HYDERABAD: WHEN it comes to applications involving the usage of Artificial Intelligence (AI), a lack of transparency in how decisions are arrived at is a major challenge, sometimes even raising ethical issues.
Now, researchers from the Indian Institute of Technology, Hyderabad (IIT-H) has developed a method to help resolve this issue, paving way to implementation of applications in areas where an explanation is mandated by law ((like the European Union General Data Protection Regulation), on how an AI-based system arrives at a certain decision. The practical implications of this research, whose findings were recently published in the Proceedings of 36th International Conference on Machine Learning, will help tackle the lack of transparency in AI on devices like phones, tablets and television sets. This could even and prevent the end user’s loss of trust in Artificial Intelligence systems.
Where the problem arises
AI systems are based on Artificial Neural Networks (ANN), also known as Deep Learning (DL) models, which mimic the working of the human brain. This helps machines learn to make decisions in a more human-like manner. In this process, machines can train themselves to process and learn from data supplied to them, and based on this, arrive at a decision.
However, how exactly they arrive at the decision is usually unknown, making them less useful in applications where the reason behind decisions is necessary, like in the health domain.
Dr Vineeth Balasubramanian, associate professor at the Computer Science department in IIT-H, said: “The DL models, because of their complexity and multiple layers, become virtual black boxes that cannot be deciphered easily. When a problem arises while running the algorithm, troubleshooting becomes difficult.”
The IIT-H team resolved the issue of lack of transparency, using a specially-devised model called the ‘Structural Causal Model.’