Categories
Writing

Meaning through truthfulness in language

Liked Eleven Urgent & Possibly Helpful Things I Have Learned About Writing From Reading Thousands of Manuscripts by Jeannine Ouellette (Writing in the Dark with Jeannine Ouellette)

Our most important job as writers is—I believe—to make language capable of telling the truth. The essayist, activist, and poet Wendell Berry has written about this idea many times. In 2010, he addressed it in a letter to an English teacher and her class, writing: “By taking up the study of writing … you are assuming consciously … a responsibility for our language. What is that responsibility? I think it is to make words mean what they say. It is to keep our language capable of telling the truth. We live in a time when we are surrounded by language that is glib, thoughtless, pointless, or deliberately false.”

We must reject overly easy, overly familiar images and phrases and push ourselves instead for the slight adjustment that can make a world of difference.

This is what it means to defamiliarize language enough to let it hold truth. If the language is so familiar it washes over us, any truth it contains will be lost.

Categories
Technology Writing

Bias is baked into the current state of AI fiction writing

Bookmarked Wordcraft Writers Workshop (g.co)

The Wordcraft Writers Workshop is a collaboration between Google’s PAIR and Magenta teams, and 13 professional writers. Together we explore the limits of co-writing with AI.

Interesting assessment of co-writing with an AI — it’s limited by its inability to perceive / remember context, a very generic, stereotyped and mainstream understanding of genre and stories, and mediocre prose without voice.

Allison Parrish described this as AI being inherently conservative. Because the training data is captured at a particular moment in time, and trained on language scraped from the internet, these models have a static representation of the world and no innate capacity to progress past the data’s biases, blind spots, and shortcomings.

The computer trying to insert a man into a lesbian love story 😬 We see time and again technology incorporating and reflecting real world biases. It feels like they think preventing bias is an afterthought, something that can be fixed after the fact. These tools will likely become quite important in the future.  Can someone integrate people of color and queer people into their design process upfront?

I am intrigued by co-design, and feel like this project could benefit from it: learning upfront from writers what their biggest struggles are and where they wish they could have assistance. This feels a bit like, “we made a thing that makes words, let’s have some actual writers try it out and see what they do with it 🤷‍♀️”

One thing that writers often need is bit part characters. With existing biases, will the AI suggest all straight white men to fill these roles? When they create characters of color will they be caricatures?

Again, the training set proves itself essential to the tool — and behind many of its failings.

I read Robin Sloan’s short story, which was a clever little work that capitalized on the program’s strengths while critiquing reliance on shortcuts (and maybe poking a bit of fun at GRRM).

Categories
Art and Design Society

Will “good enough” AI beat human artists?

Replied to

The problems of relying on AI art

AI leads towards visual convergence when trained on generic material not unique to different cultures or styles, always going to come up with the go-to visual and nothing unique unless instructed by a human. Will continue to allow the current visual paradigm to dominate. Sometimes the archetypical rendering is fine, the unique elements are somewhere else, but relying only on that will not create new visions of the future for sci-fi renderings.

The computer is limited by the input it receives, and cannot make estimations outside of 1) what it is given 2) what the scientist-academic nudges it to do 3) the scope of the project…

It cannot adequately have the dataset to make everything, because it’s limited to who can give it that data and how that data is acquired. So much of what artists are inspired by come from non-digital, non-archived sources: stories from our ancestors, inherited cultural modes, language (which affects our metaphors and perceptions of time and philosophies), animals wandering around, sensory experiences, memes, etc…

Basically, what I am saying is that just like humans, the AI is limited by its inability to access information it doesn’t have.

— Reimena Yee, The Rise of the Bots; The Ascension of the Human

Will good enough win when it comes to art? If it’s between free and paid, the free version may be good enough for a lot of commercial uses…

Is convergence enough to stop “good enough”?

In other creative fields, art is already converging to homogeneous looks and sounds:

To minimize risk, movie studios are sticking with tried and true IP, and simply adding onto or remaking existing works.

Will illustration and the visual arts follow the same trend? For some commercial art needs, the purpose is to fit a tight-fit visual niche — think romance book covers, or organic food packaging, where the goal is to communicate quickly what category of product it is.

But, some art — like magazine covers — does need to stand out. Distinctiveness is part of the goal. This is where creative work can persist despite “good enough” in other areas.

Will AI-created artwork achieve its goals?

Example: cover illustration

The art on these covers is pretty enough but the type is bad:

If you just need a placeholder cover these seem fine, but I’m curious whether these are enticing enough to sell books. Probably something you could use for a lead magnet, something you’re not selling but just want to have a cover in the Kindle library.

Example: comics

Some fine vibe-setting panels for a comic, but not super useful for storytelling, the panels are too similar, and how good will it be at action? I can’t imagine it will naturally generate unique poses and dynamic angles to keep scenes visually interesting. Just a few pages of this feels slow-paced.

If this is the only kind of art it can produce, it will only be useful for indie literary type comics. I think what’s going on is that grand vistas look impressive and are hard to draw, but the AI’s problems are also more apparent at closer scales, where it adds weird distortions or things don’t align we’ll. Our brains can ignore or fix the problems in a vista, but they’re impossible to ignore when they’re the focal point.

I would guess, like Ursula Vernon, AI will be a tool to reduce workload for artists needing to draw complex environment panels, and an asset library for rendering environments. In current state Vernon found it needed a lot of post processing.

This art style looks beautiful now, kinda Monstress – esque / movie concept art, but I suspect that the more people use it, the more generic it will feel and people will value art that’s clearly created by a human / has its own visual style.

Implications for the industry

This tech could push down editorial illustration prices so only newbies who live on starvation wages will be able to compete with AI, plus high end artists who can retain boutique clients that value uniqueness and want to signal that they are a luxury publication / brand, so the middle career folks will disappear. Or, will only high end creators with distinctive appeal be able to keep working and all junior creatives fade out?

If you’re a creator, you either have a style or you don’t. If you don’t, you’re simply a gig worker. And if you have a style, there’s a computer program that’s going to not only encourage people to copy your style, but expand it.

For some, this is going to lead to enormous opportunities in speed, creativity and possibility. For others, it’s a significant threat.

— Seth Godin, Unprepared as Always 

Not yet, but…

I’d say AI is not good enough *yet* for most use cases, but it will get better over time. In the long run there will be less work for creatives actually producing their own renderings (linework, painting, photoshoots) and more the art direction angle of knowing what prompts to give the AI to get what you want, plus correction of obvious rendering errors.

At the low end of the scale, a broader range of fields will be impacted (logo design, basic graphic design) — will enough small scale jobs be accessible to early career folks that the industry won’t collapse in 20 years, because no one was able to get the experience?