CSI-MTH-190
Start thinking about how data may be biased.
I’ll give an example then we’ll consider three additonal examples.
HP worked with NASA to develoop medical devices.
Hewlett Packard made an ear pulse oximeter for all in the 1970s.
At the same time as HP, Minolta was developing a pulse oximeter.
Takuo Aoyagi help develop the first pulse oxymeter for the finger.
But there was no discussion of different skin tones.
Small device that is battery powered for the finger tip.
People can keep it in their pocket.
Today there is still bias in these devices.
Brought to light during COVID what many people with darker skin were sent home from ER despite having breathing difficulties.
The bias in these devices is imparted by the people who design them.
HP collected data to consider skin tones.
If Minolta collected data on skin tone, it was ignored.
Those choices matter.
Let’s practice considering bias in the data.
Machine Learning (AI) example:
Start with a dataset of lots of pictures of potential cancers.
Cancer Specialists (Humans) label pictures as cancerous or not.
Data (Images of skin) is fed into a nueral network repeatedly until AI can tell appar cancer images. This makes a model.
Model is then used to take a new image and guess if it is cancerous or not.
Any Bias Here? When would it matter?
Stop Question Fisk is a policing strategy used in New York City as well as other major cities.
All the data is collected and reported here
Data is reported on a form the police officers submit.
Any Bias Here? When would it matter?
We will be considering the babynames dataset in the next lab.
This dataset includes all names given to US citizens as collected from the social security administration between 1880 and 2017.
Any Bias Here? When would it matter?
For more examples of algorithmic bias read chapter 8.6 from MDSR