COVID-19 and Excel-07: when will we fix the failed education that led to yet another data science disaster?

This is the third time in short order I’ve felt compelled to blog about a basic computational thinking educational failure shown up by the COVID-19 crisis. (See numbers one on maths-model overclaim and two on exam-grades u-turn).

Today it’s the turn of using the wrong data science tools for the job. It was announced yesterday in the UK that NHS England (the body responsible for the public health service) relied on Microsoft Excel 2007 to collate the country’s COVID test data, not only for the daily case figures, but also for feeding the track and trace service too. Catastrophically, I understand it was too big for the maximum number of rows allowed so for around a week, the authorities didn’t realise 10,000s of cases didn’t show-up and they failed to trace and contact others who might have been infected.

“‘What were they thinking?’. I’d go one step further and say ‘what thinking were they educated in?’”

They described this as an “IT glitch” but its root cause is a “CT glitch” or more accurately a core computational thinking (CT) education failure amongst those responsible at NHS England, and more generally the entire population.

When one hears that they used Excel 2007 as a central part of organising the country’s critical data process anyone who knows anything about data handling exclaims, “what were they thinking?”. I’d go one step further and say “what thinking were they educated in?”

“Selection of tools is a critical part of doing a good job and nowhere is this more critical than in data science or more generally computational thinking.”

There’s the old adage “a bad workman blames his tools”, but today, with such complexity of task and so many tools at our disposal, the more usual problem is “a bad workman uses the wrong tools for the job”. Selection of tools is a critical part of doing a good job and nowhere is this more critical than in data science or more generally computational thinking. If all you know is a hammer, every problem looks like a nail. This is a human problem not an IT one; Excel 2007 wasn’t set-up to be used for this volume of data in this way; hence why anyone “in the know” viscerally reacts that that misuse was the cause of the “IT glitch”.

In the AI age, selection of the right computational machinery is a bit like selection of the right employees. It’s hard and complex and important to know how to manage the different characters successfully!

The problem is not just that the wrong computational tool may be inefficient for day-to-day use—including limiting the scope of what’s possible. The problem is that the wrong selection can lead to catastrophic failure particularly easily—because the volume of information is high, the detail of what its doing is relatively hidden, and there are layers of automation at play.

Another topic I won’t go into here in great detail, but there’s also how you verify results whatever machinery you pick…another educational challenge that is quite critical and severely failing. Stuff will break, you need to know how to get experience of how and when it does and what to do about it, itself dependent on the machinery you’ve selected.

An excerpt from the math(s) fix on learning to Verify Computation results through corroboration

Therefore, for both of these reasons, how you select your computational tool is a hard problem, something you need education in so you can build experience from a high base. You need to know what’s available, what might be useful, how to cross-examine what its failure modes might be (not just be sold on its capabilities) and crucially have confidence to look beyond what is immediately familiar to you—throughout your life. Your education needs to equip you to be confident and proactive at changing machinery as the world rapidly changes, not get stuck in the one technology that was burnt into your psyche at a fixed point in your development.

Where at school or even college are you supposed to learn how to pick the right computational machinery for the job? Have you seen this in your maths curriculum—today’s core computational subject? I mean there’s when to use a pencil, compass and protractor; or graduating to a prescribed model of graphics calculator. Far from learning how to select the machinery, an outmoded selection is made for you that’s only for use in education! (And these calculators are a rip-off too…I’ve had to spend £100 for something that’s extremely hard-to-use, slow and with very limited functionality, certainly compared to our free www.wolframalpha.com)

Where the new subject of coding or computer science is available, I’m pleased that technology selection usually does come up, but this isn’t just about IT or coding, it’s more generally about the broader computational literacy that needs to be core to AI age education. It’s one aspect of humans having power to manage AI, not being managed or fooled by it. It’s that central.

“Schools are still trying to educate people in driving a computational horse-and-cart, only to wonder why they can’t then drive with reasonable safety on an autobahn.”

The good news is that me and my team have worked extensively on this and have specific proposals. One output is our “outcomes list”—a top level summary of what you want to achieve from a new core computational education. And within it you’ll find a category “Managing Computations” including “MC1: Choosing an Appropriate Technology”. And in turn our computerbasedmath.org teaching materials are tethered to achieving these outcomes as we build them out.

If we had a mainstream core computational curriculum that matched these outcomes, mistakes would still happen, but not the cacophony of basic CT error after basic CT error that has so punctuated the last few months. Schools are still trying to educate people in driving a computational horse-and-cart, only to wonder why they can’t then drive with reasonable safety on an autobahn.

While our students are checking their hand-working in maths for a missing minus sign, out in the real world people are delegating to the wrong machinery and spreading the pandemic.

In highlighting these failures, I have picked on the UK because I live here so am in the know. But this isn’t just a UK problem, it’s worldwide. Each country has its own different way of manifesting educational CT failures; look in any detail and you’ll find them.

My book The Math(s) Fix talks in much more detail about machinery selection. The excerpt below even features Excel, though to illustrate a slightly different failure mode of catastrophic computational results rather than data handling errors.

Yup, it’s another issue that’s just waiting to expose our societal CT educational failure. Watch this space!