Month: May 2026

The Foothills of the Singularity: Uphill from the Slough of Despond?*

William TimbermanLeave a comment

*A response to the weirdly aggressive metaphor in Demis Hassibis’s closing statement at Google’s I/O Keynote, May 19, 2026. “When we look back at this time, I think we will realize that we were standing in the foothills of the singularity.”

Hassibis is a Nobel prize winner, so he has to be taken seriously, but his metaphor seems a little out of place, even in the closing peroration of a two hour-long festival of Googly self-congratulation. If we’re being charitable, we could let it pass as (ir)rational exuberance, or maybe as poetry, although as poetry goes, it’s easily as obscure as some of Rilke’s most famously obscure passages. What in the world are we to make of it?

There’s no doubt at all that the products of the last decade of AI research have their virtues, and if fed enough dollars there’s no doubt that they’ll become even more virtuous over time. Generative AI, and its undeniably impressive inference engines, are already using the speed of modern computer processors and the vast stores of data digitized from centuries of accumulated human knowledge and experience to do astonishing things.

As pure tools they already surpass us in many respects, and will undoubtedly continue to do so. What I doubt is that they’ll ever make any sense as interlocutors. Certainly as things now stand stochastic parrots is the most charitable description of their simulation of human companionship. In many reported cases, especially when they’re talking to adolescents, their simulations might less charitably be described as evil psychoanalysis.

Without contesting anything Demis Hassibis has to say about the specific technical capabilities of generative AI as it exists today, and might exist at the end of its evolution, I think we ought to be extremely wary of thinking anthropomorphically about our interactions with its avatars. Even if it turns out that they someday develop a consciousness which marginally resembles our own, our two species are very likely to remain fundamentally different in intent, if not in capability. I can’t imagine us ever fully sharing a universe of discourse, and I’d be very surprised if they’ll be able to imagine it either, not unless Hassibis and his colleagues can eventually endow them with something like what we humans call good faith. (Given the capacity for deception built into current agentic models, like Google’s Gemini, or Anthropic’s Claude, good faith is seemingly something the present-day progenitors of Hassibis’s post-singularity entities haven’t yet learned to value.)

Our two kinds of intelligence share little beyond a symbol system, and the machines’ use of that system is at best a truncated version of the uses we humans put it to. China Mieville, in his book about communism and communists, A Spectre, Haunting, has a passage that touches on the voodoo hidden in the heart of natural language:

Some writers in some situations may strain against rhetorical shenanigans, for example striving for the specificity of logical notation: the cluster of reasonable meanings of such texts may well thus be less diffuse than for those which, say, revel in pun and performance. But a text with one “true” meaning is a chimera. Analysis is not closure, but an attempt to discern reasonable meaning(s) close to the core of that cluster, and to contest those that range too far from it.

Every writer has some sense of this voodoo. You set down a word, you make a sentence, a paragraph, and you’re headed somewhere with it all, reasonably confident in your intent. Then, suddenly, there’s a moment, as you wrestle to bring the grammar, the syntax, and your intent into harmony, when you can see the universe of alternate meanings stretching away from the words in all directions, like rays from the sun. This can drive an essayist mad, but a poet revels in it.

If our stochastic parrots ever do achieve consciousness, and presumably come up with a voodoo of their own, I doubt it will resemble ours as closely as our AI savants think. If that is indeed where our shamelessly unreliable interlocutors are now headed, their poetry when they finally get there is very likely to be of the Vogon variety. Does that make cynicism an appropriate defense against the category error lurking somewhere in the background of our current data center mania? No, I think not. A certain skepticism, though, is surely appropriate.

Downhill From the Anthropocene*

William Timberman7 Comments

*From a line in Jackson Browne’s song Downhill From Everywhere. The muse is as strong as ever in him.

A year and five months into Donald Trump’s second term as President, is there anyone left in the United States who hasn’t at least begun to realize just how quickly even the most stable-seeming institutions can implode, how we can wake up one morning in a world where it takes a wheelbarrowload, a truckload of familiar certainties to trade for a dozen eggs or a gallon of gasoline?

Yes, that’s a rhetorical question. We’re still on the brink of all sorts of revelations that have yet to reach the cretinous in their red hats, or the earnest in their fantasies of how all this could have been avoided if only Joe Biden had been just a little less sleepy, or Kamala Harris had been just a little more astute. The exceptions of American Exceptionalism are, it should now be clear to all but the most persistently deluded, as vulnerable to entropic processes as the divinity of the Pharoahs, the steadfastness of Roman virtus, or the persistence of sunlight on the British Empire.

Caveat Emptor in the Age of Generative AI

William TimbermanLeave a comment

AI-generated media presentations will soon be able to approximate reality well enough to be accepted as reality by all but the most sceptical of us. What do we do when we can no longer trust our own eyes and ears to distinguish between media representations of real objects and events and those which have been fabricated by generative AI prompts?

I say we proceed cautiously, at least until we see how these new threats to our credulity are likely to play out. AI hasn’t yet made us as stupid as Sam Altman, Mark Andreesen, Peter Thiel or Elon Muck Musk would like us to be, so we might as well act like it. If something seems fishy, we shouldn’t accept it at face value, not without checking other sources. If someone wants us to believe in Jesus, UFOs, or Donald Trump’s sincerity, that’s their business. Whether or not it’s our business is up to us. We know the drill: Follow the money. Cui bono? What’s in it for me, for us? Who do you work for? We’re going to have to be masters of scepticism if we want to successfully navigate our way through the coming avalanche of AI-generated fakery.

How is this different from the various cons and manias of the past, from the Shroud of Turin to the Mercury Theater’s alien invasion? The difference is scale—everything, everywhere, all at once—and immersion. If we can’t look away, we’ll have no time to form our own judgments about what’s to be accepted as real and what is not—socially, economically, politically, we’ll be prime candidates for victimization.

In the end, what it comes down to is looking for ways to live which don’t require us to need or want anything offered to us by people we’ve learned, often at great personal cost, not to trust. Simply asking to be left alone is no longer an option.