Genre Grapevine on Why They Want Us to Call it AI
The Deceptive Language Around Machine Learning
Note: My regular Genre Grapevine column will be out in a few days. Apologies for the delay but I've been managing the recent SFWA Auction and am still finishing that work (a big thanks to everyone who took part or placed a bid). Also, this is the third part of my ongoing coverage on what machine learning tools mean for artists and writers. Here's part 1 and part 2.
Words are power. The words we speak shape our understanding of the world around us. The words we learn craft our understanding of ourselves.
And the words describing much of the conversation around artificial intelligence are self-servicing corporate doublespeak.
This isn't a critique of machine learning itself. As I covered in my first report on what AI generated art and writing might mean for creative professionals, machine learning will likely become an essential tool used by many writers and artists in the coming decades. However, the downside to machine learning is how these tools will affect the livelihoods of writers and artists. The justifiable fear is that corporations and those with power will use these tools to cut artists and writers out of their already marginalized positions in society.
And the language used to describe machine learning tools is an essential part of how this could happen.
Let's start with the term "artificial intelligence." There is no actual intelligence behind art platforms like DALL-E 2, Stable Diffusion and Midjourney, or writing platforms like ChatGPT or Sudowrite, aside from the humans who created and manage the programs. Instead, these creative systems are examples of machine learning, using algorithms crafted from data samples (also called training data, but more about that term in a moment).
I personally believe humanity will one-day create true artificial intelligence. But as this must-read essay by Ted Chiang says in describing computer "recursive self-improvement" and the singularity, we're still a "long way off from being able to create a single human-equivalent A.I., let alone billions of them."
But if there's no intelligence behind these platforms, why are these tools called artificial intelligence? Why do the creators of ChatGPT call their chatbot an AI instead of a proper term like large language model?
For the same reason that Disney World calls their employees "cast members," corporations use the term "right-sized" when they lay people off, and workers are given a "friendly reminder" right before a manager fires you. In short, all these are examples of corporate doublespeak, where language is used as "an agent of power and control."
Calling ChatGPT an artificial intelligence is a textbook-perfect example of this type of corporate doublespeak branding, especially when there are very legitimate concerns that large language models are unable to lead to an actual human-like intelligence. Calling ChatGPT an AI instead of machine learning captures the human imagination and attention. And for the companies creating machine learning platforms, calling their creations AI also equates to a massive influx of funding.
But let's also circle back to how corporate doublespeak is an agent of power and control. As covered in my previous report on AI, a major aspect of the current Writers Guild of America strike is that Hollywood writers don't want to be replaced by or forced to use AI systems. The Guild has also made a strong push to highlight the plagiarism and theft of copyrighted works that are a central aspect of current machine learning tools.
As this Guild statement says, "It is important to note that AI software does not create anything. It generates a regurgitation of what it's fed. If it's been fed both copyright-protected and public domain content, it cannot distinguish between the two. … To the contrary, plagiarism is a feature of the AI process."
Which takes us to another language choice in the machine learning field. Instead of saying that machine learning tools copy, steal, or duplicate works created by human artists and writers, instead the term "training" is used.
If you say a "machine" is "copying" millions of copyrighted works, people would wonder why those in control of the machine are allowing this to happen. But when you say an artificial intelligence is "training" on data, the process sounds more organic and not a deliberate attempt to utilize works that the company lacks the right to use.
In short, the language used obscures how the people and corporations who created these machine learning tools are deliberately copying legally copyrighted works. The language used also makes it sounds like these machine learning tools have agency and are choosing on their own what they copy.
And another term we're hearing lately is how these AI tools can "hallucinate" or suffer from "delusions," which is doublespeak for saying machine learning can easily generate outright mistakes that aren't supported by the data set supporting the tool.
As Benj Edwards wrote in Ars Technica, use of the term hallucinations is increasingly controversial because it "anthropomorphizes AI models (suggesting they have human-like features) or gives them agency (suggesting they can make their own choices) in situations where that should not be implied."
And as Naomi Klein wrote in The Guardian, "why call the errors 'hallucinations' at all? Why not algorithmic junk? Or glitches? Well, hallucination refers to the mysterious capacity of the human brain to perceive phenomena that are not present, at least not in conventional, materialist terms. By appropriating a word commonly used in psychology, psychedelics and various forms of mysticism, AI's boosters, while acknowledging the fallibility of their machines, are simultaneously feeding the sector's most cherished mythology: that by building these large language models, and training them on everything that we humans have written, said and represented visually, they are in the process of birthing an animate intelligence on the cusp of sparking an evolutionary leap for our species."
I agree with all that. I also think what Edwards and Klein are saying about hallucinations can easily apply to much of the language used to describe machine learning.
Unfortunately, the doublespeak language used to describe machine learning tools has already filtered out into the general population. If you call these tools machine learning, many people will have no clue what you're talking about. For now, the term AI appears to have stuck.
And this is despite leading figures in the history of machine learning such as Michael I. Jordan also wishing that people would stop using the term because it isn't technically correct.
Ted Chiang has been doing amazing work speaking out about what all these developments mean for humanity. In a new essay in The New Yorker, he writes "When we talk about artificial intelligence, we rely on metaphor, as we always do when dealing with something new and unfamiliar. Metaphors are, by their nature, imperfect, but we still need to choose them carefully, because bad ones can lead us astray."
The same can be said about language. And especially the language used to describe machine learning.
Later in his New Yorker essay, Chiang asks "Is there a way for A.I. to do something other than sharpen the knife blade of capitalism?" In that statement Chiang is talking about the specific relationship between capital and labor, "in which private individuals who have money are able to profit off the effort of others."
He adds, "As it is currently deployed, A.I. often amounts to an effort to analyze a task that human beings perform and figure out a way to replace the human being. Coincidentally, this is exactly the type of problem that management wants solved. As a result, A.I. assists capital at the expense of labor."
When we use the language and terms those creating machine learning platforms want us to use, we're buying into their beliefs around who has power and control over these creations. But it doesn't have to be like this.
If we want these tools to actually benefit everyone and not be yet another method for corporations and the powerful to take away our livelihoods, we should start by not using their language. Yes, sometimes we'll have to use the term AI because that's what people understand these days. But don't say a machine learning platform is "training" when it's actually copying copyrighted works, or that it is "hallucinating" when it is instead producing mistakes.
There's a reason members of the Writers Guild of America are using the term "plagiarism software" to describe machine learning.
The reason is that language matters.