Do you think it’s possible to be the ethical voice of AI while working inside Google or building a $5bn startup? Let me know your views in the comments and I’ll crack on with explaining how Dr Fei Fei Li’s story might help us to answer the question.
This is part 2 of my deep dive into Dr Fei Fei Li and her $5bn business World Labs, in part 1 we covered her early years which culminated with events that are now attributed to thawing the AI winter.
This Substack is adapted from the corresponding YouTube episode, if you’d prefer to listen to the story.
At the end of part 1, we left off with her being firmly established as a key AI player within SIlicon Valley, thanks to her role as director of the Stanford AI Institute and her reputation for groundbreaking work in AI.
In the early 2010’s alongside other researchers she brought together 3 very important things: 1. Data, 2. GPUs & 3. Neural Networks, through the ImageNet competition. It yielded powerful results. ImageNet may have helped to give her a reputation for technical innovation but this was not enough to give her a reputation for being an ethical voice in AI, in fact her exponential scale approaches are, in my view, contradictory to being ethical.
So how does a computer vision researcher become deemed by some to be the conscience of the AI industry? And does that reputation hold up under scrutiny?
So we rejoin this story in the mid 2010s, the tide was turning, companies who had previously shunned the “AI” moniker, in favour of specific terms such as deep or machine learning were re-embracing AI. As someone who was in the digital tech landscape at this time, I remember just how hyped AI suddenly seemed to become I joked that I could probably raise a lot of money with a click of my fingers with an on the spot pitch about AI, blockchain and drug discovery.
Industry was taking it seriously and Fei Fei Li noticed.
In fact this is where she began to form a valuable insight…
With an exponential view of AI, foundational AI models could no longer be built in academia because the public sector could not afford the compute required to train these monolithic beasts!
This meant that the most exciting AI development was, and continues to be, happening in the private sector, an observation that motivated Fei Fei Li to take a secondment from Stanford from January 2017 to September 2018 to Google as Google Cloud’s Chief Scientist of AI/ML and Vice President.
She clearly didn’t agree with Elon Musk that Google are the worst stewards for AI (at least that was his view at that time, he’s obviously added several others to that list now).
Personally, having spent time in both academia and business, I think that they have a lot of things in common, but that’s because I’ve always been striving in my own startup or consultancy. In both cases those who thrive are those who are happy and able to spend time navigating complex, organisation politics. It might seem that when a scientist moves into a corporate environment a lot of things change, but for Fei Fei Li, who was brought in to a small but growing lab, it wasn’t too dissimilar, at least in the day to day. She just got to build something with exponentially more money, talent and food than was available at Stanford.
But her time there was not without controversy, and it begs the question, what changes for a curious researcher when they enter industry, we will come on to that in part 4.
Several things that happened at this time both inside and outside of google led to new insights for Fei Fei Li and a true emergence of her impetus around Human-Centric AI.
But are ethical AI and human-centric AI the same thing? This was actually the question that triggered me to do this whole series so stay tuned for my views on that.
Human-Centric AI was not just a fancy tagline for Fei Fei Li, she started to develop real initiatives to look at some of the issues across the AI sector.
Stanford Human-Centered AI Institute (HAI)
On return to Stanford after the sabbatical Fei Fei Li became the founder and director of the Stanford Human-Centred AI Institute
Stanford HAI was developed to do what it said on the tin, building on her ideas for how to put human wellbeing at the centre of AI development. Not only are they doing research but they also focus on the implications of developing AI.
They have several initiatives covering:
Research: Funding hundreds of scholars to study AI’s impact on society, healthcare, the economy, and the workplace.
Policy: Educating lawmakers and advocating for legislation, such as the creation of a national AI research resource, to ensure responsible, ethical AI use.
Education & Industry: Preparing future AI leaders, hosting executive training programs, and publishing the widely cited annual AI Index Report.
I have learned a lot from some of the initiatives that have been promoted as best practice by Stanford HAI, now under the leadership of James Landay. My particular favourites are the Affinity Group projects, where students from across the university come together around challenges that they think are important and figure out cross disciplinary ways to tackle the problems identified.
