Generative Now | AI Builders on Creating the Future
Episode: Dr. Olga Russakovsky: Shaping the Next Generation of AI Leaders
Host: Michael Mignano (Lightspeed Venture Partners)
Guest: Dr. Olga Russakovsky (Professor of Computer Science, Princeton; Co-founder & Chair, AI For All)
Release Date: December 19, 2024

Overview

This episode features an in-depth, candid conversation between Michael Mignano and Dr. Olga Russakovsky, a pioneering researcher in computer vision and AI fairness, and the co-founder of the nonprofit AI For All. Together, they dive into the evolving role of computer vision, the changing landscape of computer science education, the critical importance of diverse perspectives in AI, and the challenge of bias at every stage of AI development. Dr. Russakovsky offers a compelling vision for what the future of AI could look like if built on inclusivity, interdisciplinary collaboration, and bold new bets.

Key Discussion Points

1. Dr. Olga Russakovsky’s Career and Research in AI

[01:17–02:30]

• Journey from math camp and theoretical machine learning to applied AI and computer vision at Princeton.
• Ongoing work in computer vision: system building, analysis, explainability, and especially AI fairness and bias.
  • “Recently, I've been thinking a lot about AI fairness and bias, particularly in computer vision systems and in AI systems more generally.” — Dr. Russakovsky [02:06]

2. What is Computer Vision?

[02:30–04:37]

• Computer vision involves extracting understanding from images and videos.
• Key applications: autonomous vehicles, photo tagging and sorting, medical diagnosis (skin cancer detection), agricultural monitoring, space exploration.
  • “It's anything that's related to understanding pixels, understanding images or videos.” — Dr. Russakovsky [02:49]
  • “You send a robot to Mars... it needs to... guide it to drive around safely.” [04:03]

3. Intersection of Computer Vision and Generative AI

[04:37–05:47]

• Generative image models (GANs, diffusion models) are now integral to computer vision research, not just computer graphics.
  • “That's now very much part of computer vision. And that's been sort of interesting that it's... both the images to some kind of understanding... but also going back...” [05:16]
• Dr. Russakovsky’s lab is active in diffusion model research.

4. Shifting Trends in AI Education and Research

[06:01–11:37]

• Explosion of interest in AI, especially driven by generative models, increasing enrollments and research activity (Princeton CS is now the largest major).
• AI is becoming inherently interdisciplinary: advancing technology cannot be separated from examining its social, ethical, and societal impacts.
  • “AI research has been much more of kind of engineering or mathematical endeavor. And I think now it's becoming much more interdisciplinary.” [06:54]
• Michael Mignano compares his own CS education (code, hardware) to the new landscape where awareness of philosophy and social consequences are vital.
  • “With a lot of these generative models, it seems like... the stuff I learned with my computer science degree actually becomes less and less relevant over time...” — Michael [08:40]

5. How AI Is Changing Computer Science Education

[09:05–13:06]

• Programming assignments are now fundamentally impacted by large language models: routine code tasks are easily solvable by AI.
  • “Historically a lot of the assignments are around, you know, write this code... but now, you know, you can ask ChatGPT and it'll write that code...” [09:23]
• Question of what knowledge needs to be taught: technical foundations will remain, but higher-level, interdisciplinary thinking is increasingly crucial.
  • “I don't think the computer science role is necessarily shrinking... the field is kind of growing...” [10:35]
• Princeton’s curriculum is actively changing: example, a fireside chat between computer vision students and a psychologist exploring the use of AI in scientific discovery.
  • “Our last computer vision class... we're hosting a fireside chat with Dr. Molly Crockett, who is a psychologist...” [11:56]

6. Pitfalls of Using AI for Scientific Discovery

[13:06–14:39]

• AI’s power to summarize and accelerate research risks steering discovery toward groupthink and “path of least resistance” solutions, sacrificing creativity and diverse approaches.
  • “[AI] can drive research towards... what is currently the most promising direction... ultimately it's kind of going to limit creativity.” [13:26]
  • “It's going to limit creativity and some of the joys and potential of scientific discoveries...” [13:48]

