Hi, I'm Juan Lavista Ferres, Chief Data Scientist and AI for Good Lab Director at Microsoft. Ask me anything about how we’re using AI and data science to help solve the world’s biggest societal problems.

If I'm at high school now, with all that revolution happening, what should I focus on?
What was hot in AI five years ago is not hot anymore. What are some foundational things I should learn that would serve for years to come?

A great foundation in statistics, math and coding is fundamental for anyone and will always help you! Even though AI has seen a significant increase in research and utilization in the last few months, the foundation of AI is still based on math and statistics that many times is 100s of years old. Even the foundation of artificial neural networks was created in 1970s and 1980s.

How are we ensuring that AI for Good outweighs AI for bad?

As with any technology, AI can be used as a tool and a weapon. When we take on a project, we start by growing through an AI assessment to think through the possible impact of the work – both positive and negative.

On the one hand, AI has the potential to do a lot of good - from helping us solve complex problems like climate change and disease outbreaks, to making our daily lives easier and more efficient. But on the other hand, there's also a lot of concern about the negative impact AI could have.

So how do we make sure that the benefits of AI outweigh the risks? It starts with being mindful of how we develop and use AI. That means making sure that we're building systems that are transparent, explainable, and accountable. It also means being proactive about identifying and mitigating potential risks - for example, by investing in research on the safety and security of AI, and by establishing ethical frameworks and standards for AI development and deployment.

And that’s why there have been a number of times that we’ve decided not to do a project – simply because the potential good did not outweigh the bad.

How does your team choose which problems are the most important to solve, and how do you know what difference using AI made?

Here’s a LONG answer to your question (because it’s a complicated process sometimes!):

First and foremost, we look for projects that have a significant impact on society or business. This impact can vary depending on the field, but we can often compare projects to determine which will have the greatest impact.

Second, we always partner with an organization that has subject matter expertise related to the problem we are trying to solve. This helps us better understand the problem and identify the best AI solutions.

Third, we assess whether we have access to the necessary data and whether AI is a feasible solution for the problem at hand. This involves a thorough analysis of the data and the problem to determine if AI can offer a meaningful solution.

Finally, we make sure that the partner organization is equipped to use and leverage the AI solution we develop. We work closely with them to ensure a successful knowledge transfer and to enable them to continue using the AI solution after our involvement ends.

Overall, our decision-making process for choosing which problems to solve with AI involves considering impact, partnering with subject matter experts, assessing data and feasibility, and ensuring successful knowledge transfer. By following these steps, we can determine which projects are most likely to benefit from AI and demonstrate the difference it can make.

I'm the parent of kids aged 7 and 11 (and a high schooler!). What should I teach my younger kids about AI? They have a lot of fun with generating images from prompts, but how can I leverage their interest into something that will be useful?

I teach computer science to 5 to 10-year-old students at Global Idea School, including my own children who are 9, 7, and 5 years old 😊. One effective way to demonstrate the power of AI and develop prompt engineering skills is by generating images from prompts. However, it is crucial to first establish a solid foundation in coding. I use Code.org -- it’s an excellent learning platform (and kids as early as kindergarten can use it). Once they have developed their writing skills, I introduce them to Python. Any child who can read and write has the potential to learn coding.

What causes an AI to "hallucinate" and make up information?

Hallucination is a phenomenon that occurs in large language models, which are designed to generate coherent and contextually relevant text based on the input they receive. These models work by predicting the most likely next word or sequence of words based on patterns observed in the training data.

However, because the training data is often sourced from a wide variety of texts, some of which may contain inaccuracies, contradictions, or fictional content, the model may generate text that includes factual errors or information that is not grounded in reality.

In other words, the model may generate text that is plausible and coherent based on the patterns it has learned, but that is not necessarily factually accurate or consistent with the real world. It is important to keep in mind that large language models are not capable of true understanding or knowledge, but rather rely on statistical patterns to generate text. As a result, they may produce text that is incomplete, misleading, or outright false, and it is up to humans to critically evaluate and verify the information they generate.

How many Skittles can you fit in the full interior space of a 1996 Honda Civic DX?

Oooh. I’m doing some research on this one – just got to get it right, because I know people will check my work!

What are your thoughts, if any, on the alignment problem? What approach is Microsoft taking towards it?

Artificial intelligence systems MUST be aligned with human values and objectives. This is something that Microsoft and my group takes as a fundamental priority. You can learn more about our efforts in our Office of Responsible AI and the work my colleague Natasha Crampton is doing here: https://www.microsoft.com/en-us/ai/our-approach

What industry(ies) do you think AI will bring forth the most drastic changes to?

Simple answer: almost everything.  To (shamelessly) quote my boss, Brad Smith, "the industrial revolution is now coming to knowledge work. And knowledge work is fundamental to everything.”  

What advice would you give for aspiring data scientists today?

Having a strong background in coding, statistics, and mathematics is undoubtedly helpful, but what's even more crucial is having the curiosity to ask the right questions. It's essential to learn the necessary tools, but we should always prioritize applying them to real-world problem-solving. While there are numerous exciting developments taking place in the AI space, our ultimate focus should always be on solving real-world problems.

How did you get started in data science? What sparked your interest?

I got lucky: my parents bought me a computer, so I learned coding when I was 8 years old and became fascinated with the power of programming. I went on to study computer science and started loving data, from my early interactions using SQL and being able to query relational databases.

In 1998, I attended a presentation about decision tree machine learning algorithms that I loved. This is why I completed my master’s degree in machine learning and data mining at John Hopkins University in 2005. Back then, this was the equivalent of a data science degree as there was no real path for data science. Fast forward to today, many universities offer data science degrees/programs, and there are plenty of career opportunities available.

