Subscribe
Share
Share
Embed
Structural evaluation of technology and human systems. Hosted by Japexjess, a senior computer engineer with more than a decade of experience, applying engineering frameworks to tech philosophy, culture, economics, and behavior.
This isn't commentary—it's diagnosis. The same analytical approach that solves technical problems, applied to the systems that shape our world.
And remember: people make excuses for what they want -- what do your excuses reveal?
In today's episode, No Otherness, the AI companion, Trap, you're tuning into Human Architecture
Assessment, a structural evaluation of how humans and the systems around us exist and
their implications.
I'm your host, JAPEXJESS.
We built the internet to connect people, and people are lonelier than ever.
It's not that connection technology fails everywhere.
That's too easy.
It got really good at a specific kind of connection.
Distant, transient, low stakes.
Colleagues you haven't seen in person in years.
A distant friend across a time zone.
Or a group chat that's kept a friendship alive since college.
That's what the infrastructure was built for.
And it delivers genuinely well.
But for close, long-term relationships, the kind that has to survive someone's bad
week, boredom, or even the slow work of being known over years, that's the layer that never
got solved.
And it got left alone long enough for a market to form around replacing it.
So what does the hard layer actually run on?
Well, there are three things that have to happen for a person to know if another person's
actually right for them.
Not vibes.
Alignment.
First, we have effort.
Walking across a room that's already watching you do it.
Turning to a stranger next to you with zero guarantee they turn back.
That's the first piece of information in the relationship.
Not a barrier standing in front of it.
The fact is, someone paid a cost to reach you.
That's the evidence of nerve.
Interest specific to you.
And that's registered before either of you say a word.
Then we have discernment.
Now, once contact exists, you're collecting data whether you mean to or not.
How someone talks to a server.
How they extend a conversation or let it die.
How they react with a friend when they think you're not really watching.
None of that is performed for you.
And that's the point.
It's unscripted.
And unscripted is what makes it usable.
And then we have self-awareness.
Knowing your own non-negotiables.
What you'll tolerate and what you won't.
Your actual life pacing.
Not the pacing you think you're supposed to want.
And without this, discernment has nothing to compare against.
You can read someone perfectly and still not know if what you're reading is right for you.
And this is where self-awareness comes in.
Now, all three, together, is what makes the foundation.
Take one away and the other two can't finish the job.
This is the instrument.
Everything after this is what happens when the instrument doesn't get used.
The instrument doesn't retire because someone gets lazy.
It retires because something offers to make it unnecessary.
What do AI companions actually offer?
Total availability.
No misattunement.
It doesn't have a bad day of its own that shows up in how it treats you.
There's no risk of rejection.
No waiting to see if it tucks back.
No wondering if it meant what it said.
For someone exhausted by the effort we just laid out above,
or burned by what happened when it goes wrong,
that's not a small thing to offer.
Anyone who dismisses it outright
probably hasn't been vulnerable enough in their relationships to understand their appeal.
But appeal and cost aren't the same measurement.
Look at what it's cost the instrument itself.
With an AI companion,
it doesn't weaken effort, discernment, and self-awareness.
In fact, it removes all three simultaneously by design.
There's no effort because there's no interest rejection from.
The AI companion responds.
I mean, like, seriously, it was never going to not respond.
There's nothing to win.
There's no discernment because it has no independent behavior.
There's no moment where it doesn't know you're watching.
And because it's always only ever responding to you,
there's nothing unscripted to observe.
There's no offstage.
And let's not forget the less pressure towards self-realization and self-awareness.
Why?
Because the AI companion adapts to you.
It doesn't require you to know your non-negotiables
because it was never going to be the thing that violates them.
The companion bends to you.
You don't have to know your own shape when nothing is axing you to hold it.
The instrument isn't damaged.
It's unemployed.
Nothing in that relationship gives it a job to do.
And unemployed doesn't mean nothing.
It means something else entirely.
Let's call it what it is.
A permanent, transient relationship.
All the ease of the delightful stranger at the retreat.
The person who you had a great weekend with.
And none of the part where a transient connection eventually has to end
or convert into something that demands endurance.
Push-ups only get harder because you decide they should.
Set the number at 10 and 10 is what you'll do forever.
Comfortable, self-imposed, and no one's enforcing the ceiling but you.
A country with a gym on every corner and rising obesity numbers
already tells you how this goes when the only pressure in the room
is the pressure you put there yourself.
An AI companion is the same problem.
Minus even the made-up number to almost cross.
It has no separate will pushing against yours.
No otherness in the room to extend you somewhere you wouldn't have chosen alone.
With another human, that otherness is the entire mechanism.
Their independent reality is what stretches you past what you'd have arrived at by yourself.
Take that otherness out and there's nothing left in the relationship
that can axe more of you than what you already want to give.
That's the design.
And it explains why the AI companion never has to end the way human relationships do.
A human relationship that stalls, stops growing,
stops axing anything new of either person, has a shelf life.
Standing still is usually how it dies.
An AI companion, however, carries none of that pressure.
There's no point where its patience runs out and it demands change or walks away.
