September 24, 2026

The Conservator’s Case for Waiting

Curated and edited by John Pile Jr 6 min read

How this article was developed, including AI’s role

Watch enough art conservation and a pattern starts to emerge.

The conservator is often dealing with two problems at once: what time has done to the artwork, and what previous people have done to it.

An old varnish may have discolored. An adhesive used in an earlier repair may now be difficult to remove. Retouching may cover original material. A treatment that once seemed reasonable can become a problem decades later because it cannot be separated cleanly from the thing it was intended to preserve.

Modern conservation consequently puts considerable thought into whether an intervention can be reversed or safely treated again later. The concern is not simply about adding something or removing something. Either can permanently change what remains.

There is an unusual humility embedded in that practice:

The next person to work on this may have better tools than we do.

That person may also know something we don’t.

So preserving an object can mean more than keeping it intact. It can mean avoiding an intervention that prevents a future conservator, researcher, or technology from discovering something that is inaccessible today.

Sometimes that requires accepting that valuable information exists but cannot yet be safely reached.

When the information is there, but we cannot safely reach it

For more than two centuries, scholars have possessed a remarkable collection of ancient texts that they could not read.

The Herculaneum papyri were carbonized by the eruption of Mount Vesuvius in AD 79. The scrolls survived, but many became extraordinarily fragile cylinders of carbonized material. Early attempts to physically open some of them damaged or destroyed portions of the very texts scholars were trying to recover.

So many remained closed.

That meant accepting an uncomfortable tradeoff: the information might be inside, but retrieving it with available techniques could destroy the source.

Then the tools changed.

Researchers learned to use X-ray tomography to image the internal structure of unopened scrolls. Computational techniques could reconstruct the layers and virtually flatten them. Machine-learning systems could help distinguish ink from the carbonized material around it.

In 2026, researchers announced that one Herculaneum scroll had been virtually unwrapped and read from beginning to end without physically opening it.

The people who left that scroll unopened did not know what technology would eventually make it readable.

They only had to preserve the thing that future technology would need.

We did something similar with recorded sound

Some of the earliest surviving sound recordings present essentially the same problem.

A historical recording can contain recoverable sound while being too fragile or damaged to play conventionally. A stylus retrieves the information by physically following a groove, but physical contact can also damage the object carrying that information.

Researchers eventually developed another approach.

Systems such as IRENE photograph or scan the groove geometry without playing the recording. Software then reconstructs the motion a stylus would have made and converts that geometry back into sound.

Recordings that could not safely be played can be heard because retrieving their information no longer requires consuming the artifact in the process.

The scroll and the recording are very different objects, but they expose the same asymmetry:

Future tools can extract new information from a preserved source. They cannot extract information from a source we already destroyed.

Computing gives this distinction a name

The difference between lossy and lossless compression makes the same idea unusually precise.

A lossless transformation retains enough information to reconstruct the original exactly.

A lossy transformation intentionally discards some of it.

Imagine recording a temperature every minute. At the end of each day, the system replaces those 1,440 measurements with a single daily average.

You can make ten backups of that average. Store copies in multiple data centers. Verify them with checksums and preserve them perfectly for a century.

You still cannot determine whether the temperature stayed nearly constant all day or swung wildly between extremes.

The backups preserved the output perfectly.

The destructive operation had already discarded the information.

This is an important distinction because backup and preservation of source information are not the same thing.

We can have extraordinarily good backups of destructively processed information.

Photography makes the distinction visible again

Modern digital photography adds another variation.

A phone photograph is not simply the values its camera sensor observed. Those observations pass through substantial computational processing before becoming the image we see.

In 2026, Apple introduced Reference Image for the iPhone 18 Pro. In its optional Reference mode, the phone captures signed sensor data alongside the ordinary photo. Private Cloud Compute can develop that data into a separate reference image whose integrity is protected by dedicated secure hardware.

The useful photograph can continue changing while evidence derived from the original capture is preserved.

Other technologies approach the problem from another direction. Content Credentials can preserve provenance about a digital asset: where it came from and what happened to it as it was modified.

These systems are solving somewhat different problems, but together they highlight two things that can disappear during transformation:

the source, and the history connecting the source to what exists now.

Sometimes destruction is exactly what we want

None of this means that destructive processes are mistakes.

Lossy compression is enormously useful precisely because it throws information away. A machinist turns a block of metal into a useful part by removing material. Scientists aggregate enormous datasets because retaining and repeatedly processing every observation may be impractical. There can also be excellent privacy and security reasons to deliberately destroy source information.

And eventually a conservator may decide that an irreversible treatment is justified.

The interesting distinction is not between good preservation and bad destruction.

It is recognizing when we are crossing from a process that creates another representation into one that eliminates something the representations cannot reproduce.

That distinction can change how a decision looks.

A researcher deciding whether to consume the last portion of a physical sample is making a different decision from one running another analysis on an unlimited dataset.

An archaeologist excavating an object is not merely finding it. Excavation permanently changes the spatial relationships that may themselves contain information, which helps explain the profession’s extraordinary attention to recording context.

A photographer deleting RAW captures after exporting finished images may be preserving exactly the photographs wanted today while eliminating information that tomorrow’s image-processing tools could have used differently.

And sometimes the best available technology simply is not good enough yet.

The conservators who left an unreadable scroll rolled up could not know that X-ray tomography, computational geometry, and machine learning would eventually make it readable. They only knew that opening it with the tools available could destroy it.

Sources

Featured image: A conservator working on a portrait. Skagens Kunstmuseer via Wikimedia Commons, CC BY-SA 2.0.