Automating the Pointless Argument

A short story about fake news, public discourse & structured data

Lucy’s performance review wasn’t going well.

“I appreciate you’ve done some quality work this year. That latest batch of medical research updates for example. Must’ve been hella difficult to decide what is and isn’t true for the media and the public, you sure did a good job. But…”

There was always a “But”.

“TruFact’s workload model says that as a Level 6 Fact Curator, you should be completing two packages per quarter. Now after annual reporting, you know how things are”. He paused to look at her, eyes wide open. “We won’t have the funds to employ anyone who isn’t keeping up.

“Personally I’m here to help, of course. Are there any barriers in your work environment we can help address?”

“No, that’s not the issue. It’s just that, to do a good job, I felt this package needed more time.”

“Perfectionism doesn’t always help when you’re a Curator, you know. Is there any way you could have — how to say this? Not delivered shoddy work, obviously, but you understand the public database doesn’t need the rigour that the scientific journals do?”

“People make life decisions based on the medical updates, sir.”

“They used to make life decisions based on fake news though, remember? I’m sure that, well, anything you came up with at all would be an improvement on that. Think about it. Now I can put something in under ‘mitigating circumstances’. How’s your family situation? Home life?”

“My brother Raf…”

There was a pause. “I’m afraid there isn’t a nice way to say this, Lucy.”

“You’re going to say that Raf isn’t helping my career.”

“TruFact shows that Disability Credit is sufficient to cover daily needs in 98% of cases, and 100% of cases where Reassessment was requested. You really don’t need to spend so much of your time looking after him.”

*

For years now, Lucy had been angry at her government’s incompetence. Several governments, to be precise; but their guiding principle had always been the same: the government that governs best is that which generates the least social media outrage. At least, to that end, they had all provided extensive funding for fact checking, which had kept her in a job.

Lucy was never outraged — a cool head was a useful asset for a fact checker. Her anger, triggered strictly by her own experiences, came and went in waves; today however was the end of the financial year, and nothing quite frustrates like filling out cumbersome paperwork on pain of imprisonment. She remembered how Raf had been a successful entrepreneur before the politicians destroyed his business through overregulation. Yet since his health deteriorated, the state had failed to provide for his basic needs. How could any system screw up so badly? Online, the various political bubbles would argue bitterly about what governments should and shouldn’t do; couldn’t they both agree that the main thing was not to be worse than useless?

Today, Lucy resolved to find the truth about Disability Credit. She didn’t have paywall access to the underpinning research herself, but there are things you can do when you’re a Curator. Log into the public portal. Quick search: 98%, 100%, same stats the boss had parroted. Check the URL, what’s the fact ID? Copy. Back on her work system, search version control. Paste. Commit logs signed off by Amos Judge, downtown campus. Write an email, arrange a meeting.

*

The building was high rise, with a great view of the hills behind the city, so she didn’t mind the ten minute wait in reception. Amos, when he arrived, didn’t seem to notice it though.

“Always nice to have a visitor from Lakeside. I can put that down under ‘external engagement’ for my next quarter. Now how can I help you?”

“I’m working on the next batch of medical updates”, she lied. “There are some studies on health effects in cases of insufficient Disability Credit.”

“Bit outside their normal remit, isn’t it?”

“I know!” she rolled her eyes, hoping she wasn’t overdoing it. “They must have got multi paywall funding. Alright for some, huh?”

Little could Lucy know she had hit on his pet subject.

“There are odd cases where the paywalls hinder research, of course. Still, better than back when everything was open. So many results misinterpreted outside their own fields. And that’s not to mention the press. I suppose you’re too young to remember the vaccine scares? Half of Ducktown must have died.”

“So how is Disability Credit set?”

“Well there’s the Standard Credit, then Additional for any special needs.”

“Why is Standard set so low?”

“Well, it’s linked to minimum wage you see. And you can’t raise that, or you’ll up the level of unemployment.”

“Why would that be?”

“All those low wage jobs, they don’t suddenly get more productive at the stroke of a pen, unfortunately. If an employer can only afford 5k per hour for a cleaner, well, increase the minimum wage to 7k and the cleaner loses their job. Or just gets hired off the books, without any legal protection.”

“But most employers could afford more for the job if they had to, right? They still need clean offices?”

