Much like deeplyflawed police profiling tools, biased AI algorithms can also skew results.
It’s free, every week, in your inbox.
So explainability is harder as you must do something to explain the decisions of your own model.
You must continuously monitor it to see if its within bounds of the applicable privacy framework.
A lot of biased AI systems have been reactive rather than proactive.
The key to tackling ethical AI issues is to.
For example, reusable software packages like AIF360 enable you to measure bias.
At the same time its developing policies to ensure fairness in AI applications.
Wiggerman questions whether its okay for big tech to process the bang out of data theyre gathering.
Do they need the data to make their product work?
Could they do it with less data?
These questions are also relevant for machine learning.
I can put in your exact GPS location and predict your movements.
But do I really need this information for my product to detect fraud?
Maybe I only need to know whether youre in the Netherlands or in Belgium.
Privacy, ethics, and fairness should be added to the product development life cycle from the beginning.
In other words, fairness and privacy by design.
This isnt something that should be added afterwards.
Consider security by design.
Security measures shouldnt be added afterwards, but taken into account from the beginning.
AI ethics needs to trickle down into all the risk policies of any company.
They also lack experience with financial processing.
and big tech just doesnt have it yet.
The outcome of this computation is usable for each bank but the raw data remains secret.
In this way it allows you to maintain user privacy.
When something happens, information is shared between banking services rather than treating security as a competitive area.
This is where big tech companies may differ.
We dont automatically think about AI and financial services.
