Interview: Falaah Arif Khan, AI and machine learning scientist by day, comic book creator by night

With years of experience in machine learning, both academically and in the industry, Falaah Arif Khan is building up her expertise in the field. But in the process of developing this knowledge and know-how, she noticed some concerning trends in her field. For one, she saw how the media exaggerated and hyped machine learning and AI. Moreover, she saw ‘experts’ overstate the implications of their own work and the general public buy into this hype..   

Frustrated, Khan found a unique outlet for approaching this problem: comic books. She began writing a comic about Machine Learning, explaining her field in a new and accessible way, in the hopes of widening the discussion. 

Khan spoke with Hayat about this intersection of art and tech, her plans in academia, and more.    

You worked at DELL EMC as a software engineer for 2 years, before applying to NYU’s Master’s program and a few other academic appointments. What was that transition like?

I worked in the industry straight out of college, for 2 years. In that time I built Machine Learning models for authentication and security applications. My experience in the industry showed me firsthand the limitations of current machine learning capabilities. In working towards making these models reliable and robust in production, I became interested in the theoretical foundations of robustness and generalization of Deep Learning models and hence decided to switch to academia where I could pursue fundamental research in this direction.   

Another thing that happened when I was realizing the limitations of current Machine Learning models was to see the stark contrast in the media reporting on the very same technology. The interest and hype around AI and ML just did not match the ground reality. I quickly came to realize that the problems of media hype, gold rush mentality in publishing and exaggerated reporting were in fact well discussed and regularly called out by researchers and experts in academia. 

Machine Learning Yearning, Falaah Arif Khan and Zachary C. Lipton

Unfortunately those conversations seldom reached the general public and hence there was emerging this void in “AI Literacy” — where those who actually worked on this technology and thought deeply about how to improve its capabilities engaged in critical discourse in good faith (for the most part), but those who were looking to ride the hype-wave and make a quick buck regularly exaggerated the capabilities of the same technology.

The media in turn exacerbate this problem by conflating science with science fiction. The ones who were most adversely affected however, were the general public who regularly jumped from techno-optimism to techno-bashing within their bubble of ignorance.   

To me, comics were just my way of trying to widen the range of these discussions beyond the technical community. At some level, I created my first comic, also a therapeutic exercise, without ever imagining that so many people would relate to it and it would grow to become a thing of its own.   

I’d like to really dive into your comics, because that’s such a fascinating intersection. Did the artistic and technical sides of your life develop separately, then merge, or did you always have an interest in that intersection?

I’ve drawn/created art for as long as I can remember. My sister is an artist too and we’ve made a ton of art over the years, just for ourselves, with no plans to go professional or anything of that sort. So, I grew up in a super artistic household, and art has just always been a part of my life. I even had, as an item on my bucket list, to ‘Create a Comic Book’. I just could never figure out about what.

And looking back now I can see the itch to experiment with styles and methods was always there- I did realistic pencil sketches, then switched to polygon art and more abstract stuff, then switched to line art, later did some cartoon pieces, and would always be playing with different mediums. I only switched to digital art once I started ‘MachineLearnist Comics’. The first volume, “Meet AI”, was actually done entirely by hand, on paper and then scanned and put together digitally.   

Meet AI by Falaah Arif Khan

The intersection with my technical background really came together when I was doing research on different theories of consciousness and came across the work of Riccardo Manzotti. I saw one of his comics in an APA Newsletter and it was unlike anything I’d ever seen before or even thought could exist. I excitedly reached out to him and we discussed how sometimes it’s easier to show an idea than to write about it. He also explained how he made his comics for fun and for clarity of thought and not for any formal publishing.   

After finishing the comic I relentlessly cold-emailed prominent researchers who frequently spoke about the themes in the volume. From their response, I realized that this was actually something people needed/wanted to read, as much as I needed/wanted to make them! I’ve never looked back since.   

