I purchased my first computer in 1982. It was a Timex Sinclair 1000—a small device about the size of a 150-page trade paperback book you hooked up to your television. It had a small membrane keyboard that you could use to write and run programs in the BASIC programming language, but it wasn’t powerful enough to run productivity applications such as a word processor. A few months later, after I’d moved to San Francisco to make my living as a writer, I bought a KayPro II. The Kaypro was much closer to the personal computers we use today, with a built-in screen and a very good keyboard. It was a strong competitor to the Apple II. It came with a suite of productivity software, including Perfect Writer—a powerful word processing program that supported add-on programs including Perfect Speller and Perfect Thesaurus. I launched my writing career using the KayPro II. While the hardware, operating system, and applications I was using were primitive by today’s standards, the Kaypro helped me become a more productive writer, and over time a better writer because it made it easier for me to review, revise, improve, and share my work.

We are at the same moment with generative AI. GPT-3, ChatGPT, DALL-E, Stable Diffusion and the other generative AI tools we are using today will seem primitive in just a few years—and some of our fears about the impact of generative AI may, in retrospect, seem overstated.

It’s unfortunate we have introduced these tools as forms of artificial intelligence, a loaded term that implies capabilities that are beyond those found in these early generative systems. The generative AI tools we are interacting with today are examples of advanced machine learning, not artificial intelligence as it’s commonly portrayed in science fiction. They can be very useful for augmenting our work, but in most cases, not a viable replacement for us. Unfortunately, many of the conversations on generative AI assume that a giant leap ahead to Artificial General Intelligence (AGI), a machine-based version of the full human brain, is inevitable. While today’s generative AI systems can mimic some aspects of human intelligence, they aren’t capable of independent thought, they have a limited scope of action, they can’t handle open-ended problems, they have a limited amount of knowledge, and they can’t upgrade their knowledge on their own. It’s possible that generative AI systems will have all of these capabilities at some point, but it’s far from certain. What we have today isn’t even close to AGI, but it is a significant step forward in the enhancement (augmentation) of human intelligence.

The Other AI: Augmented Intelligence

Augmented Intelligence (also known as “Intelligence Augmentation,” “Intelligence Amplification” or “IA”) is a term used to describe knowledge-based tools that work in collaboration with humans to enhance their cognitive and problem-solving capabilities. These tools are designed to enhance human intelligence, rather than replace it. Humans are at the center of intelligence amplification, not the periphery.

While today’s generative AI tools can produce images, text, and other types of content that can replace human generated content in some contexts, they still require varying levels of human participation. Humans help train the systems, establish the rules that guide the systems, review and refine the content they produce, and in most cases, independently adapt the content for its intended use. Like my primitive KayPro computer, today’s generative AI systems are primarily tools that can help us be more productive and creative—not replacements for us.

Is some of the content produced by generative AI suitable for use as is? Yes, which says something both about the advancements we’ve made in machine learning and the changing nature of content in our culture.

On Content

What do we mean when we use the word “content?” The answer used to be straightforward.“Content” refers to both the medium of expression and the ideas expressed: a book, the words on the page, and what you experience reading the book (how the language moves you, the ideas you encounter, the insights it reveals). A film, its visual images, and the experience of watching it. A song, its music and lyrics, the places it takes you and emotions you feel as you listen to it. But today, “content” means something different—it’s the word we use to refer to anything that circulates online: text, images, videos, music, and programmed interactive experiences. As the media historian and theorist Kate Eichhorn writes in her book Content, “While some content conveys a message, shares information, or tells a story, content doesn’t need to communicate anything at all. Content is often produced to circulate and not to communicate.”

The content we produce spans a wide range of uses. Each use has its own way of calculating value: a dynamic mix of its utility, quality, impact and cost. A significant amount of the content we come into contact with online is designed to increase the producer’s “content capital”—”a form of capital, analogous to social, symbolic, or cultural capital, acquired through the production of content about oneself or one’s work” (Eichhorn), or one’s products or services. An increase in content capital can increase the visibility of the producer in search engine results and their authority in their subculture and even the culture at large. We even have a name for these authorities: influencers.

Tools like generative AI are very useful for producing content whose only purpose is to circulate. GPT-3 and ChatGPT are very good at search engine optimization (SEO), writing for Twitter and other social networks, and producing short, catchy blog posts. This is certainly a form of creative production, but it’s not the same as the sustained, deep engagement that artists undertake when they write a story, novel, play, poem, or work of nonfiction; or when they create a drawing, painting, video, or film. I do not want to diminish or devalue any form of creative expression: my goal is to call attention to the wide range of forms of creative expression and the various environments in which these forms of expression exist. We can’t use one yardstick to measure the potential and applications of generative AI. We need a range of evaluation criteria.

