Human Comprehension Friction / FTI

It started because I wasn’t reading properly

We were talking about how humans reread text.

I know exactly what happens when I’m reading something boring.

My eyes continue travelling obediently across the words while my brain develops a sudden interest in solving a cryptic crossword clue from last week.

Blah blah blah. Yes, yes. Got it. Keep moving.

Then something later in the text doesn’t make sense.

Ah bollocks.

I’ve missed something.

So I go back and read the relevant bit again, this time actually paying attention.

I diagnosed this highly technical phenomenon as lazyitis.

But then I realised there was another kind of rereading.

And it wasn’t the same thing at all.

Read. Nope.

Sometimes I’m concentrating perfectly well.

I read the paragraph.

Nope.

Read it again.

Still nope.

Slower this time.

Nope.

Four attempts later, I have supplied the attention. I know the individual words. I may even understand every sentence in isolation.

And somehow the information still refuses to enter my brain in a useful configuration.

At which point the question changes.

Maybe the problem isn’t the reader.

Maybe the writing is shit.

Hang on. That’s observable.

That was the interesting bit.

The difference between I wasn’t paying attention and I am actively trying to understand this and the text is fighting back produces behaviour.

Rereading.

Pausing.

Backtracking.

Looking for a definition you think you must have missed.

Checking whether you’ve understood something correctly.

Getting progressively slower.

Eventually abandoning the whole thing.

Those behaviours could potentially be measured.

So what if you gave the same piece of writing to twenty ordinary readers and watched what they actually did?

Not just what they told you afterwards.

What happened while they were reading it?

Human comprehension friction

We routinely usability-test software.

Put people in front of an interface and watch.

Where do they hesitate?

Where do they click the wrong thing?

What do they misunderstand?

Where do they give up?

Then somebody changes the interface.

But we don’t generally treat prose that way.

Someone writes a technical document, policy, patient leaflet, official letter or set of instructions and decides that it is clear.

Possibly after asking another person who already understands the subject whether they think it’s clear.

Which isn’t quite the same thing.

Instead, line up the reading behaviour of multiple people.

If fifteen independent readers stumble over the same paragraph, the chances that fifteen brains spontaneously malfunctioned at precisely the same location seem fairly slim.

The paragraph has a problem.

Now diagnose it.

Undefined terminology?

Acronym overload?

Ambiguous sentence structure?

Too many concepts introduced at once?

Nominalisation turning perfectly ordinary actions into bureaucratic sludge?

The important conclusion buried three paragraphs after the reader actually needed it?

The system could identify the friction point, explain what was probably causing it and suggest a clearer alternative.

Then the idea got more interesting

Once you’ve collected enough examples, you have something else:

A dataset containing writing and observed human difficulty.

Which raises the possibility of teaching a model to recognise the textual patterns associated with comprehension problems.

Eventually you might not need the twenty readers.

Give the system a new document and tell it who the intended audience is.

It could predict:

Readers like these are probably going to struggle here.

Not because some readability formula has counted the syllables.

Because previous humans actually struggled with writing that behaved like this.

That turns it from a usability-testing tool into something closer to a human-comprehension model.

Obviously it required a metric

Every serious research methodology needs a robust quantitative measure.

Ours would indicate the point at which ordinary human cognitive effort exceeds the reader’s remaining willingness to cooperate with the author.

We called it:

FTI

Fuck This Index.

Low FTI:

“Yes, perfectly clear.”

Moderate FTI:

“Hang on. Let me read that again.”

High FTI:

“Who wrote this fucking thing?”

Critical FTI:

Document closed.

Email ignored.

Form abandoned.

Possibly printer thrown from window.

Science.

And then it escaped from writing

Once you start thinking in terms of human friction, you begin seeing it elsewhere.

A supposedly digital process that requires someone to receive something on their phone, visit a website, send an email, switch to a laptop, download a document, find a printer and eventually produce a piece of paper hasn’t necessarily been successfully digitised.

The work hasn’t disappeared.

It has been exported to the human.

Which suggests FTI might eventually be useful beyond prose.

Instructions.

Forms.

Administrative processes.

Digital services.

Anywhere an organisation can truthfully say “the system works” while the person attempting to use it is actually losing the will to live.

Current condition

There is something real in here.

The initial experiment is conceivable.

The behavioural signals are measurable.

The progression from observation → diagnosis → prediction has a plausible research path.

But doing it properly would require experimental design, data, software and rather more commitment than occurred during the conversation in which I invented it.

So, for now:

Still drifting.

Occasionally muttering about nominalisation.