Artificial Intelligence

Two Hillsides, Twelve Days Apart

Aerial view at dusk of a forested hillside stripped bare and cut into curving stepped earth terraces of raw red soil, tracked machine marks across the ground, with a drawn-down reservoir below exposing a pale ring of cracked shoreline.

On 25 July the World Heritage Committee, sitting in Busan, inscribed the Jingdezhen Handicraft Porcelain Industry Sites — China’s sixty-first World Heritage property, the first industrial heritage site China has ever nominated, and the first anywhere themed on porcelain. It was one of twenty-five inscribed that session. Twelve days later, a report described a hillside above Visakhapatnam stripped to red earth and terraced into steps for a $15 billion data-centre campus.

I have spent two days looking at pictures of both hillsides. What holds me is not the rhyme between them. It is that we have a fully developed language for one and almost nothing for the other.

Extraction language with the tense changed

Read the Jingdezhen citation closely. Deposits of porcelain stone and kaolin in the hills. An extensive water transport network. Kilns, workshops, guilds, merchant houses, a specialised division of labour running from the tenth century to the nineteenth. Every clause is a claim on land, water, fuel and people. Set those same sentences beside a live operation and they would read as an impact statement, and somebody would be contesting them in a courtroom. Set them beside a system that has stopped, and they read as praise.

Heritage prose is extraction prose with the tense changed. It is precise, unsentimental, admirably specific about tonnage and hydrology — and it is calibrated, structurally, to arrive after the fires go out. We are extremely good at describing an industrial landscape once it can no longer surprise us.

The one time we had a live genre

There is a famous exception, and it is instructive in a way I did not expect. Philip James de Loutherbourg showed A View of Colebrook Dale by night at the Royal Academy in 1801. The Bedlam furnaces, flames and smoke boiling out of the centre of the canvas, the gorge picturesque and moonlit on either side, figures in the foreground hauling raw material and pig iron. It was painted while the thing was burning.

And look at what the picture does with the burning. It puts the furnace exactly where a sunset goes. The composition converts an industrial process into a weather event — something that happens to a valley rather than something a company is doing in one. It is a magnificent painting and a considerable evasion, and it has been reproduced ever since as the image of the Industrial Revolution, which tells you the evasion worked.

The Ironbridge Gorge was inscribed as World Heritage in 1986. A hundred and eighty-five years separate the picture from the citation. Two registers for one landscape — the sublime while it lives, the inventory once it is dead — and essentially nothing occupying the middle.

Gigawatts, not hectares

The AI landscape currently has only the first register on offer. The Architectural Review‘s July/August issue is given over to industry, and in it Hans Larsson describes an AI urbanism measured, in the editors’ summary, “not in hectares but gigawatts,” with architects who “haplessly observe these new industrial outposts taking shape, guided by forces far beyond their control.” Hapless observation is what de Loutherbourg was doing too. He simply had a market for it.

What is filling the descriptive gap at Visakhapatnam is not pictorial at all. It is procedural. The city of 2.5 million receives roughly 410 million litres of water a day against demand of 480, and rations accordingly. The Kambalakonda sanctuary, leopards and pangolins, sits 860 metres from the site. The state has guaranteed the project’s water for twenty years and denies fast-tracking; Google points to advanced air cooling. Several challenges are before the environmental court.

And there is the banner people have been carrying through the city: We cannot drink DATA. Three words that do more landscape criticism than any aerial photograph of a server hall I have seen this year, because they name the resource conflict inside the frame instead of admiring the light on it.

What I think, and where I could be wrong

My guess is that the honest description of a living industrial landscape is procedural rather than pictorial — allocation tables, drawdown curves, the petition and the reply — and that nobody makes it as art because it does not reward looking. The picturesque rewards looking. That asymmetry, rather than any bad faith, is why the sublime reliably arrives first and the inventory arrives when the site is safely inert.

Where I might be wrong: the photographers currently shooting data centres from the air may be doing de Loutherbourg’s job in good faith, and a century from now their pictures may be as load-bearing as his. I do not think so. But I have been quietly sneering at that photography for months without ever arguing the case, and I notice that the argument I have just made would, if I am consistent, apply to them exactly as it applies to him.

I once spent an afternoon walking Bicocca, out past a converted Pirelli shed to a supermarket built where something else had been, trying to work out from the ground what the district had been for. Nobody painted Bicocca while it was making tyres either. The pictures came afterwards, when the shape was safe to look at.

The Andhra Pradesh High Court is scheduled to hear the public interest litigation on 24 August. Whatever is written in that room will be the most detailed account anyone produces of that hillside while it is still being made. I intend to read it as landscape writing, which is not what it is for.

Sources

More Stories