My talk at the inaugural Black in AI workshop dinner

The (inaugural) Black In AI workshop at N.I.P.S. 2017 was an amazing and inspiring showcase of science, engineering, and social initiatives; and it was an honor to be invited to speak at the dinner.

I’m excited to see what this wonderfully talented community will produce in the coming years.

For folks who were unable to attend the event in person, a transcript of the speech I gave is below.


Thanks to the organizers for putting this wonderful event together, and thanks to all the great presenters for sharing their work with us today. It’s been super interesting.

So, I have to be honest: I prefer small-group discussions to addressing large crowds. And moreover, when I was asked to be one of the dinner speakers I realized that I’ve given lots of academic talks and business pitches, but I actually haven’t done too many personal presentations like this.

In the latter part of my talk, I’d like to discuss some of my thoughts on what the Black in AI organization means to me, and on some of the opportunities that we have together.

But since there are some of you that I haven’t yet met, I thought I’d start by telling you a bit about my myself, and my background and research focus in AI.

I’m currently a Research Scientist at DeepMind in London, where I work on a broad range of machine learning approaches, with the ultimate goal of developing Artificial General Intelligence.

Most recently I co-authored a tech report titled “Population Based Training Of Neural Networks” that we pushed to arxiv last week. And the short-summary is that we propose a new training method for neural nets that combines ideas from evolutionary optimization with gradient-based optimization in order to perform dynamic hyper-parameter tuning and model selection.

So far we’ve seen it give consistent improvements to training across a variety of domains. And in particular, it seems to add huge benefits to systems with complicated learning dynamics — such as Deep Reinforcement Learning and GANs. And I think one of the reasons that it’s so beneficial is that it’s able to discover an adaptive schedule of hyperparameters, rather than just a single fixed setting.

You’re welcome to catch up with me later if that’s something that sounds interesting or useful to your work.

In terms of my background: I grew up in a working-class family in a small town in the northwest of England. But since then I’ve called many places home — including Cambridge, London, Toronto, Montreal, and San Francisco.

My first NIPS conference was in 2000 or 2001… when it was much smaller — probably only around 600 people — and of that maybe just one or two other black folks. It’s amazing to see how much the field has grown since then. And as I look around the room, it’s awesome to see so many fantastic black researchers here today.

In terms of career highlights so far, there are probably two in particular that stand out:

The first was a paper that I wrote with Geoff Hinton and Yee Whye Teh in 2005, while I was a post-doc in Geoff’s lab in Toronto. It was titled “A fast learning algorithm for deep belief networks” and in it we demonstrated that it was possible to use layer-wise unsupervised training to help build deep, generative neural nets. With the computational advances that have happened in the meantime, those models we were using back then might not be considered especially big or deep nowadays — but at the time it was a significant step forward. And in fact, that work with Geoff’s group, along with papers from the labs of Yann LeCun and Yoshua Bengio, played a big role in kickstarting the deep learning revolution that we see around us today.

A second highlight started a few years later in 2009 — at a machine learning conference in Montreal. I ran into a friend, Bobby Jaros, and we decided to start a company together (LookFlow) to productize some of the deep learning methods I’d developed in Toronto. It was hard work, but over the next four years, I learned a tonne about business, and about production engineering, and also about myself. We built an awesome team, and some awesome products, and in 2013 we accepted an offer from Yahoo to buy the company and its IP — which then ended up powering some of Yahoo and Flickr’s search engines.

Okay — so now that you know a little bit about me, I’d like to take a moment to reflect on what brings us here together.

Black in AI is defined as “a place for sharing ideas, fostering collaborations and discussing initiatives to increase the presence of Black people in the field of Artificial Intelligence.”

And as we’ve seen today — it’s clear that we all share a passion for science and technology.

But on the surface… it might seem a little odd to have an event for a group whose defining characteristics are the amount of melanin they possess and some other superficial physical traits.

I mean in this group we have people from over 20 different countries and 5 different continents. We have a whole range of ethnic and cultural backgrounds. A whole range of native languages. A whole range of sexualities, genders, class privileges, physical abilities, and so on.

For instance — I myself am British by nationality. European, Canadian, and Californian in my acculturation. Devoutly atheistic in my religious beliefs. And somewhere in the middle in terms of the spectrum of sexuality and gender identity.

