Engineered & Automated Biases

"White House is investigating Google after Donald Trump claims bias"

Bias can be a problem in all aspects of society - both in humans (conscious and unconscious) and in algorithms (engineered and learned). Whilst I think this is an area that warrants further research -particularly into algorithms- it concerns me when the state decides it wants to try and police bias.
Unconscious Bias in people and learned bias in algorithms is an extremely difficult thing to identify with any certainty due to the number of variables (along with the interpretation of context) that affect decision making.

And, even if you could develop a fault-proof way of detecting bias (in humans and algorithms) who gets to decide what biases need identifying; who decides what they get replaced with; who gets to decide what constitutes precisely the right amount of ‘fairness’ and across how many different topics, groups, mediums or areas of applications; etc....

Trump? Some of his staff? Unelected PHDs? Who do you want correcting your biases and those of society?

Algorithms, particularly those that are derived from data (deep learning) will always develop bias (like humans) because they look for emerging or strongly evident patterns in data and whether we like it or not, they shape the way we live (and have been for some time).

However, not all algorithms are created equally which is why I have mixed feelings about the blind application of algorithms to certain aspects of society... For a commercial enterprise, I am less concerned (because the market can weed out the good from the bad) but for matters of say law or civil rights, I am much more concerned. Bad algorithms (either through incompetence or design) automating large scale and damaging bias is a thought that shouldn't sit comfortable with anyone, yet there have already been some well-documented examples of this happening. One example of this is a when a large number of New York teachers had their performance interpreted by an algorithm and were consequently wrongly fired. What made matters worse, was that this algorithm was kept secret until sufficient lobbying exposed the flaws.
Even if you subject algorithms to human review, those humans or groups of humans (whatever layer of abstraction you go to) will also have their own biases that they employ in their decision making. It is a really really really complicated process.

I’m actually quite torn on this subject - can you tell?

My gut feel is that bias is an individual thing and best left alone at an individual level. You can not police peoples thoughts and whilst you can add different features and more data to computer models you can not manipulate the ‘BIG’ data itself to produce a specific algorithm.

As biases work their way up from the individual level and through the abstraction layers that make up society, I think we should let democratically elected officials figure out whether or not they should correct some of the negative effects that biases have had on marginalized people. I would also suggest that this should be done on an individual level, not a blanket based statistical level. It’s not perfect, and it will probably cost a lot of money to treat individuals as individuals (rather than a statistic) but I think it’s the only way to offer some fairness in this highly unfair existence that share... That's where I am at right now, but I am still wrestling with it...

One thing I am certain of here though is that I am very suspicious of Trump's motives.

What do you guys think?