologies, used in the identification of suspects, negatively affects people of color
This statement is outright saying that black people are mostly criminals, the only case in which facial recognition identifying suspects "negatively affects people of color".
I think it's time we put an end to subtle racism like this, just because someone is black does not mean they are a criminal.
Reality has a racist bias.
"the technology industry needs to address the darker, more offensive side of some of its more spectacular advancements. Facial recognition technologies, used in the identification of suspects, negatively affects people of color. To deny this fact would be a lie."
Is that humanity in the name of safety is voluntarily putting itself in a self imposed prison. No longer is the individual entrusted with the ability and power to fend for himself and look after his own safety. Instead we delegate this primary responsibility to big brother in law enforcement.
Since we can not trust law enforcement to protect us fairly and impartially, we are further delegating this primary responsibility to machines.
Whether or not the SJW's think machines have an inherent prejudice against minorities, is side skirting the problem. The problem is that you can no longer legally protect yourself. You have to let machines monitor humanity for evil doers. This is straight up distopia. In a free society if you get robbed you handle your business and go after the robber Chuck Norris style. In a enslaved subservient society you need to constant protection of the state to provide for you safety and needs. The individual is impotent.
Come on guys..
If msmash is posting this, ZERO CREDIBILITY
people are pissed because they thought there was a way to "pass the buck" no only with regard to the task, but to the responsibility as well..
Lazy Cops are in a uproar everywhere. Minority report theories abroad.
Admit it, the tech sucks, the people behind it also suck (as things develop).
the Process behind it sucks
and now people are complaining it dont work when, based on the info above, how can it??
Garbage in, Garbage out, Distill, shake, and produce msmash.
When was the last time any one went to Valley Fair Mall and Looked up at the entry points, see that mega camera with the Multiple occulars??? Hmm, look familiar??
STOOOOOOOOOoooooooooPiD
If a system consistently gave out a 20% likely figure, the people using the system would want their money back. So, no the system isn't able to generate that information. Make it the fault of the human and not the system to sell the system.
Nathan
This, of course, is fixable - add larger, more diverse data sets and eventually the AIs will be just as good (or bad) at their job regardless of vagaries of skin tone, face shape and the like.
No, this does not follow. If you skew the input, you also skew the output. Training the AI with more images from, than the actual ratio of, people with specific characteristics introduces a bias.
To get an unbiased result, you need to train it with a larger overall and unfiltered data set that is big enough to also get good results for the minorities.
He's saying that the ML training dataset for people of color is too small.
The complaint is about facial recognition *technologies* affecting people of color. Not about a specific application of facial recognition with a particular dataset. I really doubt the person who wrote the phrase I quoted would be placated by a better training dataset, in fact I am pretty sure they would claim it was racist to include too many black faces in a training set used to recognize criminals.
There is not one dataset for every use of facial recognition. Therefore he is not talking about training data sets, but about the *people* facial recognition technology is used to recognize.
Stop trying to cover for racists by telling us what they really meant was something else they plainly did not.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
Scenario 4: There are a few hundred thousand people who will trigger the detection routines over and over again, because that's just how they look. So they get apprehended and arrested over and over again, and are unable to lead normal lives.
No, thanks, we do not need this.
And the common man can't use this tech? Maybe you should look at what's happening to cell-phones for a start?
The police have always used facial recognition--both the police recognizing criminals from previous knowledge and mugshots and witnesses recognizing criminals--from the criminals who attacked them to photos they see on TV.
This recognition has always had inaccuracy problems--and a lot of people have wrongly suffered. The (partial) solution has always been to use it in conjunction with other evidence.
So there is no basic difference from facial recognition from software vs facial recognition by people--and with time the software recognition will be much better.
EOM
love is just extroverted narcissism
and a slippery slope to losing control of our identities altogether.
What does this mean? Our identities are dependent of our ID papers and other official documents issued by the government and private parties such as banks and their databases storing our biometric information ultimately. Identity matters to our relationship with other people and organizations. Facial recognition enforces our identities. That's the whole point of it. We already know what happens to people without or with uncertain identities.
Devil's advocate says that if you arrest every 10th person you meet, you will probably advert a lot of crimes, if you just do it often enough. 99.99% of the people you arrest won't be on their way to a crime though.
...for the Pre-Cogs to show up in the news and then we are in deep sh*t....
You're messin' with my Zen Thing, man.....
Scenario 4: There are a few hundred thousand people who will trigger the detection routines over and over again, because that's just how they look. So they get apprehended and arrested over and over again, and are unable to lead normal lives.
No, thanks, we do not need this.
Stuff like this already happens if you happen to get stuck on the TSA no-fly list thru no fault of your own.....
You're messin' with my Zen Thing, man.....
