This essay was entirely written by a human (me!). What I mean by “written by a human” is that: I don’t use LLMs to generate content for my writing; and I don’t recommend using LLMs to generate ideas. I do use different AI (mostly ML, actually) tools for technical support, and they have proved to be reasonably helpful, mostly for my individual use case of a neurodivergent polyglot writer (although, some of that experience with AI has actually been problematic - on that, later).
My perspective here is entirely biased by the various identities that I operate from.
I structured the article based on these identities to provide clarity and focus. Because when we speak of AI, Writing, Autism - all of these concepts prove to be deeper and infinitely more sophisticated depending on what we actually mean and what is the specific context of the conversation. I will talk about Writing, what it means to me, my brief history of content creation, editing, publishing and marketing my work (not necessarily all at onceI. I will talk about my Autism, from my individual perspective and experience. I’ll try to translate how I think, in writing and otherwise, and share some examples of how my brain works. Finally,I’ll share my perspective on working in AI/ML, as someone who spent over a decade at Google and played and researched quite a bit. I have to point out here that all the thoughts expressed in this essay are entirely my own.
Writing
Writing is one of the ancient technologies humans possess; as such, it’s inseparable and impossible to exist outside of the concept of language - this is something I’ve given a lot of thought to, but of course feel free to openly disagree with me and we’ll chat.
While language itself is an expression of humanity, humanity is equally shaped by the language we choose to adapt. Each form of language offers a translation of the social and cultural setting, and can function as a powerful builder of connection, meaning, community but equally as a social disruptor, leveraging tools like exclusion and manipulation.
Keeping that in mind while speaking of LLMs, we need to understand that as humans we have the power to influence how we design our technology, as technology on its own does not exist, it only serves the cultural, economical and societal needs of the humans.
There is no single actor who determines collective outcomes for humanity, unlike the popular opinion these days that places the majority of humans as passive victims of the powerful, while the powerful are seen as virtue ascribed by the popular belief that there is such thing as “the standalone virtue” without it being shaped and heavily impacted by the totality of human thought.
We all collectively contribute to the meaning of writing, to the ways we use our language, and making divisive choices only deprives us from nuance that in turn, will reinforce the bias that we’re the victim of some invisible hand.
I’m one of the people who spent most of my life writing, and in some ways, the activity of writing helped me both translate my brain to the world, and find solace and peace in the process. I was one of those people who always preferred emails to oral meetings, text to phone calls and wrote out entire scripts of difficult conversations, debates and presentations ahead of time.
Since the age of four, when I wrote my very first word my mom proudly framed and displayed on the wall for years, I couldn’t stop writing. My very first written word was “computer” (misspelled of course) but what’s interesting is that years later I dropped out of my PhD and made a determined call to move countries - I moved to Dublin, Ireland (I was living in Milan, Italy at the time) to join one of the coolest companies in the world, to work in Google Cloud, that at that time was disrupting the market dominated by the idea of “keeping your data where you can see it”.
While Cloud was fun, one of my friends lured me into joining “the best team at Google” aka Global Expansion Team. Everyone on that team spoke at least 4 languages, and for a multilingual nerd like myself, I certainly felt like I found my tribe. We operated as international market advisors, using early technologies for automation, translation and expansion- in a fascinating dance and tension between a fast-paced data-driven force of a tech company and the complex realities of real life products and services that needed to be adapted to local audiences. That work has absolutely driven me to New York, where I was Google’s first hire in the America’s Global Expansion Team, and living and working in New York has shaped my life and career in ways I wish to describe in another very long essay (not this one).
Despite my all-consuming work at Google, I have always maintained a form of authorship: I used to write actively on different platforms and usually maintained a personal blog or a website where I used to publish my thoughts. I made multiple attempts at writing a book and have kept countless drafts of unpublished work.
As a multilingual, I noticed early on that I’m writing differently depending on a language I’m using, almost as if each language involved a slightly different identity that I have but that remains indistinct until I put pen to paper. It became very clear to me that having an alternate way of thinking can produce an entirely new form of thought, a novel idea, a creative approach - all of which is impossible when we try to standardize every thought into a predictable pattern suitable for widespread consumption. (This point is central to the essay I wrote, The AI Outliers Tax, debating that LLMs built to cater to the median, tend to flatten the divergent thinkers, which by extension, flattens creativity and original thought.)
Autism
Although my early writing was always featured as the “top of the class” work, my prose was characterized as unexpected, and while “engaging” with its authenticity, yet “quirky” at the same time. I was told that I’m funny, although not everyone was always sure whether I was intentionally funny or just weird. (I’m not actually sure, either).
