Opinions expressed by: Entrepreneur: contributors are their own.
You’re reading Entrepreneur India, an international franchise of Entrepreneur Media.
There’s big data, and then there’s quality data. Bobidi is providing companies with high-quality data sets for artificial intelligence (AI) testing. Do you remember Tay? Tay was testing conversational understanding for Microsoft’s Technology and Research and Bing teams. They deployed her on Twitter, and within 24 hours, Twitter users had her saying such terrible things that they had to take her down.
Tay is a prime example of the problem Bobidi is working to fix. In short, the team at Microsoft fed Tay an enormous amount of standard conversational data. When she was let loose on Twitter, she wasn’t ready for information that didn’t fit the norm. It’s like starting a new job; you are inundated with the basics, then the boss leaves you in charge after the first day. Heaven forbid anything out of the norm happens because it could result in disaster, just like the Tay experiment.
Bobidi produces high-quality data sets that you will find in the real world. Their data sets are known for having realistic proportions of outliers. For example, if you want to test a new face detection algorithm, Bobidi can offer a collection of facial photos that fit what you see at the grocery store rather than on television or a Google search. Data like this helps train your algorithms to be more accurate when average everyday users interact with them.
Bobidi also gives AI engineers the ability to dynamically build their own datasets based on different needs. They can track different versions of the models and their performances that change as the datasets evolve, resulting in critical insights that engineers can use in order to quickly improve the model.
So, how does Bobidi get its data? Bobidi uses a contributor community. People all over the world, tens of thousands of them, have downloaded the Bodidi app and act as contributors. Once a contributor downloads the app, they can start taking pictures for money.
The app presents challenges like “trick the AI by recording newly coined words that the AI doesn’t recognize”. Users use their smartphones to record different words that include recently released TV shows, movies and new celebrities that AIs haven’t had a chance to catch up with. This algo testing method is faster and more efficient than the traditional way of utilizing ‘big’ data that is often more expensive and could increase the bias the model already has. On the other hand, Bobidi’s community-based approach helps their customers to find the data the AI only needs as it lets the customers simulate the model with real people.
“When the AI models don’t work, the problem is usually not the model itself but what kind of data the model is trained with,” shared Bodidi’s CEO and co-founder Jeong-Suh Choi.
With new challenges released weekly, there is a revolving door of fun, new opportunities to make money as a user. Alongside photo challenges, some other mediums include speech, drawing, and text.
Bobidi was founded by Choi, and CTO and co-founder Yan Bei who have been long-time friends. As they hung out over the years, they identified the problem with big data and the results it produced. After some deliberation, they decided they had something with their idea to create a community of data contributors to replace stale and static data sets — and Bobidi was founded in June 2021.
Overall, Bobidi is on a mission to show the world there are better ways to train AI algorithms. They can be more precise in a shorter amount of time; they just need the right data.