"footage made up of about 74 percent female and 26 percent male subjects;"
Why this makes me thing it's mostly porn?
Here's our latest summary of AI news beyond what we've already covered. It’s all about two favourite topics in machine learning today: facial recognition and deepfakes. Over 40 festivals pledge to not use facial recognition: A campaign against facial recognition led by the nonprofit Fight for the Future has led to over 40 …
This post has been deleted by its author
"The videos contains footage made up of about 74 percent female and 26 percent male subjects; and 68 percent Caucasian, 20 percent African-American, 9 percent east-Asian, and 3 percent south-Asian people. In order to train robust detection models, it’s important that the training dataset is diverse."
Diverse as in
: The majority of deepfakes are currently of Women + Porn hence 74 percent female. ( Although that will change as the technology becomes politically usefull).
: 9 percent east-Asian, and 3 percent south-Asian people. - What kind of percentages are these supposed to emulate, simulate...
: 68 Percent caucasian. So we are talking diversity in relation to the USA I presume, not as a representation of the world....
Shouldn't diversity relate to a global scale rather than a local scale.. The dataset could then filter down to local requirements as need be, rather the hoping that the local dataset could be extrapolated to a worldwide scale.
7 out of 10 are white people and you tout the diversity horn ? Who do you think you're kidding ?
If there was actual diversity, it should be 20% african, 10% arab, 10% asian, 10% indian, 10% chinese, 20% south american and 20% caucasian (I think that about covers it all). You can adjust the percentage points as you see fit, but 68% caucasian is a massive bias towards white people. Again.
And, given that white people are apparently rather easy to recognize, other ethnicities should be getting the lion's share of that database.
Being serious it's likely that as in the medical cases the AI will make different mistakes from the humans. A cancer detecting AI working on scans worked that way, different errors than the doctors tended to make. So they use both of them and together the mistakes are greatly reduced.
That is how AI should work, like all computing it should be an adjunct to human intelligence, not a replacement for it. Something that Japanese hotel with the robot staff should learn.
The old saw of garbage in garbage out seems to apply to AI and it is clear shortcuts have been taken the AI's are idiots as a result.
The real research into genuine AI instead of Big Data which is what this should be called came to the conclusion that a genuine AI will have to be taught much like a child is and as Larkin taught us this introduces its own biases. Even if get the first genuine AI's to teach the new ones that will just set the biases in concrete.
Did they check to see if the congress critters had family members in the 20k criminal pool?
What about the pro athletes? I have siblings that could pass for me. My cousins look a lot like my sister.
Out to just immediate family might be an interesting question for facial recognition.
I do see burkas being more popular in the future.
Doing that is 100% guaranteed to derail any career you might once have had as numberous Actors, Politicians and Sportsmen can testify even if it was 20+ years ago.
Facial recog or no facial recog you have to recognise Political Correctness when you see it.
Now if everyone did it it might be enough to make the various TLA's think again but somehow the facial recognition genie is out of the bottle and won't be going back despire various places outlawing it.
An alternative to Vantablack might be to paint your face in the same manner as they do with new concept vehicules, kinda zebra like, it supposedly stops cameras from being able to detect the exact form , shape ..
Biting the hand that feeds IT © 1998–2020