How My Housemaid`s Menstrual Cycle Killed My Machine Learning Project?
Would algorithm know what is right and what is wrong? Yes, but only if we teach them — because we are the creators of those algorithms.
Mid last year I completed this Machine Learning course and since then I have been hunting for a real-life practical problem that can be solved by a machine learning algorithm.
One day our housemaid hasn’t turned up for work. If she is not in by 9 am my wife starts predicting by running some checks against housemaid`s past absence patterns.
Patterns! Yes, that what I needed
— and after talking to my wife for a few minutes I was sure that I was going to create an algorithm to predict whether our housemaid is coming to work or not
Next challenge was data. There was no data at all — it was there in my wife’s head, and I was not ready to wait for the next technology that can dump brain to the computer
So I started interviewing my wife, many times, collecting data, features, influences … and then started visualizing patterns
Ethics in Data — And The Nature of Ethics
While collecting all kinds of data that was relevant to maid`s past absence we lost track of what kind of data we are collecting. There was some very obvious pattern and regularities in data — and that was a positive sign.
Once enough training data was collected and algorithm was up and running. I got a bit greedy and started asking my wife about some particular days and reason why our house maid was absent. My wife explained me a few interesting examples that my algorithm was able to chew easily and became more accurate
There was data about our maid`s menstrual cycle that proved to be very effective for the accuracy of my algorithm. Now, the difficult part comes — the subjective nature of ethics — and it is a very personal thing.
Do I have right to use and hold that data without permission to do so? Is there any pragmatic approach that i can take and make use of this data? I don’t know.
Well, it was very evident that for those day ( at least some days ) she is not going to come for work or won’t stay for the whole day, so let’s put it in leave policy and call it “period policy” or “menstrual leave policy”
So that killed my algorithm — and i am happy about it.
Later I started to wonder how corporate world is dealing with such issues? Are there such leave policies. What i am sure of, is that no woman is going to share menstrual cycle via outlook and putting an Out Of Office responder stating “ i am on menstrual leave “
Japan, China, South Korea and Indonesia have menstrual leave policies
If we implement any algorithm blindly there could be many ethical issues that need to resolve.
- What will be the impact of these machines on human behavior?
- How to eliminate bias that is hidden in the algorithms?
- How can we guard our self against AI mistakes?
- What about unintended consequences?