Addressing Accessibility and Bias in Our AI Design
As a team, we have decided to build an AI system for the safety of the scooter riders. In order to increase safety, we have decided to make riders more informed about the incoming traffic, road conditions, and weather situation. We are also planning to use augmented reality to show the direction to riders, which we believe would help riders to keep their eyes on the road instead of staring down on their phones while riding. There would be many more challenges down the road while building this AI system and one of the challenges that we must deal with is overcoming biasness in our system. In my opinion two of the main sectors where we must identify biasness in our system involves recognizing all sorts of pedestrians on the road and understanding commands of different languages.
Taking actions based on the crucial circumstances is must for our system. Scooter riders should pay special attention to not only incoming traffics but also to pedestrians walking on the road. Collision with pedestrian may result in severe injury to rider as well as pedestrian. In order to reduce this kind of incident it is important for our system to recognize pedestrian on the road. And more importantly, our system should be able to identify people of all background. We have seen many incident with automotive cars as well as software where they have failed to identify people of minority groups thus resulting in systems making wrong decision. In same way failing to recognize minority would result our system to be biased thus we must address this problem while building our system.
Another important sector that we should consider while building our AI system is making it usable to people of all backgrounds. Since we are implementing command system in our AI, we should build our system in such a way that it will be able to understand commands of various languages. Although natural language processing algorithm has not been fully developed till now and it can be challenging for our project to take such approach, but what we are trying to emphasize is that people of all backgrounds should have access to our system. One way of achieving this would be installing the feature of understanding various languages in our AI system.
Thus, the two major challenges in my opinion are recognizing people of all backgrounds on the road and understanding commands spoken in various languages. In order to overcome these challenges, we must train our AI system before deploying them. We should collect large amount of data from diverse fields and train our system accordingly. Our system should be able to identify people of all backgrounds and should understand commands of various languages. Thus, if we train our system on people of diverse backgrounds and on people who speak different languages, we can make our product more accessible. Making our data inclusive of minority groups can also greatly help us create an AI system which would be less biased in my opinion.
