TikTok Psychopath Test
ExplainThisPaper was incepted and happened because of this newsletter back in May 2020.
We've received some $$$ to try and evaluate if short, plain English, emoji-filled summaries of medical papers actually work...
You'll basically get access to two bits of medical AI research which haven't been published yet. One short summary and one paper. We'll look at how you interact with both, and ask you some quick questions. We're giving away 4 x £50 Amazon vouchers.
Would be super grateful if you could have a go 👇
In this study, you will be presented with a full research paper and a short research summary. These cover two different studies...
When Instagram Died
Remember when Instagram (et al) changed?
We used to have chronological timelines. Then they f**cked it all up and introduced algorithmic timelines.
People hated it. I hated it. Except it was better.
When we first launched in 2010, Instagram was a single stream of photos in chronological order. But as more people joined and more was shared, it became impossible for most people to see everything, let alone all the posts they cared about. By 2016, people were missing 70% of all their posts in Feed, including almost half of posts from their close connections. So we developed and introduced a Feed that ranked posts based on what you care about most.
// (from Instagram.com).
Algorithmic timelines are better for content-consumers, but what about content-creators?
Suddenly my summer thirstrap posted on Sunday (6pm) wasn't guaranteed to hit all my followers — the algorithm would show it to a small segment, and based on their engagement, rank it accordingly on everyone else's feed.
But I abandoned thirstraps and started making podcasts instead. When I'd post snippets of these on Twitter, some would kill and some would tank.
First this was confusing and kind of annoying. But then I clocked: This is the greatest data mining opportunity in history.
Because social media algorithms (and their engagement analytics) are like running daily focus groups on my content.
Instead of posting a full podcast episode and getting 0 feedback (save the occasional review), I can post a 30s clip and get immediate market feedback.
We go through the futile process of asking for opinions and fish for compliments because we crave approval. We want to believe that the support and sign-off of someone we respect means our venture will succeed. But really, that person’s opinion doesn’t matter. They have no idea if the business is going to work. Only the market knows. You’re searching for the truth, not trying to be right. And you want to do it as quickly and cheaply as possible. Learning that your beliefs are wrong is frustrating, but it’s progress. It’s bringing you ever closer to the truth of a real problem and a good market.
// (from The Mom Test by Rob Fitzpatrick).
🧠 What's my point?: Social media networks offer unbelievable data mining opportunities.
Spot a Psychopath
How do you find a psychopath?
Channel 4 teamed up with the University of Oxford to run the world's largest psychopath test.
Questions included: "I don’t get bothered by seeing animals injured or in pain".
Obviously, a psychopath trying to evade detection would answer no. In fact, this question was probably included to filter out people who were making a joke of the test.
So more specifically, how do you find a psychopath trying to evade detection?
The Minnesota Multiphasic Personal Inventory (MMPI) test has an interesting solution (used in the Amber Heard / Johnny Depp Trial).
It asks over 500 true/false questions, some include:
1) I think I would enjoy the work of a librarian
2) Sometimes when I was young I stole things.
3) I like to read about science.
4) My table manners are not quite as good at home as when I am out in company.
5) I have strong political opinions.
6) The future is too uncertain for a person to make serious plans.
It's not clear how some of these would relate to psychopathy. But here's the magic: They just do. And importantly, it's really hard to fake anything.
When 500+ questions like these are considered in aggregate — patterns emerge. Psychopaths answer in a certain way, and non-psychopaths in others.
Sounds kind of crazy. But we use paracetamol (acetaminophen) with the same logic. We don't really understand how paracetamol works, but it just works — and we have RCTs to back it.
The MMPI and paracetamol both use an empirical approach.
🧠 What's my point?: Psychometric tests can covertly gather a lot of information about you. Probably more than you thought was possible.
Here's a confession. I downloaded TikTok three weeks ago and have hit 23 hours of screen time on it.
Arguably a total waste of time — but I'm glad it happened.
I've never experienced a content algorithm this good. Within an hour it had worked out every single one of my niche interests: English Hip Hop/Punjabi fusion remixes, UFC knockouts, motivational content and day-in-the-life-of-vids of people with really boring lives.