They were told who to vote for and chose neo-con alt-right Trump. The FBI should invstigate these people forever, or until something sticks. That's the only way to ensure the super delegate's mandates aren't defied again. Heil Hitlary!
> Reality has a racist bias.
Correct. Reminds me of the complaints about the expert system used by courts in CA for sentencing that recommended longer sentences for blacks. It was based on actual data on recidivism rates. Is it racist when it is based on actual facts?
You should try direct democracy if you're fed-up with the people in power constantly tyring to screw you.
It might help when you the people ARE those in power.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
The technology itself isn't responsible or culpable. It's the people who set the parameters and decide on what actions to take that are responsible and culpable. We simply have to find people who aren't bigots to set the parameters (or train the machine learning) and decide on what appropriate, proportionate actions should be taken.
Meanwhile, in the real-world, wouldn't it be great if the people in power and those who serve their needs and who are ultimately responsible for creating cultures of bigotry weren't bigots and selfish assholes? Then we might get systems and technologies that serve the people's best interests.
Debate is a form of harassment. Do not question my truth.
If the courts applied those stats to individual people, then yes, that is racist.
Punishing someone because of their skin color is racist. Just because the machine says blacks commit more, doesn't mean this black person in question will commit more crimes.
Besides, everyone knows jail and prison is con college. You go there to get smarter on how to commit crimes without being caught.
What if the AI is actually telling us something useful? Maybe there is a certain set of facial characteristics that in combination predispose one to criminal activity. Testosterone for example obviously impacts facial characteristics.
White, people are more likely to molest children. Maybe if white people stopped molesting children, we'd stop arresting them for molesting children.
????
Yea that's how you sound.
https://en.wikipedia.org/wiki/Ted_Kaczynski
Darker colors absorb more light and diminish depth perception and contrast.
Facial recognition depends partly on depth perception and contrast, both in humans and machines.
It's not racist, it's Physics.
Not really. I think it is something like 5% of criminals commit all the crimes. You'd never be able to have a wide enough net without basically arresting everyone.
He's saying that the ML training dataset for people of color is too small. The machine needs to see more black people to identify them properly. This has nothing to do with black crime rate.
Wouldn't a better solution than saying it's inherently racist be to send people to largely black, Hispanic and Asian communities or countries to take pictures and gather larger data sets to train the ai with? It seems more like it's an economic issue wherein this ceo doesn't want to spend the money to improve the inputs to the system to improve it.
This is going to go to a really dark place. A choice is coming. We as a society have to choose, do we want a world with crime, or a world without freedom? There is no room for both to co-exist. We can have a colourful world with choice and the crime that comes with that. Or we can be controlled, every negative thought known to the government, dissent suppressed and control handed over to a few elite who are already in power to do with as they please. China shows that this control will not be in the hands of a benevolent or even democratic government.
To use the sentiments of the currenlaw enforcement of the US, at the highest levels: "It's public info. Why shouldn't we be able to use it?"
Why shouldn't they be able to use public info to build an automated panopticon to track definitively where everyone is at all times?
Answer: Because that is a dictator's wet dream. Tracking phone "metadata" without a warrant would trivially allowed The Tyrant King George to round up all the founding fathers.
Facial recognition live tracking, license plate live tracking, all feeding into the computerized panopticon tracker, well, China and Russia are well on their way to "Imagine a boot stepping on a human face...forever."
US constitutional design orients around not building these things to begin with, and certainly not using them without a warrant.
The 4th Amendment forbids the king from filching through your papers not because "you have nothing to hide", but rather most people, especially the powerful challengers to power, might very well have broken a law here or there.
Government does not get to seek out those violations because you are a politically uppity citizen.
We've had two disgusting displays by the past two presidents. Trump and his jail Hillary stuff, and Obama, who, when S&P downgraded the US' credit rating in response to $1.4 trillion a year borrowing, ordered the SEC to look into them, amd announced it publically. And they found some dirt and prosecuted...this politically uppity private group who dared.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
There are more poor white people than poor black people. The ratios of poor to middle class to rich might differ, but trying to claim that poor = black is false. The poor black population is likely more concentrated in urban areas where crime rates are also higher. Whereas poor whites are likely more concentrated in more rural areas. The crime rates likely correlate to higher costs of living in cities as well. More desperation leads to more extreme measures. Crime also tends to be lower in smaller areas where it's more likely that most people know or recognize most other people. Then there is gang crime and violence which is a completely different circumstance.
at least here in the states where everything is based on screwing everyone else, is there a point to trying to keep it out of law enforcement hands? it will just get subcontracted out 5 layers deep so they can claim they arenâ(TM)t using it while they use it.
It depends on what the problem is. If darker skin tones make recognizing various features more difficult, a disproportional data set does not necessarily mean skewed results. it might help an AI discover other features for unique identification. Also, if you're looking at a global populace, the race numbers change drastically from looking at it from a US-centric perspective.
