Explainable AI with Machine Teaching | Murat Vurucu

Discover how machine teaching can revolutionize AI development by creating a system that asks the right questions to the right people, involving users and experts in the labeling process, and leveraging game theoretical approaches for optimal results.

Key takeaways

The current approach to AI development is inefficient and prone to errors, often requiring tedious interviews with experts to understand the problem and engineer features.
The solution is to create a machine that asks the right questions to the right people, allowing data science teams to be more productive and scalable.
The bottleneck in data science is not the lack of data, but the inability to label and engineer features correctly.
The key to solving this problem is to involve users, scientists, and experts in the labeling process, using active learning components and weak supervision.
The goal is to create a system that can learn from human intuition and reasoning, rather than just relying on data.
The speaker’s company, Latentine, has developed a system that uses game theoretical approaches to define the data space and allows experts to label and engineer features in a collaborative manner.
The system has been tested and shown to be effective in producing high-quality labels and features, even with a limited number of examples.
The speaker believes that this approach has the potential to revolutionize the field of AI and make it more accessible to a wider range of industries and applications.
The key challenges to overcome are the complexity of the problem, the lack of expertise in certain areas, and the need for scalable and efficient solutions.
The speaker’s company is looking for partners and friends to test and develop this technology further.
The presentation is not technical and is meant to be an introduction to the problem and the solution, rather than a detailed technical presentation.

Explainable AI with Machine Teaching | Murat Vurucu

More talks