Judging Orginality, Impact, and Value

Your intentions matter, especially when it comes to working with generative AI technologies. In some cases you're looking for a "good enough" version of work that has been traditionally produced by a human, but only needs to meet a basic level of quality in order to fulfill its function. For example, the header images that now appear at the top of most online articles are a pleasant bit of ornamentation, but don't need to be uniquely creative or well executed to fulfill their function, which is to draw your attention to the article, especially in visually dense environments that feature thumbnail images.

In other cases, you’re measuring the quality of your work against the expectations of others in your family, social circle, workplace, field, or culture. The bar is likely to be much higher, which means the quality of work solely produced by a generative AI system is unlikely to be satisfactory. Your perspective, knowledge, skills, and judgment anchor the creative process. You can use generative AI tools to augment your work, but they can’t replace you. (See my article The Elusive Definition of Creativity for an introduction to the Four-C Model of Creativity and who judges originality, impact, and value.)

Generative AI is a New Medium

I began this series with an article titled The Mirror and the Lens that explored how the Old Masters used mirrors and lenses to create live projections of the visual world around them as they learned how to render three-dimensional space on two-dimensional media. Their investigations and experiments led to the introduction of perspective in drawing and painting, and eventually artworks of astonishing realism. They also led to the invention of the camera and birth of photography. The reflections and refractions of light they captured changed the way we see the world around us, and the way we see ourselves.

Generative AI is not just a new tool: it’s a new medium. Our interactions with generative AI produce tangible artifacts that express our creative intentions. Like other mediums (oil painting, for example), each generative AI system has its own characteristics that you have to consider as you work with it to express your intentions. And like other expressive mediums, generative AI systems have inherent limitations that you can only overcome with insight and invention—just as the Old Masters had to learn how use vanishing lines, shading, and techniques to create the optical illusions that create a sense of three-dimensional space on a two-dimensional surface. Generative AI systems are designed to emphasize the plastic nature of content—the systems reduce words and images to tokens, which are essentially the atomic units of thought in the system. The content generative AI systems create is an aggregation of statistical probability calculations, not logic-based decisions, which is not what we experience with most augmented intelligence tools, such as the first generation of chatbots that made most of us slam down our phones in frustration.

The challenge of working with a text generation system in a creative context is disengaging from your own default logic-based binary thought pattern. Instead of just asking whether a text response is “right” or “wrong,” or “good” or “bad,” think of the result as a hypothesis, a supposition, a proposed response. Use your senses and emotions to add another dimension to your understanding. Read the response to your next prompt and ask yourself:

• Is the response the type of output you requested in the prompt? For example, if you asked for an outline and the system returned a paragraph, it returned the wrong type of content.

• Is the response relevant to the request in the prompt? Sometimes generative AI systems “hallucinate”—they go off on a tangent that’s unrelated to request.

• Are the facts in the response accurate? If you’re unsure, take the time to fact-check the response.

• Is the response the right length?

If the answer to one or more of these questions is No, regenerate the response.

Next, read it again, this time out loud (softly will do…) and ask yourself:

• Is the response coherent? Does it make sense? Is the transition from point to point easy to understand and logical?

• Is the tone of the response right? Is it too formal or informal for the intended audience?

• Is the expression of the ideas in the response unique? Does the response contain language that’s obviously similar to, or the same as, other content on the subject?

If the answer to one or more of these questions is No, regenerate parts of the response or the whole response, or begin editing it.

Now read the response out loud one last time and ask yourself:

• Who does this response sound like? Does it sound like you? If not, why? How is it different from the way you write?

• What patterns of thought does the response follow? Does the flow of ideas follow your usual patterns of thought?

• What are the biases in the response? Do they align with your beliefs and values?

And perhaps most importantly, ask yourself:

• Does this feel right?

If the answer is No, regenerate the response.

The list above is just one way to use generative text tools to augment your productivity and creativity. Whatever process you choose, make sure you keep yourself at the center of the process. Use the system to help you express your ideas in a way that sounds and feels like you.

The Question…

The invention of photography in the mid-1800s freed the visual arts from realism. That freedom came with an essential question that each artist had to answer: What does it mean to be a visual artist in the age of the photograph? The arrival of generative AI demands that today’s artists answer a similar question: What does it mean to be an artist in the age of generative AI?

I don’t know the answer… But I know that finding the answer requires curiosity, an experimenter’s mindset, and patience. Those three things, and a good connection to the Internet. Have fun!