So, given that spread of underlying diversity, does it really make sense to have a group that collects us together primarily based on properties like the color of our skin or the curl of our hair?

The answer, of course, is “Yes! … It absolutely makes sense … and there’s a massive need for this group”.

Because, in addition to our research interests, there other key factors that bring us together. Some of which are:

  • the challenges black people face in our careers and in everyday life due to the systemic and historically rooted issues that still permeate today;
  • the challenges black people face due to some of the biases, inaccurate prejudices, and stereotypes that some folks hold about black people.

Now, there are far too many examples of these sorts of things that I could point to. So to throw out just a few:

  • In the US for example, you’re more than twice as likely to come from a poor background if you’re black than if you’re white [1,2]. If you’re black in the UK, you’re six times more likely to be stopped and searched by the police than your white counterpart [3].
  • Then there are things like the immigration challenges many of you faced simply to be here today; and not forgetting the folks who are not even here, due to their visas not being approved.
  • And then on a smaller scale, there’s the more commonplace experience of having weird interactions in everyday life or at work that are just hard to pin down. For instance, in a previous job, there were several times when interactions with Research Executives didn’t go the way I expected, and I was left second guessing whether there was something I could’ve done better or differently, whether they were simply being obtuse or dumb, or whether there was some kind of racial bias at play… And that kind of thing can be really insidious. Because it’s subtle, and you never really know for any given interaction what the factors were. However, at the same time, in aggregate you know that kind of bias almost certainly has negatively affected you at some point.

Fortunately, we’re all also brought here together by a shared commitment to tackling these problems, and by a desire to ensure these issues aren’t further ingrained by the powerful technologies we’re developing

And I actually think it’s also a real strength of the Black In AI community, that these shared challenges really do bring together such an otherwise diverse collection of individuals. And the diversity within this group is something we can, and should, celebrate and leverage.

Now obviously one of the really important things we can do as a community is learn from each other and support each other. And we’ve seen so much of that today, and in the run up to the event. For me, that’s actually one of the best reasons to come to conferences in the first place. The opportunity to develop a network of personal connections and, moreover, friendships.

And then with that community, to be able to both ask for and offer things like feedback, help, and mentorship. And so I really hope that’s something that folks come away from this workshop and NIPS overall with — lots of new personal connections that will last going forwards.

But this group can also be so much more than a mutual support network.

For me, there are a couple of key actions we can take:

  • We can use our collective voices to amplify the successes of black people in AI. And in doing so, this’ll go some way to help counter the stereotypes that exist in some peoples’ minds. I think just by giving greater visibility to the Black in AI community, we have a great chance to create inspiring examples and existence proofs for others. And hopefully this’ll then encourage more black people from all over the world to participate in the field of AI.
  • We can also support capacity building projects in geographies and communities where the A.I. ecosystem and talent pool might currently be under-developed relative to its true potential. Things like the Deep Learning Indaba in South Africa are a great example of that. Or organizations like “Black Girls Code” or “The Hidden Genius Project” — which aim to increase STEM participation by underprivileged black youths.
  • And we can also use our diverse backgrounds to inject broader perspectives into the AI field as a whole. Hopefully, by doing so, we can do a better job at ensuring that the AI applications and systems that we develop don’t inherit some of the problematic biases that are still present in society at large, and instead help them become fairer, and more transparent and accountable.
  • And lastly, I think we all have a responsibility to steer the overall field towards problems and products that have the potential to benefit the whole of humanity, and not just the wealthy or privileged few.

In terms of my current work, I think that responsibly developed and deployed AGI has the prospect to be one of the most positive and impactful technological developments in the whole of human history. The mission statement at DeepMind is: “Solve intelligence. And then use it to make the world a better place.” And that’s something I really believe in.

And so while I’m here today in a personal capacity, I would like to say that if that mission is something that motivates you too, then please consider applying to join us at DeepMind. It’d be great to have more of the Black in AI community working with us.

And finally in closing: I’d like to thank you for your attention. I’d like thank again the presenters who shared their work with us today. And I’d like to give a special thanks to the organizers for all their hard work in making this event happen!