“It’s interesting you say that, because the research isn’t really conclusive either way. Plenty of studies look at minimum wage and find a link to unemployment. Just as many studies find no link.”

“That sounds like quite a mess” she said; outwardly empathetic, internally fuming. “What did you settle on for TruFact?”

“What I said first time round, after all it makes logical sense, as facts always should. Imagine the confusion on social media if we started endorsing all those counterclaims. Hours of productivity lost in argument, feeds the demagoguery, rocks the political boat. Doesn’t do the markets any good at all. People are so tired of politics.”

“So the Verified Fact is that raising minimum wage raises unemployment; best we all agree on it and get on with our lives?”

“Exactly.”

Lucy didn’t argue. She picked her battles wisely, and there was nothing to gain in this one.

*

I’m going to stop the story here, as it’s clearly a thinly veiled vehicle for my thoughts on where naively implemented fact checking could take us, and there’s no point stretching it out any longer.

I’m sorry you’re a cardboard cutout character, Lucy, but if it’s any compensation, there’s a happy ending in this for you and your brother.

Dear reader, I’d like you to imagine Lucy and her brother escaped to a better country, one that — like the first — also uses an enormous database to help people make sense of all the conflicting information in the world, but this time allowing for nuance and disagreement.

Like Lucy, you’re probably wondering how that would even work.

*

Lucy was halfway through her training on the LogiCheck system.

“Can you help me a moment please Paul? Filling out these fields here. Source, date published, date accessed, that’s easy. ‘Supports’, those are the facts this research provides evidence for, right?”

“Not facts. Propositions.”

“Right, so where are the facts?”

Paul paused for a moment.

“Well that’s an interesting philosophical question. One for the pub maybe? We don’t document facts here, just what evidence supports what.”

“Well in this case doesn’t the question of what supports what depend on what your point of view was to start with?”

“What is it you’re looking at?”

“Minimum wage.”

“Ah, yes. Search for all propositions with that tag”.

“Ok, I’ll have to have a read of these. Where do I see which ones are taken as true?”

“You can’t. For a start we might disagree, whose opinion do we take?”

“Don’t we have some kind of accepted standard? Sample size, p values under 0.05, metastudies?”

“Sure, yes, but not everyone would agree on how important each of those things is. Or whether, perhaps, there are other ways of judging which are more important than any of them. Sure, ultimately somebody is going to be right about this, and somebody else is going to be wrong, but I doubt there’s enough research done that we can conclusively decide yet.”

“Ok, but, this work needs doing. How can I input what the research supports, without deciding one way or the other?”

“See top left there. Tap ‘New’ then ‘Deductive Reasoning’. If X and Y, then Z. I think the gist in this case is that minimum wage does raise unemployment unless there’s significant market friction, in which case it doesn’t. You’ll need to dig deeper obviously. Which parts of the argument does this evidence support? Which does it contradict?”

“Oh, ok, I get it. But how is this any use if we can’t use it for a fact check of what’s true and false?”

“Well firstly the search engines use all this stuff to train their AIs, and bring up relevant material whatever you’re doing. If you install the checker plugin, it can tell you when anything you write contradicts itself. Surprisingly often, if my own writing is anything to go by. You can throw out arguments that aren’t logically consistent without needing truth consensus at all.”

“That does sound handy.” She paused for thought. “Suppose we really need a yes/no answer though, for policy or something, what then?”

“You not signed up for an account on Axiom yet? You can put truth models in there. All encrypted, personal to you. Here, I’ll share one of mine to get started.”

They turned to Paul’s screen.

“Compassionate Buddhist Governance?”

He blushed. “I think it’s got something going for it, actually. But you can go with one of the mainstream defaults if you like.”

“Nah, send it across. I’m interested.”

*

Compassionate Buddhist Governance had the importance-of-compassion propositions all set to 100%, of course. Lucy liked the sound of that, but it could wait.

The Social Darwinist model left compassion unset altogether, in Bayesian terms a uniform prior that could take any value from 0 to 100% with equal probability. A helpful message told her that it would take a specific value when she clicked ‘Update’, but the result would vary depending on requirements dictated by all other propositions in the system.

