Everybody has ethics. We make daily choices whether to have a takeaway pizza, or donate the same sum to a homeless shelter. We vote, if we are lucky enough to live in a country where that is both possible, and impacts policy decisions. We make choices about respecting the law, or infringing it - everything from stealing bread for starving people, to whether it’s ethical to have a billion dollars while others starve.

When you use LLMs in any way, you are making a very complex ethical choice.

Empathy

One of the hardest things to do as a person, is to put yourself in the perspective of another, and to see their reasoning and feelings. To understand how they see the consequences of the choices being made, and what factors are relevant to them, is a truly difficult task to do, let alone to do it well.

In discussing what a group like the FreeBSD Project should do as far as AI/LLM policy is concerned, even the framing of the question and choice itself is significant.

Over the last few years I have had many many conversations with people from all across the world, and I’ve tried to understand how people have arrived at their various positions.

This short piece breaks down a few of the choices and framing that people have grappled with, and hopefully gives you a way of understanding other people’s positions, and reforming and informing your own.

Organisations and Harm

As individuals, our responsibilities and our ethical choices are usually reasonably easy to navigate. But it is not so easy, to decide how far these choices should be brought into the communities we encounter. For each of the subsequent choices, we need to decide if our community should concern itself with this, or not.

If a corporation spills oil in a nature reserve, is the company liable for the harm, or the individuals within the company? Or, is this a collective risk borne by the country?

Should a State be held accountable for the actions performed by its military, or should the individuals who approved those choices? Maybe both?

In the same light, how much of the negative consequences of LLM usage should be considered by the FreeBSD project? Is it “just a tool”, or should the project itself take account of the widely acknowledged societal harms?

If you’re interested in exploring this sort of issue, then read at least the first chapter of Organizations as Wrongdoers, a recent book by Stephanie Collins.

Framing

As an individual, our direct influence over government policy, and global changes, is minimal. But as a group, these differences become much more significant.

We cannot ignore those consequences, whether as individuals, or as part of larger groups, without also accepting the consequences of those choices.

If you believe that a project or business shouldn’t concern itself with externalities, then that framing is still making an ethical choice, and also the very lack of making a decision, is also an ethical choice.

So, the first major framing is to decide where you draw the boundary. In the age of The Anthropocene, should LLM’s negative impacts be included, and if so, how much and how far?

Just like choosing a pizza, using LLMs means that you also partake, however diffusely, in the higher order consequences of the inputs and outputs to the systems that build LLMs.

Copyright

FreeBSD has had a long and interesting relationship with copyright and ownership, stretching back to the very earliest of days of the UNIX wars. While I wasn’t personally involved at all during those days, it seems to me that concerns of copyright, licensing, and ownership have been very much part of the history of FreeBSD, and should still be so today.

People have successfully persuaded LLMs to regurgitate, verbatim, content of newspaper articles, songs, and software. Should Copyright and LLMs require that the provenance, no matter how awkward, be traceable?

Both BSD license and GPL variants require strong attribution, amongst other things. Is it correct that an LLM, that can provide code, stripped of context and license, in such a way that the code appears to be unencumbered?

Should FreeBSD be concerned that sloppy provenance could introduce legal risk in future, if it were to be included in an official release?

If you are a downstream builder, of a commercial product, should this concern you? I am sure that large corporate consumers of FreeBSD, such as Apple, Dell, Hewlett-Packard, Netflix, Nintendo, Sony and friends, can probably afford to lawyer up and drag the debate out in court, but this is not the same thing as the status quo of having a broadly trusted, BSD Licensed code base.

If you are an end user, could you trust the license as much as you used to?

LLM training set data has, by all accounts, been a massive land-grab of data, not restricted to code licenses, and not supporting attribution. Copyright and LLMs and subsequent by Lee Gesmer, tech lawyer https://masslawblog.com/about/ , covers this nicely.

Scraping

The FreeBSD cluster admin team, and many FreeBSD committers, are regularly prevented from using our Bugzilla bug tracking system, accessing our Phabricator patch management system, due to massive predatory load from LLM systems that do not respect /robots.txt and require active punitive measures to block and dissuade crawlers that are slurping up every commit, every bug tracking comment, and every patch revision. These are presumably being post-processed by any and all LLM vendors to improve their ability to produce reasonable looking code for the masses.

This is a considerable waste of our precious volunteer community time, and its effort that we cannot spend on productive activity. It’s also not at all fun, to be an incident responder trying to clean this up.

Clearly FreeBSD is not the only project that is suffering, and that has concerns about this ongoing criminal misuse and abuse of a public service.

The lack of ethics of the companies involved here is evident. Should we accept LLM generated content with one hand, while the other hand is being repeatedly attacked, with no support for managing that load? Are the AI companies willingly providing the FreeBSD project with free CPU, memory, and storage, all of which are being wasted?