The Knowledge in the Hands

In yesterday’s The Guardian, Dalya Alberge reported that researchers at the University of Sheffield have reproduced the colour-changing effect of the Lycurgus Cup, a fourth-century Roman vessel that appears jade green in reflected light and ruby red when lit from behind – the earliest known example of a complete Roman two-coloured (dichroic) glass object.¹

»We like to picture ancient crafts as recipes that were lost and can be found again, as though the knowledge were a set of instructions sitting in a drawer somewhere«

Despite knowing why the cup changes colour, recreating glass that does it took the scientists months of experimentation. If knowing why tells you where to look, why doesn’t it tell you how to get there? Materials technician Lisa Hollands, who led the work with Roman-glass expert Caroline Jackson, jokes that if the Roman makers could see her experiments, they’d ask: »Why does she not know how to make this?«

Hollands can name the nanoparticles and explain exactly what they do to light. She understands the physics in a way no Roman glassmaker did. In ordinary coloured glass, metal oxides are dissolved evenly through the molten mixture. This makes the glass absorb the same wavelengths however you look at it. Dichroic glass is different: it suspends tiny metal particles – gold and silver in this case – which scatter light one way and absorb it another, making the glass shift colour depending on whether light passes through it or bounces off it. All that knowledge wasn’t enough: Hollands still had to make glass, cool it, inspect it, adjust the conditions, and try again.

And then again.

In The Concept of Mind, philosopher Gilbert Ryle distinguished between »knowing that« and »knowing how«.² Skill, he argued, can’t be reduced to applying rules: if every rule required another to explain how to apply it, the process would never even begin. Knowing how to do something is a different kind of knowledge from knowing facts about it. The sociologist Harry Collins reached a similar conclusion when he observed physicists failing to build a working laser from a complete written specification until someone with practical experience showed them how.³

Analysing the original cup told the Sheffield team to look for gold and silver nanoparticles, which narrowed the search from everywhere to somewhere. What remained was a narrower, but still not trivial, problem: finding the right furnace conditions that would actually produce the glass.

Perhaps nobody ever fully mastered the process, not even the Romans. Ian Freestone, an expert in archaeological materials, notes that the colour is acutely sensitive to furnace conditions. The makers may have »hit the jackpot«, producing an extraordinary object without mastering a repeatable technique. If Freestone is right, the Lycurgus Cup isn’t evidence of a lost skill so much as a rare accident, struck a handful of times across a century, inside a process that nobody ever fully controlled. A letter attributed to the Roman emperor Hadrian describes sending a relative »parti-coloured cups that change colour« for use »at banquets on feast days«. The effect seems to have been admired, whether or not its makers could reliably reproduce it.

Today’s machine-learning systems increasingly acquire competence the Roman way – by trial, not by rule: a robot arm has learned to fold fabric into target shapes from about an hour of its own trial and error, with no human demonstration and no hand-written rule for how.⁴ If a machine can get »knowing how« without »knowing that«, Ryle’s distinction looks less like a fact about minds than a fact about method: two separate roads to some competence, neither guaranteed to find the other.

That changes what »rediscovery« means here. We like to picture ancient crafts as recipes that were lost and can be found again, as though the knowledge were a set of instructions sitting in a drawer somewhere. But what was lost is in fact harder to recover than a recipe: not an answer, but a tolerance for an answer that wouldn’t always come. Hollands’s lab can now explain, more precisely than any Roman glassmaker ever could, exactly why the cup does what it does. It still can’t promise, batch to batch, that the next piece of glass will do it too. Sixteen hundred years of physics bought her the explanation. They didn’t buy her the guarantee.

References