This Week in ML & AI avatar

Dexa/This Week in ML & AI

Learn more
Published Feb 13, 2018

Security and Safety in AI: Adversarial Examples, Bias and Trust with Moustapha Cissé - #108

Moustapha Cissé, a Facebook AI Research scientist, tackles the critical issues of racial bias, adversarial examples, and trust in AI, advocating for fair and secure AI systems. He emphasizes inclusivity and accessibility in AI research, suggesting self-criticism mechanisms and initiatives to ensure global participation and robust, reliable AI models.
Episode Highlights
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) logo

Topics covered

Popular Clips

Episode Highlights

  • Inclusive AI

    Moustapha Cissé highlights the importance of inclusive AI programs to address the challenges faced by researchers from underrepresented regions. He mentions initiatives like Women in Machine Learning, Black in AI, and Data Science Africa, which aim to make the AI community more open and diverse 1. These programs are crucial for building AI that aligns with societal values and avoids biases 2. Cissé emphasizes the need for AI to reflect the diversity of human experiences, stating:

    It's critical that as a community, we become more open and more diverse, because the models that we build and the data sets these models learn from.

    ---

    Such inclusivity ensures that AI technologies are representative and beneficial to all, not just a select few.

       

    Global Accessibility

    Global AI accessibility remains a significant challenge, with geographic and resource barriers hindering participation from diverse regions. Cissé shares his personal experience of being unable to attend a conference due to visa issues, highlighting the systemic obstacles many researchers face 1. He argues that fairness in AI starts with considering global problems, not just those relevant to specific populations 3. Cissé stresses the importance of addressing these barriers:

    Fairness and bias is not only in the models that we design or the data sets the models learn from. It starts with the problems that we consider.

    ---

    By broadening the scope of AI research, the community can ensure more equitable access and representation.

       

    Conference Access

    The accessibility of AI conferences is a critical issue, as they are often held in Western countries, limiting participation from global researchers. Cissé points out that these events are not easily accessible to those without the "right passport," creating a barrier to entry for many 2. He calls for more inclusive practices to ensure diverse voices are heard in AI development 1. Cissé's commitment to building AI that aligns with societal values underscores this need:

    I'm committed to building axiological artificial intelligence. And by axiological, I mean an artificial intelligence that is aligned with the value of the society in which it operates.

    ---

    Such efforts are essential for fostering a truly global AI community.

Related Episodes