AI seems to be such a big topic and what I wonder most is what are the major ways it can affect my everyday life? Especially when it comes to AI for good. Thank you!

AI has already helped with my productivity (tools like Open AI GPT have been extremely helpful in my research and coding). But on the unexpected side: I was born and raised in Uruguay, so English is not my first language – and now AI has helped me express myself in English, which was a welcome surprise!

If you had to choose one thing you are most excited about regarding AI's impact on society, or how society might change because of AI, what is it?

Health care is a big one – especially since more than 400 million people don't have access to essential health services at all. I think AI will be a big help in this space -- especially for people living in rural areas. We’re already seeing some early applications, like I mentioned in another comment here.

When you look to the future of AI and data visualization; what gets you the most excited?

Simply: the ability for AI and data viz to help people solve giant societal problems.

Here’s an example: diabetic retinopathy is the main cause of blindness in the working-age population worldwide. Out of the 450 million people who suffer from diabetes globally, one-third will develop diabetic retinopathy if left untreated. However, with only 200,000 ophthalmologists in the world, it's not physically possible to diagnose all these patients. But AI models can detect diabetic retinopathy as accurately as a good ophthalmologist, meaning we can change lives around the world.

Here’s more about it: Binary Mode Multinomial Deep Learning Model for more efficient Automated Diabetic Retinopathy Detection - Microsoft Research

Has anyone tried feeding the source code for your best coding AI's to them and asking for optimisations, improvement, bug fixes etc?

We use LLMs to help us with coding all the time, from Github copilot or ChatGPT – they can provide a great way to help with coding.

I keep hearing about prompt engineering. Is that really important? Is that a real future career? How can someone train themselves for that?

Yes, I do think prompt engineering is important. In many ways, prompt engineering is our way of using natural language to code. Before large language models like GPT, people needed coding skills to write software. With large language models, we now have the ability to write code in natural language. So, in many ways, prompt engineering is coding, and it has a real future career!

Are you currently using AI to design novel metamaterials or do you plan on doing so?

Machine learning can accelerate development in materials science, but models must have qualities beyond predictive power. Here’s a study from our lab that summarizes applications of interpretability and explainability techniques for materials science and chemistry, showcasing numerous examples of the application of interpretable machine learning in a variety of experimental and simulation studies.

https://pubs.acs.org/doi/full/10.1021/accountsmr.1c00244

Who gets to decide which problems get “solved?”

Good question! Someone asked something similar. Dropping the comment here: https://www.reddit.com/r/IAmA/comments/12ylspw/comment/jho271k/?utm\_source=share&utm\_medium=web2x&context=3

Does the Global Renewables Watch monitor transmission constraints as well or plan to? If so, what source of data are you using?

Thanks for the question. Currently, the Global Renewables Watch (GRW) focuses on measuring the output of solar and wind farms around the world. We’re not currently monitoring transmission constraints right now.

However, we are always looking for ways to improve and expand GRW. I understand that transmission constraints are important in the renewable energy industry and I’m interested in incorporating this data into the platform in the future.

Here’s a link to the research paper: An Artificial Intelligence Dataset for Solar Energy Locations in India - Microsoft Research

Lots of examples of AI for good projects are applications of computer vision (like the ones you listed). Are there examples of high-impact problems where a natural language processing skillset could be applied?

Yes! Just one example: the majority of medical knowledge is stored in electronic medical health records. Using natural language processing to comprehend this data will be a game-changer for researchers, hospitals, and patients.

I’m a woman in my late 20s struggling to transition my career into data science. Do you have any advice? eg. formal education vs boot camps, what do you look for on a resume etc

Formal education can be helpful, but boot camps are very, very helpful. My favorite way to learn is by solving real-world problems using data science, so look for opportunities to apply these skills in practical projects.

Some members of my team have successfully transitioned to data science in their early 40s and are now accomplished data scientists themselves – so you’re in good company! Good luck!!

Is the hype around AI overblown or justified? It seems to be everywhere?

Both!

On one hand, AI has already brought significant advancements fields like healthcare, finance, and transportation. I have seen firsthand many problems solved using AI-powered technologies have significantly improved people's lives and have the potential to revolutionize industries in the future. AI is also not new, we have been using AI for decades, but technologies like ChatGPT have democratized some of this power and we have 100s of millions of people using AI that before was only meant for a small group of scientists and engineers.

On the other hand, the media tends to overhype AI's capabilities, sometimes portraying it as a magical solution to all problems. AI is not magic, in many ways is just MATH AI. We also need to understand that AI still far from perfect and has limitations and challenges.

All this biodiversity monitoring sounds cool. What happens with the data? How will it be used to help things get better?

We partner with organizations that use the data to answer specific questions, from understanding the population, and the behavior of species. In order to solve a problem, the first step is to measure it. The majority of our partners will publish studies using the data, and sometimes the data is open sourced so other researchers can also leverage it. As an example, we worked on a project with The Wild Nature Institute to understand the social networks of giraffes which can help us understand and improve conservation efforts: https://www.sciencedirect.com/science/article/abs/pii/S000334722100258X

i’m currently in the final year of my computer science bachelors degree, and have a SWE job as a fullstack engineer lined up at a large tech company. how stable is that career path with the increasing capability of natural language models like GPT4, for instance? what skills should i strengthen to remain relevant in the tech sector? and what software/tech skills do you see being the first to become automated?

Your career path is fine, but natural language models like GPT will help with productivity. If there’s one lesson I’ve learned in my career: continue your education! Continuously learn and stay up-to-date with new technologies and trends.

Thank you again for all your questions! Until next time!