The mainstream model had priors set, 95% certain that the importance of compassion was between 30% and 100%. Quite a wide range, then. Again, a specific value would be deduced from linked propositions.

Lucy flicked back and forth over her day’s work, watching truth values change depending on the assumptions she chose.

“On another topic, Paul, what about outright fake news? Propaganda?”

“All reported facts are auto linked to the source reliability models. Most will rank it low based on past performance of the source, methods used, corroboration, what have you. Of course not everyone accepts those models, but that’s a good thing, the models themselves would never improve without criticism.”

“Same goes for nutjob claims that the president is literally a lizard, right?”

“Yes, except they’re more easily dismissed because that would conflict with all the biological data. Of course if there was overwhelming evidence the other way…”

They both laughed.

*

The interface was remarkably slick. Which facts — no, she corrected herself, propositions — did the majority of models agree upon, then? That particular search led Lucy to Philosophy.

Proposition: There exist truths of which we can be certain.

According to LogiCheck this scored 0% true (with an exception in all truth models for various refinements of Descartes’ claim “I think therefore I am”, and other defining exceptions in some of the religious models). Really, nothing else? She supposed not, if she really thought about it.

Proposition: There is only one, unique, underlying truth.

100% true, said LogiCheck. Wait, doesn’t that conflict with the last one? Double tap, identify conflict…

Proposition: We are able to unambiguously identify the truest opinion from a set of opinions.

False, said LogiCheck. Ah, that sounded better. Some propositions are objectively true, but we can’t always tell which.

Proposition: LogiCheck is an arbiter of truth.

False, said LogiCheck. And therein lies the difference between that and TruFact, she thought.

Proposition: Provided LogiCheck is kept up to date, all information in LogiCheck is universally agreed on.

True, said LogiCheck. Hang on, I thought we couldn’t identify the truth, yet everything in this database is agreed upon?

But of course! As with any civilised conversation, we can agree to disagree. Even fanatic rationalists can be persuaded, if there isn’t enough evidence to know for sure. And when we do disagree, a new proposition is added to LogiCheck to record it. The computer helpfully tells each side what their position does, and doesn’t fit with: no firm imposition of right and wrong, but ideas that can’t fit anybody’s reality will never score highly.

Seen this way, LogiCheck was simply the next step in human communication. The last two hundred years had seen so many jobs automated; farming, computing, manufacturing, driving vehicles. LogiCheck succeeded at automating the pointless argument. Not all arguments, not the productive ones, just the nonsense based on ill-informed opinions. The nonsense that made people tire of politics. The nonsense that made people crave for one right answer, whatever the cost.

*

I like to think LogiCheck could be a future descendent of Wikidata. Does this sound implausible to you? Then consider this: numerous systems a bit like LogiCheck have already been built, but they’re not in mainstream use because building a community that cares about structured data is harder than building the technology. Wikipedia has, of course, succeeded in building a community, and in Wikidata that community has started to work on structured information.

Would you like to discuss this more? I plan to host an online discussion exercise (perhaps using Delphi). If you would like to take part please drop an email to crispin@tropic.org.uk

Aside from Wikidata, all names, characters, and incidents portrayed in this production are fictitious. No identification with actual persons (living or deceased), places, buildings, and products is intended or should be inferred.

*

What of Lucy’s brother Raf, how did he fare in this new land? I don’t want to dwell on him, because that shifts emphasis of this essay from logic to my opinions on economics, which are a distraction. The important aspect of economics, for the purpose of this essay, is that the central facts are open to interpretation.

As a footnote, however, I like to think that Raf — like every other citizen of the second country — is supported on Universal Basic Income; he also gets a bit extra for his disability needs. With UBI guaranteed, the labour market is not distorted by the desperation of those who would otherwise starve; in return, employee protections can be reduced and no minimum wage is needed at all. So, Rafael (like his namesake) passes his time by learning to paint. A year later this gains him a job at a local craft outfit, on low pay — the employer can’t afford more — but taken together with UBI, it’s all he needs, and he is happy there.

Of course you might think that Universal Basic Income would never work. That’s fine by me: LogiCheck’s truth models would record both of our positions and help us both identify whence this disagreement stems. For a review of the literature on minimum wage, however, it might direct you to The Economics Anti-Textbook by Hill & Myatt.