Despite being an early writer, and generally excelling at academics, there were prominent early signs that my presumed “giftedness” did not represent a full story of my “quirky” multifaceted brain. I struggled with most things that neurotypical humans would perceive as easy, or normal.
It took me an unreasonably long time to eat my food, to get dressed, to transition from one situation to another. I daydreamed, and quickly succumbed to sensory overload and easily stressed, sensitive to the overwhelming amount of signals my brain observed and absorbed unfiltered. Like, getting dressed became harder when I was under stress, and nearly impossible, to the point where I would refuse anything other than a very specific safe uniform that offered predictable colors, shapes and fabric texture.
Most autistics, regardless of being hyper- and hypo-verbal or even selectively mute, are intellectually aware and capable (pending the specific context of stressors and presence of a safe enough environment, because under extreme circumstances you can just shut down, unable to connect to the world, like “we lost signal” kind of scenario), communication is where I tend to struggle. It’s like living inside an alternate universe in a different galaxy, fully able to grasp the paradigms of other universes yet sometimes unable to bridge the reality with the words available to us. In turn, I noticed that neurotypicals often failed to read my communication styles, like my “funny” writing. Sometimes it’s not the lack of language that hinders expression, but the challenge to convert theory into practice: the executive functioning, the motor skills, connecting what you know you need to do, to actually doing it - sensory stressors always played a huge role in that.
My response was unreasonably sensitive if I was feeling any kind of injustice. Cruelty, discrimination, social exclusion, absence of explanation, lack of logic, lack of truthfulness, ignoring the nuance. I was much more likely to have a complete meltdown from observing a social game that benefitted some, and harmed others, than from having to solve a complex math problem. A massive challenge is actually noticing what’s going on at scale, which makes gatherings of more than two people particularly tiresome.
It’s a complete myth that autistics don’t read social cues. Speaking for myself, of course, but I do read them, but also I read ALL of them, ALL the time, and it’s just so bloody exhausting. Oh, and most of them make no sense to me: ego-driven, unfair, unkind and UNNECESSARY. Yes, I’m also fully aware that I’m completely judgmental. I’ve been called out on my “back-and-white thinking”, I was called a “pompous prick”, “a snob”, etc. Yep - it’s all me.
I struggled socially. Although I would speak to anyone from a person living on a street to whoever is the highest authority in town, with an entirely democratic and equal regard for their perspective, when it came to deeper friendships, I had a hard time connecting with other children. My Kindergarten BFF was a non-verbal boy who refused to engage in any other play than chess. I remember when the teachers were trying to gently nudge him towards other forms of play, and he was having meltdowns. I absolutely felt his pain, and luckily they always left him alone and that’s where he and I just played chess, all day long. Even though we never exchanged a word, he was my best (and only) friend for a very long time.
As a kid I spent more time playing with boys than girls, because I simply could not fully follow the complex social cues of girls my age: I did not get their body language or their social references and found the unspoken social hierarchies particularly exhausting. I played with the boys quite well because their games were mostly physical: running, climbing trees - none of which required any particular emotional or cognitive effort, yet taught me that physical activity often helped regulate my nervous system, unlike anything else.
Getting deeper into what Autism looks like for me as an individual writer, is helpful once we start looking more closely into using AI-tools. One critical concept I’d like to unpack in this regard, is the self-translation: what does it take for an autistic author to be able to share their authentic voice, without being misunderstood or dismissed?
The quick surface level answer I expect from some readers is that I should absolutely attempt to share my authentic voice, and never flatten it through the application developed primarily for neurotypicals. I can’t agree with this more, as a matter of fact, after spending many painful years masking and adapting myself to the extremely tiresome: because, sensory overload, unnecessary illogical practices like look me straight in the eyeball - neurotypical modes of communication, I’ve become an advocate of bilateral education, or double empathy: neurotypicals need more awareness and understanding that their experience is not the default. However, unfortunately the reality is such that in most social situations requiring communication and penmanship, the dominant culture prevails. Even the most talented piece of writing will be easily dismissed or misunderstood if it simply does not offer the cultural signaling required for the majority of the population to grasp its meaning.
What is the point of expression if it’s meant to sit there in the dark, unabsorbed, unseen?
What is the purpose of a written word that remains unread?
I don’t know, I’m just asking.
AI (or ML?)
I guess I started using what we refer to with AI - the capabilities - long before we started calling chat bots “AI* and everyone became such an expert.
