Summary
In Season 2 Episode 12 of the podcast, Shakir Mohamed, a senior staff scientist at DeepMind, shares his colorful journey from South Africa to becoming an influential figure in the world of AI and machine learning. Shakir joined DeepMind as a startup in 2012 and has since witnessed the organization's rapid growth as part of Google. Working in a startup environment, according to Shakir, fostered innovation, adaptability, and collaboration.
A prime example of Shakir's contributions in the field was his work on generative modeling in weather nowcasting which enabled high-resolution rainfall predictions. This impressive feat has promising applications in various sectors such as industry, transportation, and outdoor events. Nowcasting posed challenges for traditional numerical weather prediction methods but provided opportunities for machine learning and generative models to showcase their potential.
Generative models, a branch of unsupervised learning, are used to create and explore multiple generations and alternative plausible scenarios. Shakir explained the adaptability of these models, as they can be applied to various applications, including generating images, audio, and text.
Shakir's work in healthcare has also left its mark. Collaborating with the United States Veterans Affairs Department, he used electronic health records to predict acute kidney injury (AKI), aiming to create an early warning system to help clinicians detect and treat AKI before it worsened. Responsible innovation in machine learning and AI requires careful examination and vigilance to avoid biases and ensure the best outcomes.
One concern Shakir focuses on throughout the podcast is the racialized dimensions of data. This issue is illustrated by the distinction between black and white patients in the U.S. medical system regarding AKI. He expresses the importance of analyzing the sources of racial data and addressing such biases.
Apart from his research, Shakir is also actively involved with Deep Learning Indaba, an organization striving to reinforce Africa's presence in AI's future. The goal is to enable a new kind of technology that caters to the diverse population of the continent. Deep Learning Indaba's mission has expanded over time, promoting global communities taking ownership of their training and development.
Another fascinating concept that Shakir discussed is the idea of "queering" in AI research. This entails challenging assumptions and exploring new directions in the field by applying lessons from queer life to ensure fairness in machine learning. Queering can serve as a potent tool to identify and dissect implicit assumptions in scientific research.
Lastly, Shakir enthusiastically predicts exciting developments for AI in the next five to ten years. These include integrating AI in weather and climate science to enable breakthroughs, viewing AI as a socio-technical system that emphasizes the safety and well-being of individuals, and fostering a deeper understanding of equity, inclusion, and participation in AI design and implementation.
