When Disappointment Disappears, So Does Delight

By Linn Tonstad ’03 M.A.R., ’09 Ph.D.

They’re at it again. 

The same cast of Silicon Valley characters who gave us the “ensh*ttification of everything[1] are back, threatening us with their vision of a frictionless humanity in a world where everything is known, all is optimized, no desire disappointed, and where AI even helps people pre-vet each other by means of personalized avatars before inviting them to parties

I want Pulp to make more music, in whatever style they would have developed had the band not fallen apart as a result of getting famous and getting older. But the band did fall apart, as people often do.

They promise that their newest neural networks trained on unimaginable troves of data that still fail to match the complexity of the human brain will transform (and destroy) the world as we know it. Whether or not that destruction is also a potentially species-ending event is the primary concern of highly compensated AI alignment researchers and of the hypemen (gendered noun used metaphorically) whose breathless capacity for adulation is geometrically correlated with their claims to have corrected for their own rational biases (with a certainty given in percentages derived under the aegis of a misty cloud, excuse me, by Bayesian reasoning).[2]

Hypeman, doomer, or luddite: the only three positions technology-concerned discourse permits. How boring! Any of those positions can be undermined ex ante: whatever the drastic change expected, it (at least supposedly) didn’t happen the last time such pronouncements were made. After all, TV, the internet, smartphones, newspapers, and the railroad neither destroyed the world nor saved it. And this time it will/won’t be different (please choose one). A few months ago, the hypemen were all about the AI-aided release of the “last” Beatles song. In their infinite wisdom, someone “completed” a Van Gogh; then a serious museum made available a pretend Van Gogh who will share his “mental health struggles” with you. As it turns out, AI can generate anything in the style of anyone. For no reason. 

Compulsively Singable

Oh, it seems like fun! Imagine if I could get my favorite band—the late 90s Britpop band Pulp—to keep releasing albums indefinitely. Would I want them in the style of Different Class, the album containing their one big hit? Of course not. The earlier Separations and His ‘n’ Hers were less polished, and thus more satisfying, although the songwriting on Different Class, in particular the compulsively singable “Common People” (one of the only songs I might be able to sing at karaoke without causing uncontrollable laughter or mass flight), reflects stunning storytelling ability. But what about This is Hardcore, the album in which the darkly comic sensibility that made “Pencil Skirt” one of the greatest revenge songs of all time reaches its apotheosis in “Dishes,” with its image of a not-Jesus Christ doing the dishes rather than miracles?[3]

I don’t really want AI to make music in the style of Pulp. I want Pulp to make more music, in whatever style they would have developed had the band not fallen apart as a result of getting famous and getting older. But the band did fall apart, as people often do. That’s one of our hallmarks. Another is living in a world that escapes our control and disappoints us constantly. Yet without such disappointment, our lives would have no texture, no shape. Because it is unsatisfying, Pulp’s ending is more satisfying to me than its artificial continuation would be.

Subversive Surprises

How many of you are thinking, “Another boring jeremiad from a luddite!” But I was reminded the other day how much I love having a washing machine in my NYC apartment. The last place I lived was a fifth-floor walkup with no laundry in building, so following tradition, I delivered a weekly haul to the laundromat across the street. (I moved when that laundromat shut in the early weeks of the pandemic.) Although doing the laundry in intervals between Zoom meetings is so much easier, it’s also one of the many changes that combine to insulate me from the city, its jostlings, its frustrations, its cityness. 

Properly functioning cities are full of surprises because they are full of other people. The cityness of a city can almost be measured by its capacity to deliver surprises. And our bodyminds are shaped by surprises, all the way down. The complex neural networks of our brains evolved over millennia of paying attention to: 1) a world filled with things that might kill us (the jaguar that leaps out unexpectedly, the log that turns into a crocodile); 2) a world filled with things that might feed us (the colorful berries that catch our attention almost before we realize); and 3) a world filled with other people who might also kill us, surprise us, or engender intense, overwhelming pleasure and satisfaction in us, resulting in our acting in ways that surprise ourselves and make no sense from the perspective of restrictively calculative reason. 

