 Hi everybody. My name is Eva Black, currently an open source PM at Microsoft, but my talk is not about that. It's like we're here to talk about how to see your own perception bias. There we go. So this talk, like some of the traditions of my childhood, will follow a bit of a spiral structure. I've included some signposts to help us navigate a couple different topics that I'm going to weave together, such as facial recognition and perception. It's going to go pretty fast. This is really meant to be a 20-minute talk, and I've kind of compressed it down. So let's begin with a relatively simple-seeming question that philosophers have thought about for a long time. What is perception? And is it objective or subjective? I've always loved the color red, but I didn't really understand or think much about why until lately. I want to realize that I don't see red the same way as other people. Turns out about 10% of the world has an inherited trichromal anomaly, meaning we literally don't have the same light receptors in our eyes. So I like the color red because it's different for me. I don't see much of it, but it's well, and then I have to thinking, what about the sort of internal perceptions, memory, and imagination? Are they also different? Could they be different per person? I can envision a palace in my mind, but it is filled only with my own memories. And there's an interesting interplay between memory and sensation. When I imagine a rose receiving on a screen, it doesn't really bring up memories of my grandmother, the way that actually smelling a rose does. Waiting for the slide to load. What about my current state? Some days, I might forget to eat luncheon. Well, in the afternoon, I'm kind of snarky on an email. Turns out that my body's physical state also affects my perception of the world. That same rose might not smell very sweet if I'm angry, or it might smell even stronger if I'm feeling kind of sad or tired. And so I think we've arrived at a rough definition of perception. It is the sum of the raw sensory input mediated by the state of my body combined with unconscious recollections, memories, impressions of the past. And this is true for everything. If I'm listening to music or a job candidate's voice on the telephone, my perception is affected by all of these things. Now, I didn't come up with this framework. I learned it as a little kid growing up in South India. And ever since then, I have continued to study meditation to practice it. And I want to share with you one quote from a Tibetan tradition that really describes what meditation is. It is the act, the process of observing the arising of thoughts and feelings and memories in our mind. And by merely observing them, we actually reduce the effect of our past experiences to color them. We reduce our bias. And sitting on a cushion is not meditation. That is where we go to practice. Just like a gym, we exercise our mental muscles to build equanimity before we go out in the world. Whoa, we skipped a couple of slides here. Here we go. So about four years ago, I was running the engineering department for a hosting company. And my CEO asked me to build an unbiased hiring practice. And he handed me this work rules book from Google. And so, look, these folks have already done all the research for you. All you have to do is follow it. Here's the process. And you scrub names, gender markers, pronouns, all of that. And it's supposed to be free of bias at the end of the day. But what happened, what became obvious is that this is really just, it's hidden the bias in plain sight. All the work that we implemented just covered it up. And I continued researching bias. Wikipedia lists hundreds of different types of bias. And there's hundreds of studies on bias. And I tried to summarize them as best I could with a work frequency chart, but there's way too much to put into a short talk. All these different types of bias describe ways that our current judgment or perception or reaction to a situation is deviates from the objective norm as a result of some past experience. And what's common across all of these is generally speaking, we're not aware of our own bias. And now at the intersection of studying bias and hiring and running a company with a very particular culture, I chose to leave that company and come out as trans and take some time off work. And I planned to get back into the job market about six months later. But that six months turned into about 18 months. As I reapplied for the same types of roles that I'd already held, a lot of the same people, the job directions piled up. And they were rather not about technical skills as much as something else. And after a while, I started to get a little bit grumpy and wonder what had changed about, you know, what changed. And one hiring manager actually took me aside after an interview process and say, I just don't know how to hire someone like you. This is coming after almost 20 years in the industry. Now, what had changed about me? I was still just as skilled as I ever was, obviously, and say two things. First, the way other people perceived me changed. But secondly, the way I saw the world had changed. Maybe you've heard the phrase seen through rose-colored glasses, and this describes what's known as likeability bias. It's our predilection to judge more favorably something when it's said by someone that we like, coming from, you know, a past experience with someone like that. Like that. The inverse is also true. If we've only ever seen negative portrayals of a type of person, a category, we're less likely to respond favorably to anything they say. Well, I'd also been practicing meditation for almost 30 years at this point. I thought I knew what color my own glasses were. Imagine my surprise when one day, starting hormones, the world got brighter. We can put the slide to change. The world got brighter, and richer, and sounds got deeper, and roses smelled stronger. And I was completely surprised by this. Could it be that my own hormones had been affecting my bias all along? And they were then also? And I began to do more research into this. Turns out, there's a whole field of science