The scary version isn't the perfect AI.
The scary version is good enough AI in a
low attention environment. Someone
should clone my voice today. Not
someday. Not in some speculative future
where every creator has a fully
synthetic double doing brand deals in
the metaverse. I mean today. And if you
listen to enough of me talking through
agent memory and retrieval systems and
AI infrastructure and enterprise data
and a thousand little ways, AI stops
being a demo and starts becoming
infrastructure. There is probably enough
clean audio out there to make a pretty
convincing synthetic mate. And I know
because people have tried. So, in this
video, I want to do three things. First,
I'm going to show you a short clip of my
voice clone, clearly labeled, no tricks.
Second, I want to explain why voice
cloning is already much farther along
than most people realize, while full
human presence cloning is still really
weird in very specific ways. And third,
I want to zoom back out because the real
issue here is not was AI used. That
question is already really a crude
question that isn't useful in most
cases. The real issue is trust. What was
human? What was synthetic? Who made that
judgment? And who is accountable for the
final thing that you're watching? Before
we go any further, quick disclosure. The
next voice you hear is not me speaking
live. It is a synthetic voice clone. It
is included as a demonstration. It will
be labeled on screen. And I approve of
this use. This is not Nate speaking
live. This is a synthetic version of
Nate's voice generated from prior
recordings and included here to
demonstrate how quickly voice level
authenticity is becoming easier to fake.
The argument in this video is still
Nates. The judgment is still Nates. This
clip is synthetic and it is being
disclosed. Okay, that was impressive and
frankly a little creepy. And that
tension is basically this whole video
because voice cloning is getting very
very good. If you have clean source
audio and enough of it in a consistent
speaker, the tools are already good
enough to create something that will
pass in a lot of normal listening
environments, especially if someone is
half listening while checking email and
folding laundry and scrolling or doing
whatever else all of you are doing while
YouTube is playing. But full human
cloning is different. A voice can sound
like someone. A face can look like
someone. But the sense of presence is a
whole lot harder. You can see this
across a lot of AI video right now. The
lips are close but not quite right. The
blinking is close but not quite human.
The hands move but they don't have
weight. The expressions are there but
the micro expressions are missing. So
everything is like 90% right. And the
last 10% makes the whole thing feel
wrong. And this is the part I think
people miss. The scary version isn't the
perfect AI. The scary version is good
enough AI in a low attention environment
because if you sit there and if you
study the clip, you will catch the weird
mouth movement or the strange timing or
the eyes that don't quite behave like
eyes. But YouTube is not a forensic lab.
Tik Tok is not a forensic lab. LinkedIn
definitely is not a forensic lab.
Although people are doing their best,
most media is consumed casually. People
are half watching. They're listening in
the background. They're seeing a clip
out of context. They're catching a few
seconds and then moving on. So the
threshold is not can this fool an expert
watching carefully. The threshold is can
this create enough ambiguity that normal
people stop knowing what relationship
they have to the person on screen. The
uncanny valley used to be visual. Does
this face look real? Do the eyes look
right? Does the mouth move properly? Now
the uncanny valley is more structural.
It's more institutional. It gets into
trust territory. It's relational. Do I
believe there's a person behind this? Do
I believe there was a process? Do I
believe someone made a judgment? Do I
believe someone is accountable if this
is wrong or manipulative or fraudulent?
And this is where I think creators and
companies and frankly everyone building
with AI needs a better framework. So
quick definition. When people ask, "Was
this made with AI?" They're usually
asking at least five different questions
at once. I'll give them to you. Number
one, was the voice synthetic? Was the
face synthetic? That's number two. Was
the script synthetic? That's number
three. Number four, was the idea
synthetic? And number five, did a human
actually approve and stand behind the
final output? All of those tend to get
mashed together. Those are not the same
question. A creator using AI to clean up
audio is not the same as a creator
secretly replacing themselves with a
clone. We all know that. A company using
AI to draft a first version of a
training video is not the same as
cloning an employes's voice without
consent. An analyst using AI to research
a topic is not the same as publishing an
AI generated claim that no one checked.
So the question is not AI or no AI.
That's a bad primitive. It's too blunt.
