Yesterday
it came out
that Big Walk used LLMs in its development.
In
House House's response
(archived here), they stated that the use was small and largely
consultative rather than generative, although some
amount of LLM-generated code was used as the basis for
code that shipped.
Big Walk was developed without an internal AI
usage policy in place, and we consider that lack of
policy a failure. We will work to have a more
comprehensive policy going forwards.
Reading between the lines, this final paragraph
suggests that LLM usage was not a matter of company
policy but rather one or a few engineers making
individual decisions about their workflows. It also
implies that they will institute a policy
against this in the future, but leaves open the
possibility that the policy they settle on will allow
some amount of LLM usage.
I'm sure there's a tremendous amount of bickering
about this happening on social media; I'm no longer in
that particular torment nexus, so I don't know the
details. I do know that this news will affect how I
think about Big Walk going forward, so I
think it's worth taking the time to revise my review
and describe how and why my feelings have changed. In
doing so, I hope to articulate my position on LLM
usage more broadly.
I loved my time with Big Walk. I don't hand
out five-star reviews lightly, but I was confident
this game would earn one as soon as I began to
understand just how cleverly it created new forms of
play and puzzle solving between friends. It was
clearly a work of masterful artistry, painted with the
palette of real human interaction which is
tremendously difficult to master but intensely
rewarding to experience. That artistry was real, and
it is still present regardless of the tools used.
At the same time, no piece of art exists in isolation.
All art is in conversation with the culture that
created it, with the art that surrounded and preceded
it, with the economic conditions under which it was
produced. If history is driven by material
contradictions, as Marx and Engels taught, then we can
expect to see aspects of that contradictory tension in
the artifacts produced during that history. Within
Big Walk, an artifact of impressive artistry,
we see the use of a tool whose very existence is
inimical to art as a discipline, a tool built from the
work of artists and craftspeople over their vehement
objections, a tool whose creators depend on the
immiseration of workers for their own success, a tool
whose effects on the physical world are profoundly
detrimental.
Knowing that LLMs were used doesn't remove the
artistry from Big Walk, but it changes
everything about its conversation with the culture
around it. Big Walk is now, indelibly, in
some part an attack on the artistry of all the games
that came before it and on the very possibility of all
the games that could come after. Even House House's
response gestures at this:
We have some concern that voluntary disclosures like
this might contribute to the normalisation of this
technology, or its treatment as “inevitable”. We
support the work of those who are trying to limit the
harm generative AI is doing to videogames and our
world, and apologise that Big Walk’s LLM
usage may not meet their standards.
Of course, the best way not to contribute to the
normalization of LLMs is
not to use them at all. But I suspect by the
time any discussions were being had at a company-wide
level, the horses had fled the stable and locking the
door would do no good. I truly do not envy them the
position of knowing people would view all their work
as tainted by a few decisions made in the heat of
debugging, and that there was no real way to undo the
results of those decisions.
Some may sympathize so much with this dilemma that
they take it as a sign that people are speaking out
too harshly against LLMs. I could not disagree more
with this position. On the contrary, we must do
everything we can to make the use of LLMs culturally
radioactive, such that any use in a cultural
product is treated as putting that product in conflict
with human culture as a whole. It doesn't need to be
black or white—we can acknowledge that
Big Walk is far less in conflict than in the
mountains of steamslop that continue to pile up around
us—but we need to be clear that the only amount of LLM
use that's truly acceptable is none. While we can
grant some grace to those who don't yet realize that
this is a conflict and there are sides, we must in the final accounting recognize that
works on the wrong side do real harm that we cannot
paper over.
Others may point out that condemning all use of LLMs
incentivizes lying about whether a game uses them, to
which I reply: good! If someone is going to do
something broadly (not specifically) harmful, they
should hide it in shame and fear that the
world might one day know. It can be hard to remember
in the brazen 2020s, but shame works as a deterrent.
Forcing undesirable behaviors underground mitigates
their spread! We've all seen this happening in
reverse: as it became more feasible for trans people
to exist in public over the last few decades, an
explosion of people realized they could transition
too. We can and must harness this power to drive LLM
use into the shadows. Responsible creators will
disclose the use when asked even if it stands to harm
them, as House House did. And irresponsible ones will
be forced to balance whatever they think they stand to
gain by standing on the necks of artists against the
very concrete cost of scandal if it comes to light.
In most of the world today, the people don't exercise
political power. Good work is being done to try and
change that, but it will not change tomorrow nor will
it change next year. But we can exercise power through
culture because, despite capitalist attempts to
capture it completely, part of culture is the
people. When enough of us speak in one voice and say
"this is not what we want art to be", the world will
listen.