Certified 100% AI-Free Organic™ content

AI is going to be big. Anti-AI is going to be big as well.

One of my favorite products is “100% Fat-Free Pickled Cucumbers Fit (Gluten Free), “ which I once saw at the grocery store. I do like my cucumbers in all forms, but the thing I particularly like about this iteration is the very clear and targeted branding.

In order to lower the price of food, we have introduced new technology, manufacturing processes, and crop variations. Some are clearly ethical tradeoffs (like industrial animal farming), while others seem to have no obvious downsides but may have unforeseen consequences - like GMOs.

We have been enjoying a steady decline in food prices, effectively winning with hunger to the point where the opposite - obesity, is a problem.

As mass-produced food became easily accessible, the demand grew for food that is not efficiently produced. Since it is so hard to tell the extent to which our food has undergone processing, there is now a demand for clearly advertising and proving the origin of what we put into our mouths.

AI will be everywhere but underutilized.

Models like GPT-3, Stable Diffusion, and Midjourney are very useful tools for enhancing human creativity. They are immediately applicable to a variety of existing workflows and I’m confident that soon:

• Word and Powerpoint will generate pieces of content for you out of thin air (Microsoft now owns a 49% stake in the maker of chatGPT/GPT3 - OpenAI).

• Blogging platforms will help you write, edit, and illustrate your content. (I know because I am working on this right now)

• Your note-taking apps, editors, and image processors will sport generative tools to include in your workflow.

The limit truly is the human imagination, and we’re once again finding out it is not as boundless as previously expected. Faced with an empty text box that let’s you conjure anything under the sun in seconds, a window into the multiverse, most people type “a dog“, shrug, and continue with their day.

The way to solve this overwhelm is not to assure users that they “can do anything“, but to present boundless specific cases where the tools are helpful. Graphical interfaces work because they present all the options for the user to choose. I shared more thoughts about possible applications and user interface considerations in this thread.

So far, the App Developers were also coming up short in delivering the explosion of AI tools. The appropriate integration of existing graphical interfaces with these AI capabilities is a new problem to solve. Turns out Human Creativity works the same way as Artificial one: It mostly remixes what it has already seen.

ChatGPT (which is a little tweaked version of existing GPT-3 capabilities) is an example of how a small product effort can unlock new use cases. Much like Roger Bannister’s 4-minute mile, it’s proving that you can make a useful and successful product on top of the existing API without much effort.

And it’s kind of a problem.

AI-Free Label

I don’t think you can stop technological progress. Humanity moves in the direction of better and more sophisticated tools to make life more convenient, and it will be more convenient to introduce more and more AI-produced content into the culture.

Understandably, not everyone will like that.

1. In the Work of Art in the Age of Mechanical Reproduction, I explored the historical perspective of judging the artistic value of AI Content. Much like it was initially the case with film, some people feel really strongly that AI-Assisted content is not valuable

2. Some of the AI-generated output is factually wrong and requires closer scrutiny, especially if legal exposure is involved

3. The legal status of images is disputed, and I am sure the same will be of GPT-generated text as soon as it becomes mainstream

4. Published content will be later used to train subsequent models, and being able to distinguish AI from human input may be very valuable going forward

5. Certain jobs will get replaced without a doubt. Even my stated job description will not be safe in the long term

6. As Tim Ferriss pointed out in his latest podcast with Kevin Rose:

How much of it will be being certain that you are reading something generated by human before you're willing to commit to having an emotional response, even if the output is identical, right?
Well, there will be some authenticity of human production that becomes important. I could see that becoming important. Interesting. Like, do you really want to cry watching a movie that was 100% produced by robots?Maybe not

Long-Term effects of new technologies are always hard to predict and we (as technologists) don’t have the right to impose universal adoption.

Customers and the audience should be able to know and choose if they are interacting with AI-generated content. More importantly, they need to be able to choose to interact exclusively with artisanal, human-produced ideas.

Applying Kosher rules

There are multiple ways in which content can be assisted. It can be remixed, rewritten, or edited. AI can supply post ideas, execute human ideas or blend them together.

The question of the origin of content will be debated in the coming years, but for now, we don’t have a sensible framework. Especially for text, it will be very hard to determine which piece of text is AI-assisted and which is not. When did you take inspiration? Which sentence is your own?

If we continue to take inspiration from food, there is a long-lasting tradition that has to solve a similar problem. From the rules of Kashrut:

Utensils (including pots and pans and other cooking surfaces) that have come into contact with meat may not be used with dairy, and vice versa. Utensils that have come into contact with non-kosher food may not be used with kosher food.

If we followed a similar rule, then any content incorporating a piece of AI-produced content can’t be marked “100 AI-Free“.

Towards a web standard

As part of our AI explorations at Automattic, we hope to tag every piece of content incorporating AI-produced pieces coming from our systems. We don’t know how the future of this debate will unfold, but this is the last call to store this information before everything on the internet is of suspect origin.

At the same time, I started the discussion about a web standard for tagging such content across the platforms. I imagine a <meta/> tag or an HTML attribute that marks AI-generated parts.

If you know of similar work being done elsewhere, please reach out - I would love everybody to adopt an open standard so that customers have a choice.

Few random points

Howdy Hacker News visitor! Here are my previous posts that have trended.

• This email looks different! - You may have noticed that this email looks different from my previous ones. I am experimenting with a new platform - Substack, to learn what makes it tick. The substack team is doing a lot of things right, and I am here to learn (before AI takes over the writing of these articles).
  Here are my previous essays

• Managing time as a manager - In response to my colleague, I’ve written up my thoughts about managing my time as a Software Engineering manager with ADHD.

• This post is not, in fact, 100% AI-Free. As you can probably tell 2 images here are AI-generated.