Assuming results would skew is stupid. If the system is giving more false responses for blacks, increase the data set for blacks. If that added data decreases the accuracy of white matches, investigate why. Perhaps the solution at that point creates branches dependent on skin tone that uses different facial features as criteria. Some traits might not be a one size fits all method.
Crime statistics validate GP's claim - and invalidate yours.
In other words, the machine has a racist bias because it would only catch white people. Is the idea canned because it is racist or because it would benefit people of color?
Precog: I know what you are going to do.
Me: Good we are on the same page, while I mine and cook cracked stone bricks, you start leveling the foundation for the castle and start on the wall.
>White, people are more likely to molest children
I can't find any numbers which back up this statement, child abuse and rape stats suggest blacks are 200-500% more likely to be offenders - that doesn't factor in the lower reporting rates in the black community. Feel free to provide sources.
The 4th Amendment forbids the king from filching through your papers not because "you have nothing to hide", but rather most people, especially the powerful challengers to power, might very well have broken a law here or there.
Nonsense. Warrants are required in order to protect the innocent. And protecting the innocent is more important than catching all of the guilty.
scenario 5: people of color commit more crimes, therefore they have a higher probability of being mistaken for another person of color
city A:
population A: 93%
pop. B: 7%
pop. B commits 85% of the crime
having a smaller population and also a larger list of suspects, it's only logical the system would be bias towards detection,
now what needs to be looked at is not the totals, but rather the typical rate of misidentification on a per person basis within all groups
The only problem with your argument is that one is based on facts backed up by data and the other is not.
Assuming your unsupported premise (and I have no idea if it's true) that "white people are more likely to molest children" it would be entirely reasonable to expect more white child molesters would be arrested. In the same way, people who have a propensity for shooting other people are much more likely to be arrested for shooting other people. I fail to see where you meant to go with this.
Did you have a point to make or are you just trying to troll?
But they don't commit crimes all the time.
> Well, the fact is that most people in jail are from the lower socio economic class and in the USA that means blacks.
No it doesn't. Quit being such a racist.
A Pirate and a Puritan look the same on a balance sheet.
No. The underlying problem has nothing to do with race. The real problem is that you are trying to punish people based on future crimes. This is the problem of letting your brain rot because you can't do anything but play the race card.
A Pirate and a Puritan look the same on a balance sheet.
> Why shouldn't they be able to use public info to build an automated panopticon to track definitively where everyone is at all times?
Why shouldn't anyone? This is a basic liberty issue. What isn't explicitly forbidden is allowed in a free society. Anyone can do it. I could probably cobble something together myself. That's just the nature of technology in a sophisticated society.
You are whining about the wrong part of the equation.
It's the panopticon that's the problem.
Data that might make it more useful is just a red herring.
A Pirate and a Puritan look the same on a balance sheet.
There are clearly different levels of crime and different levels of safety. Why fight a strawman.
Learn.
http://federalsafetynet.com/us-poverty-statistics.html
"While the poverty rate for the population as a whole is 12.7% the rate varies greatly by race. Blacks have the highest poverty rate at 22.0% and Non-Hispanic whites the lowest at 8.8%.
The Poverty rate for Blacks and Hispanics is more than double that of non-Hispanic Whites."
Quit living in the clouds. Reality exists and we can measure it.
People of color (which color?) don't commit more crimes, they get arrested for more crimes.
Anecdote time. I live in Minnesota, not all that far from where Philando Castile was shot. I was driving about 10 mph too fast on a local street, and I got pulled over.
Instead of gunning me down, the officer just admonished me to slow down and let me go. No ticket, no arrest, just a gentle reminder about speed. Colored people often don't get the same forbearance.
As opposed to all the totally ethical things it will be used for by the private sector?
People of color don't commit more crimes, they get convicted of more crimes. It goes at all levels: Citizens in general are more likely to report crimes by colored people. Police are more likely to arrest colored people vs. give warnings to white people for misdemeanors. Courts are more likely to convict people of color, and they hand down longer sentences to people of color.
You've assumed ML is likely to ignore people of color due to their misrepresentation in the dataset and would therefore mostly catch whitea.
We don't know that. Maybe it is more likely to ignore, maybe it is more likely to false-positive, and maybe it's a fluctuation of inaccuracy.
If a system consistently gave out a 20% likely figure, the people using the system would want their money back. So, no the system isn't able to generate that information. Make it the fault of the human and not the system to sell the system.
Nathan
Nobody complains about false positives with drug dog searches. Why would they care about false positives with facial recognition? Just do not record them and the problem is solved.
Devil's advocate says that if you arrest every 10th person you meet, you will probably advert a lot of crimes, if you just do it often enough. 99.99% of the people you arrest won't be on their way to a crime though.
But how do you know that if you do not arrest them?