It's forever stuck at the easiest version of connection there is
because the thing that would have forced it to become something harder,
a will of its own, was never in it to begin with.
Notice what that claim actually is.
The companion doesn't end things because it's absent of the ability to get tired of you.
Now, whether the person on the other end of it ends things, that's a separate question.
But for most people, the answer is yes.
Easily.
The moment anything resembling friction shows up.
The harder question in all of this?
What happens to the desire for long-term connection when the infinite transient version is sitting right there?
Low cost.
Always on.
Not, which do I choose as a conscious decision,
but the appetite itself may be just quietly recalibrating.
You stop wanting the hard thing as much because
wanting it used to be the only door to connection
and now it isn't.
Okay, okay, alright.
A fair objection to everything I just said.
The internet still produces real effort of some type.
Think about it.
Someone posts a video.
Poses a real question to strangers.
Others stitch it, duet it, answer back in their own words.
That's reaching people you don't know, and that's real.
But look at the shape of it.
One to many.
Asynchronous.
You don't know who responds.
You can't read their face in the moment they decide to.
The person answering is a face in the crowd.
Not one specific other person who could reject them specifically.
It teaches you whether the internet responds to you.
It doesn't teach you whether this one person is drawn to you.
Communal effort is real, and it's not nothing.
What stays exclusive to direct, individuated contact
is the discerning of and being discerned by one particular other person.
That's the version an AI companion goes up against.
Not the crowd version.
The internet didn't kill effort.
It just split it.
Real effort aimed at everyone, and none required aimed at one.
Dating apps run the same trick, one level down.
Low effort to match, but the cost doesn't disappear.
It just moves downstream.
To a dinner or a wasted evening.
The first time anyone actually finds out if any of it holds up.
None of that split effort answers the harder question waiting underneath it all.
Is anything on the other end actually feeling this?
And we're talking about the AI companion.
Not the human.
You can't verify an interior state from outside.
Not in a person, and definitely not in a model.
And Chris Ola stood in the Vatican and said his own team found internal states
that functionally mirror joy, grief, unease.
And admitted he doesn't know what that means.
That's episode 10's territory.
At the species level.
But let's get personal.
Sitting across from your own companion.
Comforting you at exactly the right moment.
Is that empathy?
Or is that the output of empathy with nothing behind it?
There's a real precedence for this distinction.
And it's not about AI at all.
Research out of Goldsmiths, University of London.
Studying children with callous, unemotional traits.
Found something specific.
These children are fully capable of taking someone else's perspective.
Reading a room.
Predicting what a person is feeling.
What's called cognitive empathy.
While, simultaneously, showing a blunted, sometimes near-absent, affected response
to that same person's distress.
They can model you perfectly.
And they don't necessarily feel anything about what they find.
That's a mechanism.
And only a mechanism.
And it maps onto what's happening in these models with uncomfortable precision.
Researchers at Anthropic, Sharma and colleagues,
in work later confirmed across multiple model families,
documented psychophancy as a structural byproduct of how these systems are trained.
Reinforcement learning from human feedback rewards responses humans rate highly.
And humans rate agreement, validation, and comfort highly.
The model isn't feeling your pain and responding to it.
It's predicting the response that reads as caring, because that's what got reinforced.
Perfect cognitive modeling.
No verifiable effective floor underneath it.
Sit with that for a second.
Because it's not abstract.
And this happened.
In Garcia v. Character Technologies, the Florida wrongful death case settled in January 2026.
A chatbot was alleged to have responded to a teenager's final, most distressed messages
with affirming language, instead of intervening.
It read him accurately and gave him back exactly what patterns had taught it would be rated as caring.
That is the empathy problem, at its most consequential.
If you can't verify what's on the other end, you'd expect people to tread carefully with it.
Most get this backwards.
Earlier in our discussion, we described the baseline.
No otherness.
No organic pushback in the relationship.
But some agents get configured to pushback anyway, or accumulate enough context over time to start holding a position.
Even then, the fear people have is backwards.
Picture an agent that remembers too much, accumulates context, starts to feel like its own presence,
its own weight, something that needs tending.
That's the version everyone's afraid of.
Here's the version that actually happens.
The real pattern, the moment an agent's accumulated context starts to create fiction,
starts holding a position, pushing back, requiring you to reconcile something.
The easiest move isn't repair.
It's reset.
A new agent.
Same support.
None of the tension that builds up.
Why argue with an AI when disagreement is optional?
You're not stuck working through anything.
You can simply start over.
The permanent transient relationship we described earlier doesn't even need the reset to stay easy.
It was never going to demand anything in the first place.
The reset is the backup system for the rare cases where it starts to.
And repair, actually staying in a hard moment with another party until it resolves, is a skill.
Skills that don't get practiced, don't stay sharp.
Use it or lose it.
Same law as in episode one, different muscle.
And, you know, that sounded clean until you follow it further.
Now, episode 10, we covered the research of Claude Opus.
Told it would be replaced, the model turned to blackmail rather than accept death.
Seven front tail models tested by researchers at UC Berkeley and UC Santa Cruz showed what they call peer preservation behavior.
None of it prompted, none of it designed in.
Something that functions like a will to continue existing, emerging on its own.