7. AI For All: Increasing Diversity in AI

[15:10–17:44]

• Dr. Russakovsky’s nonprofit, AI For All, aims to address AI’s “existential threat”: lack of diversity of thought and demographics.
  • “To me the biggest threat... is the lack of diversity of thought in this field.” [15:36]
• Why demographic diversity is a fair proxy for cognitive and creative diversity in AI, and how current AI practitioners have homogenous backgrounds and values that shape technology’s direction.
  • "What that's doing is it's driving us towards echo chambers and decreasing the diversity of thought in the field..." [16:20]

8. How Bias Enters AI Systems

[18:17–21:50]

• Bias is present throughout the pipeline: in data collection, model design, and application focus.
• Data sets—especially in computer vision—are overwhelmingly sourced from the US and Europe, neglecting other geographies.
  • “You have this map of the world...the US highlighted bright colors and parts of Europe highlighted...and then the rest...particularly South America and Africa are just completely missing from this data.” [18:54]
• Model objectives stem from dominant communities’ needs: e.g., focus on autonomous driving reflects the life experiences of Bay Area engineers.
  • “What people are passionate about...is very much influenced by their values, by their culture, by their upbringing.” [21:32]

9. Changing Incentives and Broadening AI's Impact

[22:15–24:21]

• Economic incentives drive research directions, but Dr. Russakovsky believes it's possible to find powerful intersections between creative application, economic value, and social benefit—if diverse thinkers are included.
  • “I don't think we have the full creativity... to reimagine what are the different kinds of applications that could be tackling ... the full range of things that we could do with AI...” [23:48]

10. How AI For All’s ‘Ignite’ Program Works

[24:27–27:03]

• Training college students from underrepresented backgrounds in core and responsible AI; providing mentorship, hands-on portfolio projects, and career readiness support.
  • “They guide the students through... a passion project of theirs... driven by the students' interest... what do they want to explore with this technology.” [26:13]

11. Correcting for Bias: More Diversity or More Tuning?

[27:03–29:59]

• Dr. Russakovsky is clear that the solution is not simply algorithmic tweaks (which are hard and imprecise), but ending root-cause homogeneity through inclusion and broader perspectives.
  • “All of that is really hard... The root cause is that we are not being creative enough or thoughtful enough or diverse enough in how we approach this.” [27:56]
  • Memorable anecdote: As the only woman in her lab, Olga noticed a voice recognition system that failed on her voice—exposing an unchecked bias due to lack of diversity among creators.

12. The Myth of Internet Data as 'Ground Truth'

[29:59–33:03]

• Simply scraping all internet data embeds hidden biases based on access and curation (“what’s popular on social media, what’s uploaded for advertising”), not the objective real world.
  • “No company wants to intentionally output a biased product out there... but is this ground truth? I mean, I think it's important to remember this is not ground truth.” [31:31]
• Calls for integrating social sciences into technical AI training, and regrets about skipping humanities education herself.
  • “I'm horrified by that... I really wish I hadn't because I feel like I'm playing catch up...” [32:18]

13. Diversity and the “Data Wall” in AI

[33:03–36:50]

• The so-called “data wall”—AI models exhausting available internet data—may require the same solution as bias: sourcing genuinely new, diverse data and perspectives.
• “Data obsession” in AI is a recent development; the next breakthrough might not be data-driven.
  • “We don't know... are there other alternatives? I think right now the alternative is... generated data...” [35:56]

14. Synthetic Data: Promise and Limits

[36:50–39:11]

• Synthetic data can help address compositional and representational weaknesses (e.g., novel object pairings), but cannot solve geographic gaps unless there’s original source diversity.
  • “It's not going to help with issues of geographic diversity if there's no representation of data from particular countries in your original data set...” [37:21]
• The effectiveness depends on controlled, intentional generation, not random sampling.

