The F.A.T. Fairness, Accountability and Transparency for Artificial Intelligence

Almost all the institutions are increasingly exploring and implementing much more “AI and machine learning” into their processes and offerings to customers and clients, much more than ever before in the past. And as a result, a new headache for the several policy makers, regulators trying to qualify if this “tech” is responsible enough per societal/govt norms and what could be the repercussion if the governance fails.

Most of us know that machine learning runs on powerful algorithmic model, that gets trained continuously using datasets from several data sources. An old saying of “Garbage in would results into Garbage out” which implies that the quality of these data sources is crucial for ML to be effective and address the concerns of regulators.

Let us understand the above a bit more with few examples, on lack of quality data sources impact to ML outcome:

·  There is a term “Unconscious Bias” that could arise which would result in egregious examples such as image data sets not facially recognizing “African Americans” and categorizing them as gorillas. Highly ineffective that would be, isn’t? Please read the article in the link, ML amplifies the Bias.

·  We live in a society which is full of “Malicious and Mischievous Actors” sometimes also known as “Hackers”, can also influence machine learning. A perfect real-life example would be Microsoft’s “Tay Chatbot”. It was a painful experience for Microsoft as the bot ended up developing racist and sexist rant and the source of learning was Twitter which was deliberately spammed with offensive trolls that was feeding into this bot.

Problem does not end here or is it so? You would agree we all are churning out huge amount of data and as these data sources become available to track both online behavior as well as offline activities through connected devices, technology companies are using ML to convert these behavioral data into individual profiles for predictive analytics (Google does it and so does Facebook).

All the RegTech and FinTech uses ML for important processes such as credit ratings, search engines, bank loans, university applications, home/health insurance etc. Take a note, use of ML is critical here and it leads to civil liability and loss of one's liberty.

The need to combat negative bias outcomes, unconscious or otherwise, remains an important source of concern. Few examples below:

·  How do we qualify that there was none unconscious racial biasness in the COMPAS algorithm for criminal sentencing risk assessments in the United States?

·  Do Facebook's algorithms have a confirmation bias that allows fake news to be targeted, and in turn did their postings impact the outcome of the Presidential elections?

Reiterating the point, AI/ML is only as strong and as good as the data that you serve it with. When you think from a regulation perspective, it's not the regulation of algorithm that matters much but the quality of algorithm, and the natural tendency of algorithm to create an oligopoly.

Remember folks, it is data that acts as a fuel to any intelligent algorithm, and an algorithm only grows to be constructively (destructively) effective by having better (bad) quality data. I have always believed and mentioned in my previous posts that companies are doing whatever it takes them to be able to capture all the data they can on their consumer to have a dominant market position. As data is naturally driving network effects, and because network effects are themselves very naturally oligopolistic.

The role of the regulators becomes too crucial in this “get all the data” game, as they would need to find a way of breaking away the dominance of large tech firms that are controlling most of the data. For example, Facebook controls mostly your social data, Amazon controls e-commerce data and professional data is most likely controlled by LinkedIn.

They have all the information about you and you should be worried if you are not already. The only way to break away that stranglehold on your personal data is by allowing you as an individual to share your data to other people, companies when and if you want to. It is your privacy and thus it should be your choice.

Imagine when these companies reach out to you for the data and you negotiate the monetary benefits with them before you let them have your data? You get paid for your data, is it possible? Yes, it is, and this is where data regulation will go in the future. It will focus on individual data as opposed to the algorithm adjustment that will be required to be made by technology companies as requested by the regulators.

 And the Activism that can make this happen is “Blockchain”, a dream to be fulfilled for regulators and customers.