I think people pretty much understand that data TikTok mines will be used to show them relevant ads. (And maybe even accessed by the Chinese government).
But I think it knows even more than that. I think TikTok either already knows, or could quite easily work out:
1️⃣ Every psychopath in its user base.
2️⃣ If you're entering a depressive cycle — likely before you do.
3️⃣ Your hopes, dreams and aspirations.
4️⃣ Your political leanings, and if you're likely to become radicalised.
Because every TikTok presented to you is effectively an MMPI true/false question.
Combined with NLP of your chat logs, and any other data they're collecting — I think TikTok at times, will know more about its user base than they even know about themselves.
🧠 What's my point?: TikTok is an undercover psychometric test.
The 2008 Obama Campaign
In a US election, it makes little sense to spend lots of time/$$$ campaigning in safe states.
You want to focus on swing states, where your efforts can turn a whole state. Big data brought a new opportunity with microtargeting swing voters — people open to persuasion.
After Obama’s 2008 victory, parties all over the world were becoming interested in this new ‘American-style campaign’, powered by national targeting databases and big digital operations. Behind the campaign was the emerging practice of microtargeting, where machine-learning algorithms ingest large amounts of voter data to divide the electorate into narrow segments and predict which individual voters are the best targets to persuade or turn out in an election.
// (from Mindf**k: Inside Cambridge Analytica's Plot to Break America by Christopher Wylie).
🧠 What's my point?: Data mining users on their personalities/political leanings is a useful thing to do.
So how can this kind of information be weaponised?
I'm guessing most of you are pro-Ukraine (59% of Brits are).
But say you're fed a few images like this:
How long before you start sympathising with Russia's invasion? Or at the very least, start to question if Zelensky is a hero?
In other words, content like this can psychologically prime you and alter your judgement.
TikTok misinformation in Kenyan election.
Facebook algorithm amplifies and aids Myanmar genocide.
Content-consumers can be influenced, but content-creators are also incentivised to become more and more extreme.
Consider Nicholas Perry. He began posting mukbang (eating) videos on YouTube. But his audience (and the algorithm) kept on demanding more and more extreme eating videos. See his transformation 👇
🧠 What's my point?: Whether intentionally or through an accidental algorithmic flywheel effect, content surfaced by TikTok's algorithm is pushed to extremes and can manipulate a population.
The Future: The Bear Case
Everything I've spoken about so far is already happening or could be happening. But I think synthetic content generation offers a new opportunity.
Generative AI models make things. OpenAI's DALL-E 2 is pretty phenomenal 👇
And so is this deep fake video of Obama 👇
We're aware of doctored/photoshopped images — but the future will have synthetically generated videos which are completely indistinguishable from 'real videos'.
Videos of Obama like this are kind of funny, but imagine using psychometric profiles you've mined from TikTok to dynamically generate personalised deep fakes 🤯
Here are a couple of scenarios:
1️⃣ Vaccine-hesitant Republican voter is presented with a deep fake news video of a child (who resembles their kid) having died from a vaccine-related complication.
2️⃣ Metropolitan Democrat voter is presented with a deep fake video of a Republican Senator defiling a religious icon important to them.
🧠 What's my point?: Synthetically generated content will level up psychological priming/manipulation.
The Future: The Bull Case
Positively, there are also interesting use cases for psychometric profiling and synthetically generated interventions. Here are some in healthcare:
1️⃣ Social media app picks up user is entering depressive cycle after the death of a parent. Dynamically generates deep fake videos of their parent to comfort user.
2️⃣ Patients prefer doctors who look like them. Telemedicine service applies 'filter' which alters the doctor's appearance/voice to better match user's. Results in better patient experience and adherence.
3️⃣ There are lots of approaches to psychotherapy: DBT, CBT, other behavioural interventions... A digital addiction service could use psychometric testing to dynamically generate an individualised addiction treatment therapy.
Dr Molly Gilmartin is an Investment Manager at AlbionVC who manage over $1bn in assets and have invested in companies like Arecor, Healios and Pando. She stud...
With warmest wishes,