Love It, Loathe It, and Lift It: The Missing Conversation on AI
I’ve sat through perhaps too many conversations on AI this month. Whether in-person (hi Outlier), online (oh, the discourse), or anywhere else I happened to be, GenAI is specifically on the minds of so many. Perhaps I don’t engage as much because I embody a unique tension point: someone who has actually used GenAI in products, but who would be labelled a hater for sure.
Except I don’t hate AI, lest of all GenAI.
I hate how we deploy it. I hate how we dismiss all voices of concern and label their speakers difficult (usually the people conveniently most affected by the worst of AI…). I hate how we remove democratic ideals from products with AI, usurping an endless amount of data into the system at all costs (want to opt out? TOO BAD!). I hate how we’ve rushed it to production, like so many other (broken) things, solely for the sake of having it out on the market. I hate how we push binary arguments to distract from the bigger conversation (more on this later).
What I hate most, lately, is the “horse is out of the barn” argument. Every horse trainer I’ve ever watched has never just let horses that escaped linger for long. No, folks, the horse eventually ends up back in the barn and “reined in.” Perhaps, it’s also because this argument – the equivalent of “everyone’s doing it” – is a logical fallacy and one commonly used to continue to justify bad decisions worldwide.
We can have a choice, several actually. But all them start with listening – really listening and immersing ourselves as much as possible in the worlds of those typically labelled “haters.”
Revisiting our approach
The rushed and prolific deployment of AI didn’t happen by accident. Rather, we’ve steeped this tea for some time with a variety of flavors to get the cup we have today.
The modern economy shifted from a circle that included community, customers, workers, and so many others as constituents (see stakeholder capitalism) to what we recognize today – the shareholder only focus. Progressively, companies pushed for higher, constantly growing profits with commitments to other stakeholders waning. Yes, there is work to change this. Status quo remains shareholder centric.
Tech took this economic shift a step further, codifying “move fast and break things” as the holy grail. Disruption became the en vogue term. Now some two decades later, we are living with the effects of endless disruption and broken things. Parents of toddlers know this type of fatigue well – it’s why you see us flopped over on the couch while oatmeal drips from the ceiling onto our faces. We blink, but do nothing more.
Intersect these factors with what Dan Sinker identifies as a problem around caring. Business as usual and fatigue fuels the failure to care. We give way to the pressure to create, to consume, to participate, and to generally create more to where we don’t absorb. Nothing matters when too much matters all at once – it’s death by overwhelm fueled by apathy.
All of this faces a head-on collision with growing autocracy and declining democracy (source, source, another one…). Building under democratic norms attempts to accommodate more people. Building towards autocracy solidifies power for the few. We see this with the complete removal of the ability to opt out, allowance of direct theft, as well as with the hard-mode discourse of “get on board or get out.”
Within this swirl of overwhelm sits conversations around AI, specifically Generative AI. People are described as “loving it” or “hating it” with little room for nuance. These tools can create interesting things. They also add to the overwhelming slop that has completely dominated the internet (see, for example, this project preserving the pre-AI internet) at a very high cost. Much like kudzu in the southern US, we initially curated this slop by hand, but have used AI to absolutely proliferate it to new levels (“invasive species”).
The challenge with the love-it-hate-it (binary) type discussions is they completely remove the willingness to integrate accountability. I can love the idea of certain types of text being scalable while recognizing the severe impacts on climate, economic disparities, and further codification of discrimination and power imbalances. For me, these problems aren’t abstract, but have real faces tied to them, people that I feel accountable to, and communities that matter in my life.
And perhaps that’s the difference.
Beloved economies ask us to think differently about how we work, buy, and sell. They ask us to prioritize relationships, to give time to a process, and to center justice. It’s the exact opposite of what so many of us associate with the way of business today.
The loveless economy lets us live apart from those we build “for” – these abstract humans that exist somewhere that happily consume lists of books that don’t exist because they won’t buy said book in the first place. They’re as flat as our drawings, stick figures with no heart, no mind, and no relation to us. They don’t matter because they’re not real in the first place. Thus, we have no moral conflict.
The beloved economy instead demands we immerse ourselves with our constituent users, to see them, hear them, but perhaps more importantly, dare to be with them as an extension of ourselves. This immersion risks blurring the line – bias, we may gasp loudly! It also gives people we can experience with all of our senses. It morally ties us to people. What we do to one, we do to ourselves. To Dan Sinker’s point, we care – deeply, radically, and fully.
Yet, when I was interpreting, embedding within the community was exactly what we did. In order to interpret effectively, we had to understand both cultures in the first place – not just superficially, but to understand the nuance, the framing, and thought processes of a culture as much as we could. We needed to get outside of ourselves and to really live with our second language as much as possible. Doing this meant being able to understand more than just the words, but to finally be able to unwrap the cultural envelope that really held meaning.
You can’t have the good without the bad. Responsible practice demands looking at the concerns and addressing it: how this affects the carbon footprint, who this discriminates against and in what ways, how it fuels disinformation, and so on. By acknowledging the bad, we can potentially build loops to correct it. We can also decide where not to use it.
Click to unfold side note: discrimination
Often times when we discuss discrimination, we rush into various accounts of “but I’m not [discriminatory against facet]!” We build tools in societies that favor and uplift certain populations and make it much harder across so many other facets. By assuming your tool is [racist, sexist, ableist, etc], you can help correct it (anti-__-ist stance).
Unchecked optimism leads to recklessness. Our pessimists are often canaries – far more aware of the risks of the coal mine than others who can endure longer, but not indefinitely. Despite what seems to be the song du jour, we are all interconnected.
To badly paraphrase Octavia E. Butler: there isn’t a single solution against these fights, but you can be part of some of the solutions.
The TL;DR Lift It Guide
People tell me they want actionable ideas. Here’s a very incomplete list of considerations.
Assume the worst of what you create. Have others help you understand how and where it can go awry. Channel your inner dystopian writer and imagine this so many years out in the future. (See Octavia E. Butler’s advice.)
Recognize this practice doesn’t fix everything. Iterate, iterate again, and continue to recalibrate with feedback. Learn more and, yes, iterate further.
Deploy slower. Or as Mike Cisneros mentioned in his talk at Outlier: move slow and fix things.
Create connection. Put a face to the people you’re designing for. Know them, ideally love them in some way, and really hear and feel their concerns.
Create friction. Opt out, stall, or otherwise add runway if reasoning doesn’t work. This works really well when you pair it with connection. You have power. Connection is motivation.
Shrink it. Deploy smaller, in beta, and actually test before mass-deploying. Build feedback loops and the ability to opt out. Listen to those opt outs.
Use proven models. HuggingFace, for example, provides some vetted tools. This is still not a full fix. Test and do the above.
Look at the labor and ask hard questions. Hayao Miyazaki was notoriously against AI. Really consider why his art style was stolen against his will. My book, along with so many others, is helping feed Meta’s AI. Trust me, I was not paid enough for this slap. Ask those uncomfortable questions about what’s valued, paid for, and celebrated.
See the data (and problem) in context. I’ll hand this to another Outlier talk with Chani Wisdom dropping this quote as one of the most profound comments I’ve heard in some time. She presented with Obinna Iwuji and you should absolutely follow both of the people. This is that immersion piece. It changes you.
It’s hard to believe it these days, but this post was human written and intended solely for humans. I spent far too much time misspelling words, fixing grammar mistakes, and moving paragraphs around. All typos and grammatical glitches mine. I am also still – for now – a 3D human, as the folks at Outlier can attest to.