Introduction
Introduction to Existing AI Concepts
History of Centralized AI
AI(or Machine Learning) models are being used in various fields. All of those models are trained in a centralized way which is followed by these :
Infringement on Data Privacy
Personal data is essential for training, but most are private or sensitive. AI companies collect and utilize the data, which can infringe on data privacy.
No Reward
People provide their data to the company but, in most cases, unknowingly. People need help to get a proper reward for providing the data.
No Ownership
AI companies own AI models. The resulting profits earned by the models are taken solely by the company.
We Need a New Form of AI Training
To compensate for the shortcomings of centralized AI, a new form of AI Training is needed. It should satisfy the following :
No Infringement on Data Privacy
Participants can participate in the AI training without infringing on their data privacy.
Get Reward and Ownership
Participants are rewarded appropriately for their contribution rate. They can also insist on their partial ownership of the model.
Robustness to Malicious
Participants' Malicious actions are detected and stopped by the system. For example, if any malicious participant tries to get a reward by stealing another's model or if it tries to train the model arbitrarily to a specific direction, it is stopped by the system.
Last updated