studying how our bodies' sex hormones affect nerves. Some hormones can dead in pain. Others can stimulate nerve growth. They can modulate how likely we are to respond with trust. I'm not a scientist. I don't know if these are good studies or not. I'm really glad that people are asking the questions and trying to study it, because, well, our bodies are complex, and sex is a spectrum, and it's a lot more than just binary. You're probably wondering right now, this is a talk about technology. Why do I keep talking about gender? Let's talk about machine learning for a minute. What we call AI is fundamentally just a bias amplification tool. An image classifier is only as inclusive as the people who trained it. In order to train that model, categorization is done by people who label it, who remove outliers. Those people are as biased and as flawed as I and you. Now, 99% might be a fantastically good, accurate model, but to get there, you pruned outliers. And when deployed at scale, that 1%, that's a lot of people who are suddenly not represented anymore. The world is not binary, even if our machine learning models are, and our training classes teach it that way. I would not want to live in a world that had no room for this cute, fennec fox, which is neither a cat nor a dog. So while machine learning might be easy to build, this is so critical. Groups are trying to use machine learning to classify all kinds of things, to predict criminality, for example. Now, reproducing a bias that's already present in our criminal justice system, that's easy. Feed it data. But building a model that does not reproduce epistemic violence, that is hard work. I invite you all to do that. Hard work. Now, we cannot code our way out of this problem because it is not a code problem. This is a people problem. To build a less biased system, we have to learn to see our own biases first, and then evaluate the systems we're building. The best tool that I know is to live in another culture, not just another country. Do not consume it as something other, but go there and become part of it. Live there. Then come back to where you used to live. You will see the world differently. And that distance, that gap is a bias you always had and did not know. If you can't move there, learn through books, learn through movies, learn through podcasts, learn through other communities in your own city. And the second best tool I know is meditation. You don't need to find somewhere quiet to sit, though that does help. Over time, you learn to build the quiet place inside yourself. Find a tradition or a teacher that resonates with you. You don't have to start when you're five. You can start anytime. It is hard work, but engineering is hard work, and if you want to build a better system, that's less biased. Well, thanks for listening. Hopefully, you have a little time for Q&A now. Thanks for listening. Well, I ran through that pretty fast. It was originally meant to be about a 20-minute talk. Hope Shannon is able to join us. I'm looking to see if there's anything, any questions coming in? I don't see any yet. Okay, no questions have come in yet. This platform is a bit different. I don't know if anyone was even watching. I'm sure folks were. We've got questions coming in now. Fantastic. Question that we have is what resources did you utilize to start educating your co-workers? That's a challenging one. At the moment, I'm relying on hearing my own feedback. That was weird. Relying on work that other teams inside of Microsoft have been doing and occasionally gently recognizing when stuff that my co-workers are doing comes from a place where they're not aware of their biases and trying to help them see that. Educating co-workers is a challenging thing because if folks don't want to see their bias, they're focused on a deadline, then it's a distraction. I see a question. Shannon, I think I can just take a couple of these. Okay. Let me know if you need me to read any to you or feel free to take what you like. Cool. Thank you. Any tips on how to change the people around us? Lead by example. Gently. Like I said, if people don't want to do the work, you cannot make them. For those that do want to do the work, be available and show them some resources. There's a question. Will the slides be online? Yes, they will be. I'm going to continue watching the chat for a couple more minutes in case any more questions come in. There's one that came in that says, have you read the book by the art of thinking clearly by Ralph DiBulli? I have not. The title reminds me of thinking fast, thinking slow, which was an interesting read. I don't know that I agree with everything in it, but it's a good model. Another question is, how does one suss out the inability to perceive or be open to perceiving their own biases during the hiring process? That's a really good question. I have occasionally done this. It's tricky to not cross the line and make an interview a source of conflict, but testing someone s edges by presenting them with a situation that detects their biases. Let's say I was trying to administer an interview to a documentation writer. I might include things in there that challenged assumptions, but that runs into so many flaws with cultural assumptions that really don't work well globally. One of the biggest problems that I didn't have time to cover in this is how much our biases are steeped in our cultures, both our current and our culture of origin, and how the differences there are not a bad thing, what's important is to recognize and respect each other's differences. If I'm testing someone for biases based on my biases, it's really unfair to their culture and experiences. I would say to add a little more to that question, what I would look for is their reaction to being tested rather than the specific answer they gave in the test. Additional questions at this time? We still have about five more minutes if anyone has questions for you. I will be hanging out in the Slack channel and I'll also be in the Confidential Computing Consortium booth for a couple hours. If folks want to find me in either of those when these few minutes are up, thanks for coming.