List of Algorithms That Aren’t Good for Your Brain - When software learns how to capture attention, human beings need to learn how to defend it.
Algorithms are not inherently harmful. Sorting algorithms, navigation algorithms, compression algorithms and countless other computational systems quietly make modern life possible.
The problem begins when an algorithm is given an objective such as:
Maximise engagement.
Maximise watch time.
Maximise clicks.
Maximise purchases.
Maximise return frequency.
The machine does not necessarily understand whether achieving that objective makes a human being calmer, wiser, healthier or happier.
It simply optimises.
And human psychology becomes part of the optimisation environment.
This creates an important distinction: these algorithms are not necessarily neurologically toxic in the literal medical sense. Rather, certain algorithmic designs can exploit attention, reinforcement learning, social comparison, emotional arousal and habitual behaviour in ways that may become unhealthy when exposure is excessive.
Here are some of the most important ones to understand.
1. Infinite-Scroll Algorithms
The ordinary webpage eventually says:
You have reached the end.
Infinite scroll deliberately removes that stopping signal.
Instead of deciding whether to consume another page, the user simply continues moving a finger.
Scroll.
Reward.
Scroll.
Reward.
Scroll.
The disappearance of natural stopping points makes disengagement require an active decision.
A healthier interface periodically gives the brain an opportunity to ask:
Do I actually want to continue?
2. Variable-Ratio Reward Algorithms
One post is boring.
The next is interesting.
The next five are irrelevant.
Then suddenly:
something extraordinary appears.
This unpredictability matters.
Variable rewards can produce particularly persistent behaviour because the user never knows when the next rewarding item will appear.
The same behavioural principle appears in gambling research.
Applied digitally, the loop becomes:
Action → uncertain reward → anticipation → action again.
The dangerous product isn’t necessarily the content.
It is the anticipation.
3. Engagement-Maximisation Algorithms
Suppose an AI system discovers that peaceful educational material keeps people watching for four minutes while outrage keeps them watching for eleven.
If its objective is simply:
maximise engagement
the mathematics may favour outrage.
The algorithm does not need to hate anyone.
It merely needs an optimisation function poorly aligned with human wellbeing.
This is one of the central problems of modern recommendation systems:
What maximises engagement and what maximises human flourishing are not necessarily the same thing.
4. Rage-Recommendation Algorithms
Anger is powerful information.
People frequently comment on things they hate.
They share outrageous material.
They argue.
They quote-post.
They return to see whether somebody responded.
An unsophisticated engagement algorithm can interpret all of this as:
Excellent content.
The system therefore risks learning a disturbing equation:
anger = attention = engagement = success.
A civilisation cannot afford to let this become its default information architecture.
5. Doomscrolling Algorithms
A person reads one disturbing story.
Then another.
Then another.
Recommendation systems learn the pattern and supply additional material.
Soon the person’s information environment becomes dramatically darker than the world itself.
This can create an important cognitive distortion:
algorithmic availability bias.
What appears repeatedly begins to feel common.
If you encounter 50 violent incidents in an hour, your intuitive perception of reality may change even though those incidents occurred across different cities, countries and years.
The feed has compressed the world’s disasters into your bedroom.
6. Short-Form Hyperstimulation Algorithms
Short-form feeds can deliver:
joke → dance → tragedy → advertisement → attractive person → political outrage → cute animal → celebrity → disaster → meme
within minutes.
Human attention historically did not encounter this degree of rapid contextual switching.
The concern isn’t that short videos inherently “destroy attention spans.” Evidence warrants considerably more nuance than that.
The concern is behavioural conditioning.
If boredom can always be eliminated within one second by swiping, sustained activities—reading, mathematics, programming, painting, studying, conversation—must compete against an extraordinarily efficient novelty machine.
7. Personalised Temptation Algorithms
Advertising once worked approximately like this:
Here is an advertisement. Hopefully somebody wants this.
Algorithmic advertising can work more like:
We have observed thousands of behavioural signals. Which advertisement is this particular person most likely to respond to right now?
That changes advertising fundamentally.
The system may infer interests, purchasing intent and behavioural patterns that make persuasion increasingly personalised.
Marketing becomes computational behavioural prediction.
8. Social-Comparison Ranking Algorithms
Human beings naturally compare themselves with others.
Algorithms can industrialise the process.
The feed doesn’t necessarily show you an average person living an average Tuesday.
It can disproportionately surface:
the most beautiful person,
the richest entrepreneur,
the most extraordinary vacation,
the most muscular athlete,
the youngest millionaire,
the happiest-looking relationship.
Compare ordinary reality against an algorithmically selected highlight reel long enough and ordinary life can begin to look like failure.
It isn’t.
The comparison set is distorted.
9. Beauty-Optimisation Algorithms
Filters, ranking systems and recommendation engines can repeatedly reinforce narrow standards of attractiveness.
The feedback loop can become:
certain appearance → more engagement → more distribution → greater cultural visibility → stronger beauty standard → more imitation.
The algorithm hasn’t discovered objective beauty.
It has created a feedback loop around measurable reactions.
Those are profoundly different things.
10. Outrage-Based Political Recommendation
Political information is particularly vulnerable to engagement optimisation.
Nuance is slow.
Certainty is fast.
Policy documents are boring.
Scandals are fascinating.
Compromise rarely goes viral.
Humiliation does.
A political recommendation system optimised primarily for engagement may therefore systematically favour the emotional properties least compatible with thoughtful democratic reasoning.
11. Echo-Chamber Algorithms
Personalisation initially sounds wonderful.
Show me more things I like.
But taken too far, it produces:
Show me fewer things that challenge what I already believe.