Most of the time, the brain functions as a predictive-processing system upholding a sense of continuity while filtering out the vast majority of the information it receives.[4] Thus, the default—the assumed, average, most likely options—have their place not only in large language models (LLMs), but also at the center of our lives. But the surprise that is good for us, and that has its source in other people, matters differently. The surprise of finding ourselves having forgiven someone. The surprise of having made assumptions that turn out to be deeply, fundamentally wrong, and the delight we might find in that discovery (once we get past the embarrassment). The turn in a line of poetry that could neither be said in another way, nor can its saying be accounted for. A twist in a musical phrase or a video game storyline that we had not at all anticipated—and where what matters about it isn’t simply the surprise, but the questions the surprise invites us to consider, questions that matter to us because people matter to us.

A World in Need of Our Defense

The hype around AI therefore worries me. Online is already flooded with bottomless garbage, and we are only at the start. While these floods predate the onslaught of generative AI, the overwhelming scale of that onslaught will only increase: generic music, information that isn’t, boring remixes of art and authors. Maybe this onslaught will generate its correlative opposite, bringing increased interest in the local and ephemeral rather than the generic found everywhere. But what is other than generic is harder to make, and harder to access. 

Drastic decreases in time spent with friends in recent years (with the greatest decreases preceding the pandemic) can be placed side by side with research showing that four of the five most enjoyable ways people spend their time involve the bodily presence of other people. (The fifth is sleep.) AI makes the many forms of no-human-involved entertainment that can be accessed in the home even cheaper, while inflation makes everything else more expensive, and hypergentrification means that the venues people might go to to spend time in the presence of other people are continually under threat due to rising rents. The result is, in effect, a zero-sum game: cheaper home entertainment crowds out the more satisfying (and sometimes more expensive) entertainment that requires effort to access, but that offers the life-altering satisfactions of the disappointments and surprises that come with living in a world with other people in it. That world needs our defending, disappointments and all.

Linn Tonstad ’03 M.A.R., ’09 Ph.D. is a constructive theologian working at the intersection of Christian theology and queer theory. Her books include God and Difference: The Trinity, Sexuality, and the Transformation of Finitude (Routledge, 2016) and Queer Theology: Beyond Apologetics (Cascade, 2018). She joined the YDS faculty in 2012.

1. This was named the 2023 Word of the Year by the American Dialect Society. Popularized by writer Cory Doctorow, it describes how digital platforms (and much that surrounds them) become less and less useful.

2. Bayesian reasoning calculates probabilities based on available knowledge. It is a most wonderful tool. But it can offer misleading confidence in situations of low probabilities and high levels of complexity with myriad underspecified variables. Consider AI researchers and hypemen’s wildly varying estimates regarding how soon to expect the arrival of Artificial General Intelligence. For more on these debates, see https://www.astralcodexten.com/p/in-continued-defense-of-non-frequentist (in defense) and https://www.benlandautaylor.com/p/probability-is-not-a-substitute-for (in critique). The accuracy of Bayesian reasoning is also strongly affected by how information is presented in the first place. 

3. I’ve known the lyrics to “Dishes” by heart for more than 20 years. But I still make sure to double-check them on lyrics.com. My iPad is set to a European location and therefore allows me a modicum of control over the expanded surveillance that using the internet now requires. I scrolled through 27 options and moved six slider bars to the left to reject the 830 vendors wanting to install trackers based on that single click.  

4. For more on predictive processing or predictive coding, see Andy Clark, “Whatever next? Predictive brains, situated agents, and the future of cognitive science,” Behavioral and Brain Sciences 36 no. 3 (2013), pp. 181-204. In this regard, some versions of AI and the human brain have much in common.