HAI has also been doing comparative evaluations of large language models and continues to engage with Congress, the Senate, and the White House on AI policy.
As someone who exists in this ecosystem, I think this work is important, however we have to recognise that it is only one part of the puzzle, there is a responsibility for all of us to take this knowledge and create a better future.
Coming back to the Fei Fei Li founder story, her focus on societal applications of AI, specifically healthcare didn’t come in a vacuum, the main thread of her book is family and in particular the relationship with her parents stands out. Specifically health challenges. These challenges opened her eyes to how AI should be approached in healthcare settings.
Her 2020 paper in Nature titled “Illuminating the dark spaces of healthcare with ambient intelligence,” described her work in ICUs, combining smart sensors and AI algorithms to monitor patients at risk of falling from their beds.
In her book she goes into more detail about how this helped her to realise the importance of human-centric approaches, but I continue to ask the question….are these approaches built into her new $5bn dollar company? We will explore in future parts of this series.
My sense is that a few revelations came from this era, in particular, it seems that Fei Fei Li seems mostly focused on the application end of human-centric AI: ease of use, privacy (especially for vulnerable people/people who can’t consent), and the huge opportunities of World models/3D mapping and interrogation of environments.
This is what I call a founder blinker, tight focus on the end goal gives tunnel vision, so focused on where you are going that the wider context just disappears. Perhaps it needs to, because otherwise would anything move forward? This just highlights the importance of governance structures and diverse teams, people who might keep things moving forward, laser focued, but those people also have to accept that they need others to take a more holistic view.
Having said that, Fei Fei Li is smart enough to be tapped into some of the most prominant issues, in her books she discusses big mistakes made by AI, such as the killing of Elaine Herzberg by a self-driving Uber Volvo XC90 on March 19, 2018, as she walked her bicycle across the street.
Another prominent discussion point in the book is the impact of bias. In a field developed mostly by white, educated, relatively well off, technical men there was, and still are well documented bias issues.
AI and all emerging technologies are amplifiers, so they amplify issues that exist across the field.
This was known by Fei Fei Li and other people in her team, and part of the solution is to widen access of AI to a wider range of people. Following a Stanford school outreach programme in 2015, designed by Dr. Fei-Fei Li, Dr. Olga Russakovsky, and Dr. Rick Sommer in 2015, to familiarise high school girls with AI, AI4ALL was created and officially launched in 2017.
As is stated on their website: AI4ALL’s mission is to ensure that the next generation of AI leaders reflects humanity.
They state:
“AI’s rapid evolution demands a responsible, inclusive approach to development and implementation.”
So up until this point, Fei Fei Li has exhibited some of the humanity, creativity and drive that feels optimistic in AI leadership. In addition and perhaps conversely, she is also obsessed with perception of the visual world around us, she is fixated on her mission to create a Cambrian explosion for the robots…
…and perhaps that’s why she is the one that has led the charge from words to worlds.
In part 3 we will dive into the formation of World Labs and the concept of Spatial Intelligence.
We have largely discussed the sugar-coated, highlight reel of Fei Fei Li’s journey so far. Until about 2 years ago, this was how I perceived her too, with most of my information coming from reading her book. I truly believe that she is cut from slightly different cloth from some of the other billionaires that appeared on the cover of TIME magazine with her.
But we all know that light and dark co-exist, Fei Fei Li will be more familiar than most with that concept, vision is almost entirely driven by contrast. And if you want another vision metaphor, let’s add a bit of colour: Fei Fei Li demonstrates some red, green and amber flags; she is complex. Perhaps if we can understand her we can understand how to make the tech world more human-centric, not just for the humans that are users of products, or the humans that are founders & investors in tech, but a holistic end-to-end view of the benefits and negative implications of technology on humans and, by inference, our environment.
I look forward to sharing the next part of this story with you.
Resource list: https://large-maiasaura-d69.notion.site/Series-1-Fei-Fei-Li-World-Labs-378a6aa238fb81aba7a5fc79ee7e70c2?pvs=74
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.