8 Years is a Long Time in Data Ethics

Data Ethics is the topic of the moment. If your organisation doesn't have an existing ethics framework or a concerted effort to produce one you are behind where you need to be.

Ethics has been built in as a component into the latest release of the EDM Council's Data Capability Assessment Model (DCAM) and was recently incorporated into the Danish Financial Statements Act. In Denmark you need to have a statement on your Data Ethics Policy else explain why you don't have one. There is a squeeze from both directions - industry best practice as well as government based regulation.

We have come a long way and this reminded me of an article I first read back in 2012. A New York Times article on how Target was using big data techniques to segment data sets and focus their marketing material to more accurately target their customer base at a granular level - Feb 16 2012 "How Companies Learn Your Secrets"

The full article is fascinating reading whilst at the same time the concepts raised in 2012 are borderline naive when read through the lens of 2020. At the time the thought that your data could be mined to this level of specificity was a new and scary idea. In the subsequent 8 years we have had the Cambridge Analytica scandal, we know that a product search on Google results in adverts shortly afterwards in your Instagram feed and we have full awareness that Netflix knows the difference between what we claim we watch (historical documentaries) versus what we actually watch (Married at First Sight).

Vance Packard's The Hidden Persuaders, a classic analysis of the advertising industry, was shocking when published in 1957 as it exposed selling and marketing techniques to an unknowing public. What was new then is very much common knowledge now. Data Analytics has been following much the same path in terms of awareness but presents a different challenge due to the narrow focus that can be applied to the individual consumer.

Much like the proverbial frog in the boiling pot of water, we have gradually reached a point where what we can do has outstretched our thinking on what we should do.

The Ick Factor

Back in 2012 the Ick Factor was already present. Most people are fine with being advertised items that they actually want. I find Instagram spectacularly good in this regard as my recent purchases of a map, dinosaur necklace and drone indicate.

Equally we don't like the feeling that someone knows "too much" about us - the Ick Factor.

The Ick Factor is where Data Decision Making and Data Ethics Meet

Many current Data Ethics efforts focus on the build out of AI models. The model assumptions, who builds the models and how the models are used all form part of an intricate decision making and review process. In talking to an industry colleague we agreed that the Wall Street Journal rule still holds true.

Would you be comfortable seeing your name included in an article published front page of the Journal?

In the same way that the Ick Factor makes you uncomfortable due to its specificity, far more harmful but less visible is lost opportunity. For example losing access to credit, capital or schooling because a part of your identity is now encoded in a feature set in an AI model and leads to an adverse outcome for you.

Example: People with green hair are bad at repaying loans. You have green hair. You are now less likely to get a loan because that is what the feature (hair color) tells the model. Is it anyone's business that you have green hair? Should they be using this information to decide if you get a loan?

Just because data can make your model more accurate, should you use it to make your model more accurate?

Where are we now?

Much of the industry debate of the past two years has been focused on privacy. The advent of GDPR followed by the California Consumer Privacy Act (CCPA) has led to a focus on data privacy and the rights of the individual. Easy soundbites like the "right to be forgotten" take the headlines. Now its about Data Ethics - you may be complying with the appropriate legislation but are you doing the right thing?

Thoughts on Implementation

I don't think you can set prescribed rules for Data Ethics. We cannot predict the unknown-unknowns that will come up in the future. What we can do is build a strong flexible operational framework that can flex to the fast paced rate of change.

As a starting point consider the following:

• A decision making framework with board-level commitment and accountability.
• An educated organisation knowing when to apply the framework (not just AI specialists).
• Decision making bodies composed of a diverse set of individuals that can act appropriately to manage standard and edge cases.

Success is that in each case where you have used data, you have used it in a considered manner...and yes, happy to see that usage on the front page of the Wall Street Journal.

______________________________

Andrew Foster is a Wall Street Executive with a background in large scale program delivery across London and New York. He specializes in building effective data teams in complex organizations. Andrew has a passion for mentoring, building employee resource groups and advising on how to work effectively in global teams.

This blog post represents personal views and does not necessarily represent the views of current or prior employers. Any advice provided should be acted on at your discretion and with acceptance of your own risk.