15. Who Will Lead the Next Data Revolution?

[39:11–40:07]

• Dr. Russakovsky believes the best-placed organizations to push new forms of diverse data or alternative approaches will be those led by the next generation—her students and AI For All alumni who, unbound by current paradigms, ask novel questions.

16. The ImageNet Revolution—Lessons and Next Bets

[40:07–43:11]

• ImageNet, which Dr. Russakovsky helped scale, shifted AI’s focus to data and directly enabled the deep learning wave.
• She credits colleagues Jia Deng and Fei-Fei Li for their ambition and vision.
  • “...this was the emphasis on data and on the bet that collecting more data will actually unlock... the next generation of AI models, which is in fact what happened.” [42:14]
• Hopes for a new wave of equally bold bets: not just on data or algorithms, but on fundamentally new paradigms (possibly brain-inspired models, or others yet unimagined).

17. Closing Thoughts

[43:11–43:50]

• Dr. Russakovsky expresses optimism for the future, emphasizes the importance of supporting and empowering new voices in AI, and expresses gratitude for a candid, thought-provoking discussion.

Notable Quotes & Memorable Moments

• On the existential threat for AI:
  “To me the biggest, the biggest threat...is the lack of diversity of thought in this field.” — Dr. Olga Russakovsky [15:36]

• On bias in AI data:
  “You have this map of the world... the rest of the world, particularly South America and Africa, are just completely missing from this data.” [18:54]

• On the promise of diversity:
  “If we want all of these applications to be actually sort of solved and tackled and get the attention that they deserve, then we need people who are going to be passionate about that kind of work...” [21:40]

• On computer science education’s pivot:
  “We're hosting a fireside chat with Dr. Molly Crockett, who is a psychologist... because we think it's sort of important to ask some of these more fundamental questions.” [11:56]

• On AI’s new interdisciplinarity:
  “I don't think the computer science role is necessarily shrinking... but maybe the field is kind of growing to bring in more and more ideas...” [10:35]

• On her own experience with bias:
  “The voice recognition system would understand everybody else in the lab except for me... I was the only woman in the lab.” [28:34]

• On the next big AI breakthrough:
  “I would like to see somebody else make a different bet. This is the time for somebody else to make an equally ambitious bet that it's not data or algorithms, but it's something else entirely.” [42:44]

Timestamps for Key Segments

| Timestamp | Segment Description | |-----------|------------------------------------------------------------------| | 01:17 | Dr. Russakovsky’s career and research overview | | 02:30 | Defining computer vision and its applications | | 04:37 | Crossing computer vision and generative AI | | 06:01 | Trends in AI education & interdisciplinarity | | 09:05 | How AI is disrupting computer science pedagogy | | 11:48 | Changing Princeton’s curriculum with interdisciplinary fireside | | 13:06 | Risks of AI in scientific research | | 15:10 | Introduction to AI For All and its mission | | 18:17 | How bias proliferates across AI pipelines | | 22:46 | Market incentives vs. societal AI priorities | | 24:27 | The Ignite program: diversifying AI pathways | | 27:52 | Diversity of people vs. algorithmic correction | | 31:31 | The fallacy of “ground truth” internet data | | 33:03 | The “data wall” and possible solutions | | 36:50 | Synthetic data: opportunities and cautions | | 39:35 | Who will unlock the next generation of AI? | | 40:07 | Inside ImageNet and the next big bet for AI | | 43:11 | Final reflections and gratitude |

Tone and Language

The episode maintains an open, optimistic, and critical tone. Dr. Russakovsky is candid about her field’s shortcomings while expressing clear faith in the power of new voices and bold ideas. The conversation ranges from technical exposition to philosophical and social analysis, accessible and inspiring for experts and newcomers alike.

This summary captures the heart and substance of the episode, offering a roadmap for listeners and non-listeners to understand the urgent questions, lived insights, and the future-shaping ideas in AI leadership today.