Countering marketplace bias through loan exploration

Inspiration

We were inspired by past research performed on user bias based on borrower appearance in lending on the Kiva marketplace, as well as the many borrowers whose loans expire each year due to lack of funding, to see if there's a way to better balance the Kiva marketplace so that more worthy projects can get funded.

What it does

We propose a system to break lenders out of habitual modes of lending by giving them new kinds of loans that are different from the ones they currently lend to, as well as offering tools to Kiva to monitor and address distortions in marketplace funding.

How we built it

Currently we have a proof-of concept Python notebook that illustrates how giving lenders new loans to explore might look.

Challenges we ran into

We considered early on whether it might be possible to directly intervene against unconscious bias through a ranking algorithm that operated directly on extracted appearance features shown to be subject to lender bias in past research, but not only is that an emerging area of application in machine learning that can be challenging in technological scale, but there are also many ethical challenges and risks involved in even having the data that such models would produce, so we decided to defer that problem to future exploration and discussion.

Built With

• python
• tensorflow