On AI


Some AI is all well and good. If you’re making a neural net specifically
to better detect cancer, great. Make sure you ethically source your data
and don’t need several data centers to run it, and we’re good. Clearly,
for “Generative AI”, this does not apply.

The following quote, which I paraphrase entirely from Brent Yorgey,
summarizes my stance on Generative AI accurately:

  Although LLMs are fascinating from a purely technical perspective, I
  refuse to participate in or contribute to such systems that are built
  on massive exploitation of human labor and make profligate use of
  scarce resources.

To put it differently, as far as I can tell, this technology is actively
harmful to human society. Even in the best case, Generative AI working
as intended.

There are a bunch of other reasons. For a more in-depth discussion of
these, I recommend Sean Boots’ blog post on Generative AI
vegetarianism.[1] It should not be too difficult to find other sources
once you get a hang of the keywords. My AI Skepticism blog posts are
outdated, but provide a starting point nevertheless.

You’re best off not using on these tools. Rely on your brain instead.

But what if I need to use Generative AI?

I’d prefer if you reconsider.

Yet, I must also acknowledge there might be reasons for using it. Maybe
you are a teacher, and want to keep up to date with technology so you
can inform students of the do’s and don’ts. Or maybe you are a
communicator on Generative AI capabilities seeking to inform the public
of the tells and dangers. Do realize that by using these tools, you are
indirectly endorsing them: actions speak louder than words.

If after all this, you still think you need to use Generative AI, I have
failed to persuade you. Please, in your future Generative AI-related
endeavours, take the following into account.

Use only open weight models

In the same way it can be ethical to use medical data that was acquired
in an unethical way, open weight models can be considered ethical to
use.

Power Generative AI with green energy

While you cannot prevent the waste of water, you can prevent, to some
extent, the spillage of unnecessary carbon into our atmosphere. Advocate
for, and prefer, datacenters powered by green energy over others.

If it is not possible to ensure your datacenter is using green energy,
compensate your Generative AI bill by donating, in equal measure, to
green causes. Feel free to pick a local one that directly benefits you.
If you’re not sure, prefer a global heavyweight, e.g. Cool Earth.

Use open-source agentic harnesses

Avoid agentic coding harnesses provided by Big Tech, such as Claude Code
and Codex. Instead, prefer open source ones. One I’ve heard of is
oh-my-pi; this is explicitly not an endorsement. You will be less
reliant upon Big Tech, and sleep better for it.

Don’t use more than you need

Maybe a weekly or monthly experiment is enough to form an opinion on the
current state of Generative AI. I doubt it’s necessary to make
Generative AI a primary component in your daily workflow.

Talk about the consequences of using Generative AI

Most importantly: talk to people around you how they square their,
typically superfluous, use of Generative AI, with the implicit goal of
being a good person. Consider mentioning, primarily:

-   environmental harm,

And secondarily:

-   ethical harm,
-   economical collapse,
-   loss of personal agency, and
-   mental degradation and dependency.

My favourite version of this is asking someone how to, in 18 years time,
explain to my children why the world has gone to shit. Every day I’m
hoping I’ll never need to.

[1] Backup link 1, backup link 2.