Another major influence in the early days  (and I say early days loosely, because it hasn’t even been a year since MachineLearnist comics started!) was Prof Zack Lipton’s Approximately Correct blog. He truly is the master of satire and his writings showed me the merit in speaking truth to power and that there would be those in the ML Community who would appreciate my comics. It’s really come full circle with the ‘Superheroes of Deep Learning’ comic series that I now co-create with him.     

What’s up next for your comics?

We just released the Volume 2 – Fairness and Friends of the “Data, Responsibly” comic series
that I co-create with Prof Julia Stoyanovich. In this volume, we delve into algorithmic bias and
different conceptions of ‘fairness’ in machine learning. I’m very excited about this piece because
we’ve gotten Dr Eleni Manis, a political philosopher, onboard for this volume and we’ve really
gone deep into doctrines of Justice, and Equality of Opportunity from political philosophy. We
always wanted the comics to be truly interdisciplinary and so I’m very pleased with how we’ve
been able to stitch together ideas from political philosophy and law, with emerging scholarship
in Fair Machine learning.  We even recently presented a tutorial on Fairness as Equality of
Opportunity at this year’s ACM Fairness, Accountability and Transparency conference, based on
this volume!

Volume 1: Mirror, Mirror is now available in French and Spanish and a French translation of
Vol2 is in the works!

The Deep Learning Superheroes will also return with new adventures in 2021.

Finally, one pet project of mine that I really want to create but just haven’t gotten the time to do
(yet!) is an interactive, Create Your Own Adventure comic book (think Black Mirror
Bandersnatch) about algorithmic auditing.

Where would you like to see your career go? Will we see much more of the artistic side of yourself?

Oh absolutely! This is a beast that can’t be tamed now. 

In seriousness, it’s been incredible to see the reaction to the art. There’s also a push from within the community to adopt other publishing mediums, beyond the research paper, and I think it makes me excited to stay in academia.  

My immediate plan is to get a PhD. I would like to work on the Robustness of Deep Learning and on Fairness and Algorithmic justice. Robustness of ML is important because current models are extremely brittle and behave unexpectedly in settings that they have not seen during training. These ‘out of domain’ or unseen samples usually correspond to historically marginalized demographics in society that are underrepresented in data as well and so the impact of misclassifications is disproportionately felt by them.   

I’ve been extremely lucky to find mentors and collaborators who are incredibly supportive and value the comics as much as any other form of research. So I intend to continue creating comics throughout my doctoral studies and beyond! I’ve been creating art (in some form or other) for so long now that it’s pure intuition at this point. To me ‘success’ with respect to my PhD would be to develop the same kind of intuition and feel for fundamental research.   

Decoded Reality, Falaah Arif Khan

After graduating I would love to get a professorship at a University and set up my own research lab. I want to create more spaces in academia that value ‘unconventional scholarship’. One of my favorite things in the world is to see people being unabashedly creative! From speaking to people in ML about my comics, the most common thought I hear is “I used to draw/create art too, but I couldn’t find a way to sustain the work” or “I really wanted to create a comic book/ graphic novel on X, but I couldn’t find any suitable person to create it with”.

I would have never thought to put the two together if I hadn’t come across Riccardo Manzotti’s comics or Zack Lipton’s blog and so it’s important to me to keep doing this kind of work. I would really want to push for artistic mediums to become more widely used formats to disseminate research and to establish the credibility of comic books and other artistic pieces as artifacts of scholarship. My hope is, in the near future, we’ll see a comic book in the list of prescribed readings in coursework on AI! And we’ll see comic books and graphic guides feature heavily on the Google Scholar profiles of researchers!   

But then who knows, maybe in the course of my doctoral studies I’ll come across yet another bucket list item that I could never figure out how to complete and I’ll take another hard left turn to pursue it obsessively instead.  

 

Don’t miss the second volume of “Data, Responsibly”, titled Fairness and Friends:

Click to access fairness_en.pdf

 

Catch all the current comics by Falaah Arif Khan here.