Misaligned AI Triggers Trade Tariff Crisis
Key Points
- The speaker alleges that the Trump administration relied on large language models like ChatGPT, Claude, and Grok to draft recent tariff policy, citing a test by author Roit that reproduced the same errors across multiple AI systems.
- All the AI‑generated drafts mistakenly used a trade imbalance as the justification for tariffs, a fundamentally flawed approach that contradicts standard reciprocal tariff practices.
- This misalignment of AI output is presented as the first instance where an AI‑driven mistake directly contributed to an economic misstep, highlighting a broader risk of future crises caused by unchecked AI.
- The policy also erroneously applied tariffs to entities identified by top‑level internet domains—such as the uninhabited McDonald Islands—demonstrating how AI can misinterpret domain data as sovereign countries.
- The overall message warns that AI’s “keyboard‑to‑chair” errors can have real‑world economic consequences, urging greater scrutiny of AI‑generated policy decisions.
Full Transcript
# Misaligned AI Triggers Trade Tariff Crisis **Source:** [https://www.youtube.com/watch?v=GmaJsU7TQF0](https://www.youtube.com/watch?v=GmaJsU7TQF0) **Duration:** 00:03:06 ## Summary - The speaker alleges that the Trump administration relied on large language models like ChatGPT, Claude, and Grok to draft recent tariff policy, citing a test by author Roit that reproduced the same errors across multiple AI systems. - All the AI‑generated drafts mistakenly used a trade imbalance as the justification for tariffs, a fundamentally flawed approach that contradicts standard reciprocal tariff practices. - This misalignment of AI output is presented as the first instance where an AI‑driven mistake directly contributed to an economic misstep, highlighting a broader risk of future crises caused by unchecked AI. - The policy also erroneously applied tariffs to entities identified by top‑level internet domains—such as the uninhabited McDonald Islands—demonstrating how AI can misinterpret domain data as sovereign countries. - The overall message warns that AI’s “keyboard‑to‑chair” errors can have real‑world economic consequences, urging greater scrutiny of AI‑generated policy decisions. ## Sections - [00:00:00](https://www.youtube.com/watch?v=GmaJsU7TQF0&t=0s) **AI-Driven Tariff Miscalculation Claims** - The speaker asserts that the Trump administration used large language models such as ChatGPT to draft recent tariffs, which—due to a repeated error of basing duties on trade imbalances rather than reciprocal rates—constitutes a misaligned AI decision that precipitated an economic crisis. ## Full Transcript
So there's considerable evidence that
the Trump administration used Chat GPT
or Grock or another LLM to create trade
policy specifically the tariffs that
were announced yesterday. The reason we
know that is that Roit who is an author
uh he wrote uh building God which is a
fun book he actually ran a prompt for
trade tariffs through claude 3.7 through
chat GPT multiple models through Grock
what he found was the same mistake in
all of them where the basis for a tariff
was a trade imbalance and a trade
imbalance is actually not a basis for a
tariff because if you're doing
reciprocal tariffs, you want to match to
a tariff rate in another country. And if
you are doing trade imbalance tariffs,
that's not what you're doing at all.
You're just sort of making up math to
divide the amount of goods some country
buys by the amount of goods we buy here
in the US and dividing that out and
that's how you get your tariff. And
that's what they did. And that's also
what the LLM recommends and that's what
zero economists would recommend. And if
you're like, Nate, why is this a tariff
channel? The answer is it's not a tariff
channel. I just studied economics, lived
through multiple economic crises on
multiple continents. So, it's sort of
interesting to me. And also, it's AI.
It's the first case where arguably a
misaligned AI caused an economic crisis.
It's kind of a big
deal. And you can say, well, it wasn't
intentionally misaligned, but I suspect
we'll be saying that about a lot of
future crises caused by AI if that
happens.
So the second thing we know also
suggestive of AI writing this entire
policy from scratch is that it's not
actually
countrybased. Countries have political
governments and can impose tax rates,
tariffs, etc. Top level internet domains
do not map to countries. They partially
map to countries. We think they do
because our top level domain does, but
they don't entirely. And so when there
was a lot of laughter going around the
internet about the Trump administration
putting tariffs on an uninhabited island
with penguins, uh, you know why? They
have their own top level internet
domain. The herds of McDonald Islands
have their own top level internet
domain,
HM. And that's why they got tariffs
because someone dumped the name of
entities with top level internet domains
and decided that was going to be the
object against which they wrote tariffs
and that is
incorrect. That is what an LLM might do
if you didn't check it, but it's
incorrect. So, misaligned AI caused an
economic crisis. Or you could read it as
the most recent example of the problem
is between keyboard and
chair. I'll leave you there. You can
sound off in the comments. It's been a