unSILOed with Greg LaBlanc

476. AI's Potential for Positive Social Change feat. Juan M. Lavista Ferres

Listen on

Episode notes

AI is a fast-growing field full of potential insights, challenges, and ethical implications for its users and the world. How can the people behind the machines explore the ways to use AI and data technology to leverage societal benefits?

Juan M. Lavista Ferres is the Corporate Vice President and Chief Data Scientist of the AI for Good Lab at Microsoft. He also co-authored the book AI for Good: Applications in Sustainability, Humanitarian Action, and Health.

Greg and Juan discuss Juan's book 'AI for Good,' various AI projects, and the critical role of data labeling. They also discuss philanthropic initiatives from Microsoft, the transformative impact of robust data collection, and the challenges of applying AI to real-world problems. 

Juan covers innovations like GPT and Seeing AI, as well as the ethical concerns of open access to AI models, and Satya Nadella's leadership transformation at Microsoft. Listen in for insights into the importance of using AI responsibly, collaborative efforts for accurate data processing, and how AI technology can actually enhance real lives.

*unSILOed Podcast is produced by University FM.*

Show Links:

Recommended Resources:

• House of Medici
• Andrew Carnegie
• Moore's law
• Global Forest Watch
• Ruler Detection for Autoscaling Forensic Images
• BeMyEyes App
• Michael Bloomberg
• Brad Smith
• Amy Hood

Guest Profile:

• Profile at Microsoft
• LinkedIn Profile
• AIforGood.itu.int Profile
• Stanford RegLab Profile
• Social Profile on X

His Work:

• AI for Good: Applications in Sustainability, Humanitarian Action, and Health
• Google Scholar Page

Episode Quotes:

On deciding which ai-driven projects are worth doing

12:26: We first ask the questions like, can we solve it through AI? Not a lot of problems can be solved from AI. There's a small portion of them that can be solved with AI. From those problems, does the data exist? Is the data of good quality? And sometimes the answer is no. Even if the data exists, do we have access to the data? Can we get access to the data? We will usually work on the partners' data sets, not our data sets, meaning that the data set will not leave the partners, but sometimes there's no way to have a data-sharing agreement in place, where it makes it impossible to share the data. Once we have that part, the next question is, do we have the right partner? We are not subject matter experts on the point that we work. We are subject matter experts on AI, but if we're working with pancreatic cancer, we need, on the other side, a group of people that are experts on pancreatic cancer, for example. In that case, we try to partner with people who are subject matter experts and are world-renowned.

Data needs to be representative

19:55: Data is a fundamental part. I would say the majority of the success or failure will happen because of the data set, and investing in understanding the data set—making sure that there's no bias—is a critical part of the work. It's tough; it's difficult. Data needs to be representative.

What are the do’s and don’ts for companies looking to launch initiatives for good?

36:40: I would love more companies. So, this is something that we discussed with my team. Whenever we see other competitors creating something like we do, we feel proud because that would be a success for us in many ways. So I would encourage everybody to use that technology for good. That's something that I think is certainly worth the do's and don'ts; I think it's important to make sure that this organization remains clear that its objective is on the noncommercial part of the philanthropic aspect of the company because, within this organization, the objective is to be helping society and making it clear for the people that are working there. That is something that is helping us a lot. Our end goal is to help society, and I think I would encourage other companies to do it.

Is there a possibility of a zero bug project?

21:09: Some of these problems require people to really ask the question: how is this model going to be used correctly? And that takes experience. More importantly, I think it's crucial that in many of these cases, we need to be ready to find those problems and fix them, correct? And I think that this is like software development in many ways. The chances of having a zero-bug project are zero, correct? Projects that have zero bugs are projects that people don't use. What I think is important as an organization is to find those problems, be proactive in trying to find them, and be really fast in solving them.