Be More Human

Shoutout Bill O'Reilly?

Image created with ChatGPT

In a recent interview on how to prepare for advanced AI, Tyler Cowen’s top-line advice boils down to this: Be more human.

“Be more human” sounds so simple—a cliché, really—but what does it actually mean? In a labor market where AI can complete greater and greater numbers of tasks previously done by people, what people-shaped holes will remain?

Let’s take as a given that AI will dramatically change the economy, and that it will continue to improve at what it’s already good at—admittedly big assumptions. Let’s try to distill a definition that avoids the invocation of a soul or spark or X factor.¹ In practical terms then, to “be more human” is to emphasize that which is:

• Synchronous

• Social

• Embodied

Synchronous

Focusing on synchronous skills is the low-hanging fruit for most knowledge workers worried about AI replacement; simply put, this means meetings, presentations, and conversational ability.

The key insight is to remember that most contemporary jobs involve a combination of both knowledge AND communicating/applying that knowledge. AI supplements the first half of that equation much more so than the second half. If you’re a doctor, for example, AI will be able to make better and faster diagnoses than you because it has access to millions more examples than any human practitioner can see in a lifetime; the best doctors in the year 2040 will, even more so than today, be those who can demonstrate empathy and help patients think through tradeoffs in real time. If you’re a teacher, AI may be able to take on more of the grading and lesson-planning work to free you up as a coach/facilitator or presenter. If you work an email job, focus on what happens at the water cooler.

Of course experts will still need to know things—and indeed know them quite well to move at the speed of conversation. But the aim of this knowledge is not to surface hidden connections through deep monastic focus. (If there’s one thing LLMs are good at, it’s identifying patterns from a big body of data!) Instead, the knowledge should serve the practical end of increasing clarity, resourcefulness, and ability to improvise.

Social

Virtually all synchronous skills are a subset of social skills (see above), but not all social skills are synchronous.

First, some aspects of self-presentation are social even when they are one-way or separated in time from their audience. I’m thinking here of factors like voice, aesthetic, and personal brand. AI is a decent mimic, and can often produce convincing art or writing “in the style of” a given person—but by default its outputs are generic. This means that to elicit more distinctive, personalized work from generative AI, one of two conditions must be met: Either a) the style must be a copy or combination of existing ones; or b) the prompter must have the inner compass to evaluate/tweak the outputs (and the vocabulary to describe what they want).

In other words, taste is socially mediated because it is downstream of culture. We learn what is chic, crunchy, or cheugy in reference to the people around us; and the people around us understand the cultural signals embedded in our style through the same mechanism.

Additionally, there are some forms of expression which AI might have the technical ability to create but whose inherent purpose is to make a connection between two minds. Again, this is somewhat applicable to style (which is meant to convey something authentic about the individual behind it). It is even more the case with inherently social communication (synchronous or asynchronous) such as thank yous, apologies, and condolences. In a world where virtual assistants were drafting and sending these missives without the sender’s awareness, they would lose almost all meaning. Your bereaved friend doesn’t just want a string of kind words; they want kind words from you.²

Embodied

Manual labor is in some ways the most obvious form of work that LLMs cannot perform. Without ears, eyes, and hands, current AI are unable to interact with the physical world except when trained on very specialized tasks (e.g., surgery or driving). We’re a long way from robots that can make ceramics or carpentry.³

In other instances, work is embodied because we are in a sense paying for the bodies themselves. This is most literally true for sex work, but also applies to athletics, dance, theater, and all kinds of live performance. Even if we could build a robot capable of hitting dingers off a 100 mph fastball, it’s hard to imagine anyone would want to watch it play.⁴ We already have audio technology which can faithfully reproduce the acoustics of musical artists—yet for some reason live concerts are as popular as ever.

In a deeper sense, all of our cognition is “embodied” because it bottoms out in physical perception. Sure, given enough data and examples, an AI might be able to design a solid blueprint—but it is the architect and builders who must translate from two to three dimensions; this is the intelligence required to navigate the gap between abstract model and messy reality. That gap exists everywhere. And whatever fancy words or numbers we might now use, there was a time when we all learned to point and count on human fingers. If I’m honest, the reason I ordered this post the way I did was mostly just because I liked the rhythm: Synchronous. Social. Embodied.

Each time AI demonstrates ability on par with human experts, it’s easy to feel the tug of despair. As someone who finds meaning and self-worth in forming sequences of words, this despair is sometimes tinged for me with anger and resentment at the technology which makes writing look effortless and cheap. And as we continue to surpass new benchmarks, there will of course there will be winners and losers in the economy—and existential questions we all must answer:

Source

I think for most of us, though, the impact of AI will not be to render our work obsolete but rather to rebalance what is most relevant. Faced with machine supplementation, the right attitude is not rage or nihilism at what is being taken away; it is a renewed embrace of that work which the machine cannot yet replace—those which pulse with flesh and blood, which foster real connection in the moment. As a young Bill O’Reilly famously said: FUCK IT. WE’LL DO IT LIVE.

1. To be clear, Cowen does not use these wishy-washy terms; synchronous, social, + embodied is just an attempt to impose some higher-level categories on the predictions of him and people like

Hollis Robbins (@Anecdotal)

.

2. Speculation: Perhaps we’ll see a renaissance of hand-written notes as a signal of authenticity? (At the very least, this indicates someone took the time to transcribe whatever the LLM came up with!)

3. Note that this doesn’t mean that AI can’t reason about physical systems—as evidenced by feats such as modeling proteins and more everyday tasks like creating recipes or music. These outputs are ultimately for embodied beings, but a machine is doing most of the cognitive heavy lifting. (The success of AI in those domains is, as usual, the product of an immense body of training data—even though it has no direct sensory experience of the world those data represent.)

4. A telling contrast to this example: robot umpires.