There is nothing more beautiful to me in my life than Artificial Intelligence. Whoa, whoa, what about a sunrise? Or the sound of ocean waves? Nope, still Machine Learning algorithms are the things to win my heart.
I was recently introduced to the book To Siri, With Love: A Mother, Her Autistic Son, and the Kindness of a Machine. Now let me be very clear. This book is extremely problematic, as many parenting perspectives tend to be in the Parents of Autism communities. For many, autism is something to be cured, feared, adjusted for, corrected for, and unlearned. The book even mentions plans for the mother to sterilize her autistic son, Gus; eugenics has no, I repeat, NO, place in any conversation, ever. Read this tweet thread for more information on the book, from an autistic writer/editor. Please don’t get me wrong; if my child were to suffer in the ways that I did with undiagnosed autism as a child, I would want to help them in all of the ways that I could. There is some tragedy in the fear, pain, and overstimulation of being an autistic child (and adult for that matter). Crying when foods touch or being afraid of going to the grocery store; peeing yourself because you’re afraid of getting lice if you go near other children; living in a world where you express yourself differently from other children and fixatedly tear the skin off of your own lips because you like the feeling of the blood dripping down your chin (yep, got sent to the principal’s office for that one in the second grade haha). There is most certainly some behaviors that I would want to help my autistic child work on and overcome. But the big difference is embracing that the way neurodivergent individuals process the world is not a handicap but rather, a struggle because they exist in a world catered for others that are not them. I could talk forever about my utopia where everything is ASMR sounds (soft whispers and gentle tapping and other stimming videos), everyone uses Google to answer all the rhetorical questions they hear, subtitles on every video, and just let’s cancel all overpowering smells please. But that’s for another time. This post will actually delve into what I would have wanted to hear from a book with such a seductive title: My autistic relationship with AI, and the deep emotional sense I have for statistics and computing.
For a little more background, I am a hyper-verbal autistic non-binary queer adult scientist who loves their life. What a mouthful! #extra. My special interests are: honeybee communication, machine learning algorithms, packaging (like, food packaging or different kinds of boxes), Alice in Wonderland, and autism (lol so meta). I am sensory-seeking and hyper-sensitive, meaning that I actively seek out stimulation but not in the form of human touch or loud noises or other big sensory events that hypo-sensitive people might seek out. I am typically not sensory-averse meaning that usually smells and sounds do not bother me, and I actually tend to enjoy them. But I can hear everything like the ticking of a watch or the bird from across the street, which can be very distracting (just ask me and my autistic friends, we can’t carry a conversation without attuning to some random sound like meerkats). When I am experiencing hyper-focus, I can’t hear anything, and am extremely startled if people come up behind me (like, a small tap on the shoulder = a gas stove just exploded). In class, if someone asks a rhetorical question like “I wonder where that word came from…” I am compelled to look it up, memorize the answer, and interrupt the class to share a detail everyone forgot about. I remember almost everything that’s ever happened to me except for directions (oh man do those just go right out the window for me), and I achingly watch myself continue to talk and ramble even when all of my calculations are giving me the signal that it was time to shut up about 10 minutes ago.
Many of these things I find kind of funny about myself. Sometimes they are really difficult to deal with. But the thing that helps me most is that I have a really beautiful coping mechanism: Artificial Intelligence programming. Computer Science itself has been like a faith to me for many years now (yes, that’s totally weird). When I first started programming, I was utterly entranced. It wasn’t until halfway through college that I wrote my first line of code, and I attribute this to some of my feminine upbringing or whatever. But some way or another, I was able to eventually find the beauty of code. A quick sidenote: my infatuation with programming is actually part of a harmful stereotype that we need to either be obsessed with computing or can’t do it at all. And that the fixation on logic and numbers is what makes a Computer Scientist. This is absolutely not the case, and absolutely anyone who is excited to try can experiment creatively or otherwise with computing topics!
