Accessibility? Bias? AI?
At most of time humans are easily to forget the importance of empathy. Without doubt, everyone is bias because we live our own life. Every day you wake up at your place, go to your work, have your meal. Even though you did those stuffs with someone else, your feeling would between very different. Humans are individual and everyone would have their own journey of life. We have built our personalities and perspective forward to value through out our daily life. It is a unique and lonely trip that no one have done it before. However, when it comes to design tool for people, we start to think about empathy and put us in someone’s shoes.
For example, the project I am working on this semester is AI for language learning. I will articulate some of my thoughts to create an inclusive and unbiased tool.
Accessibility
1.Conversation method:
I would like to add typing as a method to interact with our AI. In our first blueprint of language learning AI, we all focus on interacting with AI through speaking. However, for someone have disability in listening or speaking, they may use other method to learn a new language. Typing is also a common way to communicate with others especially in nowadays. Thus, I believe it will be a great way to increase our accessibility.
2.Speech rate:
For some hearing impaired or partly loss person, we would allow our rate to tailor their speech rate when interacting with our AI. They can start from a very slow speech rate when they are unfamiliar with new language. After being familiar with it, they can gradually turn the speech up to close normal speech rate.
3.Cultural context:
I also think that add cultural context into our AI would be a great way to improve accessibility. The reason why languages are so unique is based on different culture. For example, although grocery shopping is a universal experience among the world, there are still lots of difference between culture. Hence, adding cultural context can help us to bridge this gap and provide the actual situation for our users to learn.
Bias
1.Diverse of data provider
Extended for the grocery shopping experience for last point, the conversation people have in Whole Food may be different from HEB. As a result, to build a diverse database as possible as we can is the first step to deal with bias.
2.Create a standard of fairness
When collecting data, we should create a standard of fairness to test how can this single of data set affect the whole database. Would it makes the database more biased or unbiased? We need to a rule to follow and test our data.
Conclusion
In sum, accessibility and bias will still exist in AI for a long period of time. I would say that there will be no perfect AI which is unbiased and accessibility-friendly. However, all we can do is to keep these issues with top priority in our development process. Although we are still far from perfect, we have become more and more close to it.
(External resources: https://www.technologyreview.com/s/612876/this-is-how-ai-bias-really-happensand-why-its-so-hard-to-fix/)
02.19.2019
Eric Chi