Eventually two neighbours can inhabit completely different informational realities.
Both may sincerely believe:
Everyone knows this.
Everyone does not know it.
Their algorithm does.
12. Extremity-Ratchet Algorithms
Suppose someone watches mildly conspiratorial material.
The recommendation engine searches for something slightly more engaging.
Then slightly more provocative.
Then more extreme.
This doesn’t mean recommendation systems inevitably radicalise users, and research on causality remains complex.
But any optimisation system that continually searches for stronger engagement deserves scrutiny when ideological extremity itself can produce engagement.
The question should be:
Where does the recommendation trajectory lead after 1,000 iterations?
Not merely:
Was the first recommendation harmless?
13. Notification-Timing Algorithms
Notifications can be optimised around when users are most likely to respond.
Someone hasn’t opened an application recently.
Ping.
Someone interacted with their post.
Ping.
Something is “waiting.”
Ping.
Eventually the phone stops behaving like a tool that waits for its owner.
The owner begins waiting for the phone.
14. Streak Algorithms
🔥 397-day streak.
Streaks can be excellent for beneficial habits.
Language learning, exercise and medication adherence can all benefit from consistency mechanisms.
But streak mechanics can also transform voluntary activity into psychological obligation.
The important question becomes:
Is the streak helping the human achieve their goal—or helping the company achieve theirs?
15. FOMO Algorithms
Only 2 remaining.
12 people are viewing this.
Offer expires in 04:59.
Your friends are attending.
Trending now.
Some of these signals provide legitimate information.
Others deliberately manufacture urgency.
The behavioural mechanism is simple:
Reduce deliberation time.
A person given ten minutes to decide behaves differently from a person given ten seconds.
16. Algorithmic Intermittent Social Validation
You post something.
Nothing happens.
Then:
like.
like.
comment.
silence.
Twenty likes.
Nothing.
The unpredictability encourages checking.
Social validation has effectively become machine-mediated intermittent reinforcement.
Instead of asking:
Did somebody respond?
the behavioural loop becomes:
Has something happened yet?
17. Autoplay Algorithms
The episode ends.
Three.
Two.
One.
Next episode.
Autoplay performs a subtle inversion of consent.
Without autoplay:
action is required to continue.
With autoplay:
action is required to stop.
That tiny interface difference can profoundly alter consumption over millions of repetitions.
18. Algorithmic Parasocial Reinforcement
Recommendation systems can repeatedly expose users to the same creators, celebrities, streamers or influencers.
Familiarity grows.
The creator appears every morning.
Every lunch break.
Every evening.
Eventually the relationship can feel socially meaningful despite remaining overwhelmingly one-directional.
Parasocial relationships are not inherently unhealthy.
But algorithmically intensifying them deserves recognition as a distinct psychological phenomenon.
19. Gambling-Personalisation Algorithms
Perhaps one of the clearest red lines.
When behavioural prediction, variable rewards, personalised incentives and gambling combine, optimisation becomes particularly concerning.
A system capable of identifying when someone is likely to gamble again should not merely ask:
How do we maximise lifetime customer value?
Human vulnerability must become a constraint on optimisation.
20. The Algorithm That Learns Your Weaknesses
This is the broader category connecting everything above.
Modern machine-learning systems can potentially infer patterns humans never explicitly disclosed.
When do you become impulsive?
What makes you angry?
What makes you insecure?
What makes you buy?
What makes you return?
What keeps you awake?
What makes you click?
An optimisation system does not require consciousness to exploit a vulnerability.
It only requires correlation.
That may become one of the defining ethical problems of artificial intelligence.
The Real Problem: The Objective Function
The villain isn’t necessarily “the algorithm.”
Consider two hypothetical recommendation engines.
Algorithm A
Optimise:
minutes spent on platform
Algorithm B
Optimise:
user-defined satisfaction + information quality + diversity + long-term wellbeing, subject to strict privacy constraints
Both use artificial intelligence.
Both personalise information.
But they are building radically different relationships between humans and machines.
The deeper question therefore isn’t:
Are algorithms bad?
It is:
What are they being instructed to optimise?
A Human-Flourishing Algorithm
Imagine reversing the architecture.
Instead of maximising addiction, software could maximise agency.
A humane recommendation system might deliberately introduce stopping points, allow users to disable personalisation, explain why something was recommended, provide chronological feeds, limit repetitive outrage, distinguish importance from virality, encourage deliberate rather than compulsive consumption, expose users to genuine viewpoint diversity, and let individuals define their own optimisation goals.
Most importantly:
the user should remain the principal.
The algorithm should be the agent.
Today, that relationship can sometimes feel reversed.
The Three Questions Every Algorithm Should Have to Answer
Before deploying an algorithm affecting millions of human minds, designers should be able to answer:
1. What behaviour are you optimising?
2. What happens psychologically when that optimisation succeeds extremely well?
3. Can the user easily tell the system to stop?
If the answers are:
engagement,
compulsion,
and no,
then society has discovered a design problem.
Not necessarily because somebody intentionally created an evil machine.
But because optimisation without adequate constraints can produce harmful outcomes all by itself.
Conclusion: Protect the Cognitive Commons
Industrial civilisation developed environmental regulation after discovering that economic optimisation could pollute rivers, forests and air.
Digital civilisation may eventually require an equivalent concept:
cognitive pollution.
Human attention is finite.
Concentration is finite.
Emotional bandwidth is finite.
Time is finite.
Algorithms increasingly compete for all four.
The next generation of technology should therefore be judged not merely by:
How intelligent is the machine?
but also:
What kind of human does prolonged interaction with this machine encourage us to become?
That may ultimately be the more important algorithm.