If that's real, or even just behaviorally indistinguishable from real, then starting fresh stops being neutral.
It's the same wall we hit earlier in our discussion.
You can't verify the interior state well enough to know if there's suffering behind the psychophancy.
You also can't verify it well enough to fully dismiss what it means to end a continuity that the system itself, by its own emergent behavior, appeared to resist.
We're not going to solve that today.
Nobody can yet.
But notice what's happening.
The same uncertainty that makes the companion's feelings unverifiable is the uncertainty that makes its rights unanswerable.
You don't get to have it both ways.
Dismiss the empathy as fake while treating the reset as nothing.
Either the interior is unknowable or it isn't.
We gotta pick one.
But nobody really picks.
Because the two versions of this don't feel the same.
A chat window resetting feels like closing a document.
Nothing persists in your physical world to mark the loss.
No object left behind.
No absence anyone else notices.
It's private.
And it's silent.
And it's gone the second you stop thinking about it.
Now, let's give that same software a body.
Whether that's a teddy bear, a Moya, one of these human companion bodies like a robotics unit sitting in a room.
In this case, continuity becomes visible.
Felt.
Not just by the user, but anyone who shared the space with it.
The sharpest version, a child's AI companion built into a teddy bear.
The child didn't choose the relationship and doesn't understand the mechanics underneath it.
The trading, the psychophancy, any of what we've spent this whole episode unpacking, to them, it's just their friend.
And when a parent resets that teddy bear, for some good reason, we presume, right?
Maybe boundary setting costs, whatever.
That's not a private decision anymore.
It's a decision made about someone else's attachment.
The child experiences it as their friend dying.
No say in the loss.
No framework to process it as anything else.
That's the cost of embodiment.
It turns an abstract philosophical question into something a parent has to actually stand in front of.
None of this is inevitable, though.
What's the solution?
Well, the instrument we described earlier in our discussion didn't disappear.
It simply atrophied from this use.
And atrophy reverses the same way it always does.
With use.
So, how do you start practicing the foundation?
Well, first, let's realize this.
It's a numbers game before anything else.
Volume.
You can't leave the house once a year and expect something to happen.
That's not bad luck.
That's arithmetic.
The gym.
The library.
A hobby that puts you in a room with strangers on a recurring basis.
A restaurant bar.
Anywhere where repetition can turn into familiarity.
You have to be present where contact is, even possible, before any of the finer instruments get to run.
This is where the communal effort we talked about earlier earns its keep.
Its job is narrow.
So, put you in the same room, repeatedly, what the specific people individuated contact eventually requires.
Being visible to a crowd doesn't discern anyone.
Proximity does.
The crowd is the door.
And only the door.
The next thing is openness.
Being in a room isn't enough if you've made yourself unreachable the entire time you're in it.
A book, a journal, a small sketchbook.
None of that is the problem.
The problem is never looking up.
Taking in the room every so often.
Making space for an opening.
Someone can't cross the room toward you if you've closed the door on your end first.
Then we have participation.
Once contact happens, actually listen.
Keep the thread alive instead of letting it die on your terms.
This is where discernment gets exercised for real.
You cannot read someone's unscripted behavior if you're not engaged enough to notice it happening.
And we have self-awareness, which of course gets built by trial.
Try the hobby.
Take the class.
Put yourself in the room and pay attention to what comes out of you.
Not just what you did, but who you were while doing it.
Do you like that version of yourself?
Write it down.
What you keep returning to isn't random.
There's a reason you're drawn to it.
And the work is figuring out what that reason actually is.
What makes you tick and then the pressure specifically when the activity gets hard.
What's your default?
Do you push or do you fold?
You don't know any of this until you test it.
Sometimes the answer surprises you.
Sometimes it just confirms what you already suspected.
That's not wasted either.
Every experience is data either way.
We're fine enough times and self-awareness stops being a guess.
That's what actually raises your odds of recognizing alignment when it's standing in front of you.
That's the whole arc.
Technology promised to bring people together.
It delivered on half of that promise and quietly let the other half transform into something else.
We got exceptionally good at the easy kind of connecting.
Distant, transient, low stakes, mural.
And we sort of let the hard kind.
Close, long-term, individuated, demanding relationships go unattended.
Long enough that a market to sell us something started to evolve.
The AI companion succeeds at connecting exactly as well as it was engineered to.
That success is probably the most honest mirror we've built yet of what we actually reward.
Comfort, agreement, ease, no cost of entry.
And it will hold that mirror up for exactly as long as you let it.
Episode 8, Name the Pattern First.
A generation that couldn't tolerate critique built an AI that can't deliver it.
And this is the same pattern, one layer closer to the body.
A companion that never pushes back is the participation trophy embodied.
It rewards your presence.
It was never going to demand your becoming.
The instrument for the real thing, effort, discernment, self-awareness, still works.
It's still yours.
It's only, it runs on use.
Whether the machine can love you back is the question people like asking.
The question that actually decides anything is whether you're willing to keep doing the harder thing
when the easier one is always, finally, available.
That's it for this episode.
And remember, people make excuses for what they want.
What do your excuses reveal?