It's a light switch. The world is not
binary. The better question is where in
the stack did AI operate and where did
human judgment take over to guarantee
the final product? I think of this as a
creator trust stack. Layer one is
disclosure. What was synthetic? Was the
voice cloned? Was the face generated?
Was the script just drafted with AI? Was
the edit assembled with AI? Well, we can
say that clearly. Layer two is
provenence. Where did the source
material come from? Was the voice clone
trained on recordings that the person
consented to? Was the avatar made from
authorized footage? Was the data scraped
and licensed and owned or just sort of
magically available in the way everyone
says when they don't want to answer the
question? Layer three is control. Who
had the ability to approve or reject or
change the output? Did the person being
cloned have control over the use of
their likeness? Layer four is judgment.
Who actually made the argument? Who
decided what this video meant? Who
decided what claims were worth making?
And layer five is accountability. If the
video is wrong or manipulative or
harmful, who owns that? This is the part
a lot of us want to skip. And I think
it's the part that matters most because
if you are building media with AI, the
audience does not just need to know that
a model was involved. That's like bare
minimum. They need to know whether a
responsible person was involved who's
accountable to the results. Now, if
you're looking for a silver bullet to
answer where the rule is never use AI,
just stop looking. This is not how this
works in 2026. We are going to use these
tools. Creators are going to use these
tools. I use these tools. Companies are
using these tools. Educators, analysts,
marketers, product teams, support teams,
everybody. The question is whether they
use them in a way that makes the
audience smarter or whether they use
them in a way that makes the audience
feel tricked. Human weirdness is going
to start looking like machine weirdness.
Someone mispronounces a word and people
say that's AI. Someone wears the same
shirt in four videos because they batch
recorded and people say that's AI.
Someone has an awkward pause, a weird
edit, a tired delivery, a strange facial
expression, and suddenly the comment
section becomes some kind of touring
test with bad lighting. But humans are
inconsistent. I'm inconsistent. Humans
get tired. I get tired. I repeat myself.
Humans say something a little wrong and
then keep going. Humans have bad hair
days. Humans blink weirdly. Sometimes
humans do not always perform humanity in
a clean, legible, perfectly edited way.
Has it occurred to any of you that I
wear a beanie cuz I don't want to have
to brush my hair on camera. And this is
going to become part of the confusion.
And no, for all of you watching, I will
not be doing a get ready with me. AI
will get accused of being human. Humans
will get accused of being AI. And
somewhere in the middle, the actual
issue will be whether there's a
trustworthy relationship between the
person making the thing and the
audience. So what should we creators do?
On the one hand, we need to disclose
synthetic media. Clearly, not in a
paragraph that is buried in the
description, not in a vague AI assisted
footnote that could mean anything. Be
specific. Two, don't clone voices or
faces without consent. This should be
obvious, but apparently we're living
through a period when obvious things
need to be stated clearly. Three,
preserve human judgment. Use AI for
leverage, not for deception. Draft
faster, edit faster, prototype faster,
but don't outsource responsibility for
what you're saying. Four, and I'm really
passionate about this, make the audience
more literate. If you use a clone, show
the clone and label it and explain what
it can and can't do and help people
understand the difference between
synthetic media and synthetic
accountability. And five, if you're a
company, create the policy before the
scandal, who can approve a voice clone,
who can use an employee likeness, what
happens when someone leaves? What gets
labeled? What gets logged? What's never
allowed? Because if you don't define
this ahead of time, you're not making a
strategy decision. You're just waiting
for the mess to make the decision for
you. So, let's zoom back out. I don't
think the future belongs to creators who
never use AI. That's a fantasy. I also
don't think the future belongs to
creators who quietly automate themselves
and hope nobody notices. The future
belongs to people who can use AI without
breaking trust because trust is becoming
the scarce asset. Not content. We're
going to have infinite content and not
polish because AI can polish. Not even
voice. Apparently, you all heard the
clone. The scarce thing is judgment and
taste and accountability. The sense that
a real person made choices and is
willing to stand behind them. The buck
has to stop somewhere. So yes, someone
can clone my voice, but they cannot
clone the responsibility for what I
choose to say with it. And ultimately,
that is where I think the line has to
be. Being human is no longer enough,
right? You have to be legibly human in
this world. And if you're gonna be
synthetic, you have to be legibly
synthetic, too. So, happy building.
Cheers.