The moon before Mars

I’ve been to Cape Canaveral twice, and what stayed with me wasn’t the Mars ambition, which is everywhere you look. It was the moon.

Isaac Newton published the Principia on 5 July 1687. Apollo 11 landed in July 1969, 282 years later, and the trajectory work that got it there rested on Newtonian gravitation. Knowledge nearly three centuries old, underpinning one of the defining achievements of the twentieth century.

And here’s the part that struck me as an operations person. After decades of Mars being the headline, NASA’s strategy now goes back to the moon first. Artemis II flew in April 2026, the first crewed lunar flyby in more than fifty years, with Artemis III targeted for 2027 and surface missions beyond that. The stated rationale is explicit: establish a long-term presence at the moon and demonstrate the technologies there, in order to prepare for human missions to Mars.

Not because Mars stopped mattering. Because you prove the fundamentals somewhere closer to home first.

The moon is your process and your data. And if you skip it, you don’t get to Mars faster. You just fail further from home.

What the data foundation actually is

Every AI application depends on data, and data quality in most laboratories is poor. Not through negligence, through the fact that it was never designed to be used at scale by anything other than the person who created it.

Supplier product descriptions vary between suppliers. Part numbers aren’t normalized. Inventory records are partial. Spend sits in disconnected systems. Compliance documentation lives in file shares nobody indexes.

This isn’t a hunch. The Pistoia Alliance’s Lab of the Future survey in 2024 asked 200 life science professionals what was blocking them from adopting AI. Low quality and poorly curated datasets came out as the top barrier, cited by 52%. Fifty-nine percent said they couldn’t access the data they needed. Fifty-four percent were blocked by unstructured data, 48% by a lack of metadata standardization, and 38% said their data didn’t meet the FAIR principles, findable, accessible, interoperable and reusable, which have been the accepted standard for research data since 2016.

Read that list again and notice that not one item on it is an AI problem. They’re all process and design problems, sitting upstream, and they won’t be solved by anything you buy from a vendor with intelligence in its name.

Fixing it is a process problem

None of this is an AI problem and none of it is a data science problem. In PURE terms it’s straightforward, if not easy.

Purpose: what data do we need, and what for? Understand: what do we actually have, and what state is it in? Reimagine: what would our data architecture look like if we designed it today, knowing what we now need it to do? Engineer: build that, starting with capture at source, because every downstream fix is a workaround for something that should have been recorded correctly when it happened.

The organizations that will get the most out of AI in laboratory operations won’t be the ones with the largest AI budgets. They’ll be the ones that normalized their supplier data, connected their systems, and made sure information was entered once by the person who understood it.

That’s the unglamorous work. It’s also the moon. And Newton’s numbers were sitting there for 282 years before anybody needed them badly enough to use them properly.