I went on to study cognitive science with an Artificial Intelligence “track” of study, and worked in a research lab studying how child development and learning can be modeled by Bayesian Machine Learning algorithms. Now I’m in a PhD program where I study ML literacy. For me, the human mind can be broken down into computational parts; the most beautiful concept in the world to me. For many, this is reductionist and maybe offensive. I can understand that. But for me, the fact that we could understand ourselves through piecing together computational components is the ultimate expression of life. (I’m slightly laughing at myself because I sound so enamored, dramatic, and dreamy about computers right now) Ever since I was little, I had to learn to reflect and question myself and the way that I fit within social interactions. I now describe it as “my defaults are all different”. If you think like a Bayesian, you might say my priors are extremely different from most of the people I interact with. Since I was young and expressing my gender as a little boy, building little fairy houses in the sand instead of talking to other children, having extreme meltdowns because parking garages are too large, and crawling underneath my mattress (between the bedframe and the mattress!) so that I could feel squeezed… well yeah, you can say that was the impetus to start questioning why I was so weird. Computing gave me some answers, some joy, some expression, some pride, and helped to define my inner world of pictures and diagrams. I connect everything to computing, and run tiny programs in my head all day long in order to succeed at social interaction (sometimes) and compute possible Theory of Mind of others. I wake up in the morning and run my “Start Day” program and do the same thing every day; drinking exactly 6 glasses of water because I should (while n < 6, you get the idea). By the end of the day, my brain is exhausted and I fall asleep in strange places. But I assume I just ran out of CPU space or something. For me, computing has been an explanation of why people are the way they are, and why I am the way I am. Let me explain a bit of my personal worldview:
We are data driven machines. As newborn infants, we come with a set of priors that have been passed down through our genes by evolutionary history. For instance, babies react with increased fear of images of spiders and snakes, even if they’ve never interacted with one. Babies will interact similarly with their parents, craving touch and affection and initiating sounds to get what they need. Various things come built-in. I personally don’t believe in universal language or innate language or anything like that, but I do know we come with a set of biological potentials that allow us to express language and perform some computations that not all animals can do (though the field of animal cognition is a fascinating one). Now, given these priors, what happens next? We get exposed to BIG DATA. The BIGGEST DATA out of all the BIG DATA sets in the world. Because life is Big Data. Every instance of a word being paired with a referent in the world (like, apple + point to that red ball object that someone is eating) is a datapoint. Some datapoints simply don’t land. The green apple might not fit in the distribution of “apple” for the concept the child is forming. Or maybe the point was too vague and the child attributes “apple” to something else in the room. But with enough data, we can train the model. With enough data, the things that become more likely are the things that are consistently classified using the same linguistic label. Perhaps there’s a distribution, like sometimes “the apple one” is referring to an apple pie. But we learn sets of rules and sets of probabilities to understand the world around us. Sometimes, our deeply set priors can actually get overwritten by data; and we learn something we didn’t believe before (snakes can be pets, Donald Trump can be president, you know). We can then compute a series of heuristics to define our beliefs, our morals, our assumptions. We don’t have to conduct the computational search every time, but have come up with heuristic probabilities for how we respond to certain types of situations. Our data shapes us, influences us, comforts us, and surprises us. And with such overwhelming flow of information, our algorithms can be quite sophisticated. How to understand a set of human emotions? I’d argue it’s a nice combo of emotional processing priors and labeled data. How to understand a linguistic pun? I’d argue it’s a big search through words we have seen and their statistical probabilities of occurring together and the statistical unlikelihood of them being used in the particular punny phrase. Again, I know that this is most certainly not everyone’s world view. And for some, it may be seen as shallow and lacking true measure of experience. But for me, it is the best experience I have ever lived through; computing my world to make more sense and to learn how to share love and affection in a complex and computable way. The AI technology that is all around me is affirmation and validation for the way that I process humans. I perform searches, classify images and phrases, respond with pre-computed small-talk, have difficulty with sarcasm, and am constantly learning from data. Not only do I find comfort in the abilities of assistants like Siri or ALEXA, like informing me right now of the distance from Moscow to Singapore (8415km)… but I find reflections of myself that make me feel better about the strange and wonderful way that I process my world.
Let’s look at a few examples from classic AI/ML education: Tic-Tac-Toe, Regression, and the notion of Cost functions. Tic-Tac-Toe seems like magic until you realize it’s a simple set of rules with amazing speed and memory. When I play games or have conversations I often try to predict possible paths that me and the other agent could go down. Eventually I’m constrained by memory and the possible outcomes become more like a semi-solved mystery (with probabilistic likely outcomes, of course). But I know that I could map out simple possibilities like options in Tic-Tac-Toe. It’s actually a beautiful experience to deconstruct the magic sometimes. Pull back the curtain and see that AI is a reflection of ourselves.
It’s actually a beautiful experience to deconstruct the magic sometimes
As for regression, I think of it like this: In the world, there are weights on predictors to help us make sense of the world. Just as we value different priorities or pay attention to different things, predictor Betas combine in such a way to influence outcomes. For me, I’m particularly interested in how we influence social change. What factors lead to an individual caring about social issues and human rights? There exists some predictors and I try to map those out so that I can be effective. I also try to keep my body relatively healthy, and I know that some predictors carry more weight than others; 8 hours of sleep probably matters more than that half a cup of carrots. We must also recognize the error terms and the failures of model fitting; we can’t really predict the future. But we can learn more about what matters and what helps.
There lies a great freedom to accept that we are all working towards different models of happiness, and that they can often coexist.
And of course, my new favorite thing to talk about: cost functions for humans. I’ve recently started to explore the idea of personal cost functions. I think of everything in terms of maximizing goals and minimizing costs and error. Some of us want to maximize things like money, or family, or creativity, or happiness (more likely a combination of several of those things!). We want to minimize sickness, loneliness, boredom, etc. I find my life purpose in my pre-defined cost functions; knowing that my main goal is to have fun with what I’m doing. I have to remember that everyone’s cost functions are different; some of us seek publications first, a good job, money for our families, intellectual pursuit, productivity, relaxation, time with family, etc. We are all maximizing and minimizing different things. Of course, we can interact in such a way where we can serve each other’s goals and help reduce struggle; but at the end of the day I can’t compare my success against anyone else’s cost model. There lies a great freedom to accept that we are all working towards different models of happiness, and that they can often coexist.
In conclusion for this haphazard blog post, I’d just like to reiterate that the relationship between computing and autism can be such a beautiful one. For me, it is intricate, dependable, explanatory, and quirky. By no means is that kind of relationship limited to just people on the autism spectrum, and by no means do all autistic individuals feel this way towards computing. But I wanted to put this all down in writing, so that my special friendship with computers might reach someone else, too.