You may not know it but you are probably already contributing to the development of AI. When you do a search query and find a result, this helps train the search engine. If you have ever completed a captcha by identifying objects such as traffic lights, bicycles, or busses, you could be helping improve computer vision systems. Using voice assistants, GPS navigation, and interacting with social media posts are all engagements that may help train AI.
The field of Data Annotation is big nowadays. Datasets can contain tens of thousands of data points or more. The datasets used to train Large Language Models such as ChatGPT contain billions or trillions of tokens.
You may occasionally find me working with the CrowdGen or Upwork platforms — labeling, judging, or rating of images, audio, search results and other kinds of data annotations. This kind of work tends to be kind of tedious but is very helpful to the big companies. They work with very large datasets to train their AI models.
On the robotic frontier, a new and exciting area of data annotation involves making videos of a person doing common household tasks such as cleaning the kitchen, taking out the trash, or vacuuming the carpet. Contributions to these kinds of datasets will help train robots to do the same things. There are already companies like Kled.ai that are paying people to perform daily household chores and submit a video of these workflows. Kled.ai has an App that you can download and start contributing to the collection of these types of datasets.
One interesting way to think about it is that AI learns not only from deliberate training efforts but also from the digital footprints people leave behind. Every search, click, rating, review, photo, and interaction can become a tiny piece of information that helps computers better understand how humans communicate, move, work, and make decisions.
— ChatGPT
If you have any desire to contribute your voice to help train AI, there is a project called Mozilla Common Voice that the Mozilla Foundation started in 2017. Instead of paying participants, they rely on individual volunteers from around the world. You can read sentences into your microphone, submit those recordings, or listen to and validate audio recordings. The publicly available datasets include over 130 different languages.
To me, it is a pretty fascinating world that we are all a part of in today’s digital society. I am eager to learn all that I can and to see where it goes next.