My early use cases included working on automation tools that supported global expansion by translating and adapting enormous volumes of content to hundreds of languages. Those early tools allowed for machines to take on tedious tasks considered too repetitive and monotonous for humans, who would lose focus and make simple errors. While these tools offered humans immediate relief and time savings, they weren’t just replacing human jobs, like those or translators, analysis or expansion consultants. Because while part of the task involved automating bulk translation, actually adapting the content to cultural relevance and actual complex market dynamics, industry trends and user experience required careful human oversight.
Translating words and meaning are obviously two different things. Because a language is not simply a collection of labels, it consists of a complex web of social and cultural code, unspoken assumptions, shared history and jargon.
On an entirely personal branding front, I was using early content generation tools to design and build my website. The machine was better at putting my unfiltered “bottoms-up” processing thoughts into a sleek business marketing language. I was using Jasper for my content marketing copy and advice on website branding and design. It was good enough for generating helpful suggestions, in such offering a different way of thinking and filtering the world that was not immediately accessible to me. This is an example of why people talk about technology being helpful with accessibility. Kind of like a wheelchair ramp for the mind.
Fast forward a few years I started working on AI/ML more deliberately, and ultimately launched my personal Anthro-AI podcast on YouTube to help debunk some of the prevalent myths and clarify what seemed to become an incredibly chaotic and polarizing public discourse.
AI is not a single technology that serves a single purpose. There are multiple applications and use cases and seeing the public discourse shifting more and more into the confusing and unhelpful conversations, rather than focusing on the actual technology vs useful applications - made me want to step up.
One of the biggest miscommunications was the confusion between AI and ML. While all LLMs are built using machine learning technology, the technology itself is not new and is definitely not as hyped as the latest models. Gen AI is one application of these technologies, and is definitely not the entirety of AI.
I care about this deeply, because I care about value and nuance. When we collapse the discourse about technology into the latest chat bot, we risk having a very unproductive conversation.
Machine learning has powered useful tools behind the scenes for many years. Like search engines, recommendation systems, speech recognition, translation, spam filtering, fraud detectiom and so on and so on.
And even today my favorite applications are completely boring and not sensationalist at all.
I like using voice to text while walking and thinking out loud. Transcription and organization tools help me capture and refine my thoughts.
I also use tools to quickly understand information I usually struggle with, such as administrative, legal, insurance etc - information I don’t necessarily want to read and even if I did read it all, I probably would spend a lot more time and energy to truly understand what I need to do and why it matters. In an age of information overload, AI/ML tools actually help me reduce my mental load so I can have more free time to do my human stuff.
I still use tools to translate myself to the world. As I explained earlier, it’s not for the lack of ideas or languages. Where I struggle the most is translating my thoughts into socially/culturally dominant processes and my internal models into a linear form of communication most people can actually follow along. An editor can help uncover blind spots where I lose the reader simply by assuming that they can draw the same connections - while the reader is already thinking, wow, she really jumps around topics a lot.
I did notice that AI editors usually attempt to moderate my delivery and structure my reasoning in ways that undermine my points so I can essentially sound a bit more mainstream. I describe this phenomenon in my essay An AI Outlier’s Tax - where the dominant systems are optimized around the mainstream patterns and the divergent voices might be treated as ones that need optimizing.
When I was editing my book I got some feedback from one of the AI editors suggesting that my writing is “too intense”. Well, yeah. The feedback was something like: “it’s currently 10/10 on the intensity scale, maybe let the reader breathe”. I did not rewrite my scenes to make them softer and I did not write in “filler” chapters to let the reader “chill” (sorry!). If you’re not into that, it’s not for you.
At a different time my human written prose was flagged as AI assisted or even AI generated. So, sometimes my writing is too intense to sound like a normal human. And sometimes I just sound like a bot. Cool.
But of course everyone loves AI detectors and just like everything “AI”, it can be fun to play with but somehow it amplifies my autistic experience with the world: we’ve built tools that learn from human expression and we’re now using these tools to police human expression. My authentic communication style has been corrected, misunderstood, scrutinized so many times - this is just another layer to that experience.
While reviewing this piece, my AI editor effectively trashed my writing: the intro is too much like “terms and conditions”, the “Writing” section is “too academic” - we must simplify everything to you, the reader, who has an attention span of a goldfish, and also, my ending is too abrupt - I need to spell this out.
I’m not going to edit it. This is the ultimate example of how the burden of self-translation continues to rest on the divergent author. But you made it this far, so thank you and let me know what you think.