AI Changes Everything and Nothing
The marketing frenzy surrounding the recent flood of AI-powered apps and services has caused some observers to question the long-term viability of creative roles.
However, most sweeping technological change overpromises and under delivers, the responsibility for which lies with overly enthusiastic marketing departments, and our own insecurities. Technology never replaces designers, it either finds a home in our toolbox, or it doesn’t.
Once you strip away the hyperbole and take a look at the feature set, most ‘revolutionary’ technologies have been intended as supplementary to existing workflows. Human beings define product roadmaps; AI is no different.
AI certainly does an impressive job of imitating creative work. Some of its output is superficially convincing. But it lacks depth, and it is (quite literally) incapable of thinking outside of the box.
AI could not have written the last four paragraphs; that’s not how AI works.
How AI Works
In every social group, there’s a know-it-all who’s an expert on everything. No matter the subject, they will correct anything and everything from the best brand of tinned tomatoes to the capital of Eswatini. And we tolerate it because we know that something traumatic happened in their childhood to make them feel small and foolish, and everything they say and do is an attempt to win our approval — they were programmed to always know the answer.
That person is AI, pontificating on subjects with a tenuous understanding because it has been programmed to always know the answer.
Stephen Wolfram wrote a fascinating and detailed explanation of how ChatGPT works. In overly-simplistic terms, it looks at all the possible next steps and selects the most likely outcome. For example, if AI had generated the previous sentence, it may have inserted “outcome” as the last word, or “result,” or “conclusion,” but it was very unlikely to insert “spinach.”
The problem arises when there is no clear choice, as occurs when a concept or topic is new and it doesn’t have sufficient ‘correct’ answers to steer the language model’s direction. Prompt AI to generate something it can’t find a pattern for, and it will find the closest pattern it can. Even if the chances of the last word in the sentence being “spinach” are 0.000000001%, in the absence of a better option, that is what AI will insert. What is more, having opted for “spinach,” any future output will be distorted to normalize the text. Eventually, AI will output that its creator was Popeye.
The Real Threat From AI
The real threat from AI isn’t that it’s so good that it will replace creatives. The real threat is that we will be collectively gaslighted into thinking it can. The technology press has been heralding the “AI revolution” for months, but revolutionary is precisely the opposite of what AI is.
Let’s prompt ChatGPT to answer a query that clients often ask me in kick-off meetings: What’s the best CMS for our site?
Here’s ChatGPT’s response: “The choice of CMS (Content Management System) for building a website depends on various factors such as your technical expertise, budget, website goals, and personal preferences.” So far, so vague, but then it went on the recommend six possible CMS in this order: WordPress, Drupal, Joomla, Wix, Shopify, Squarespace.
ChatGPT isn’t exactly wrong. WordPress is a CMS, and it is the first choice for most sites. It’s also out-of-date, there are more professional solutions available. AI, by its nature, reinforces the status quo. It can only see the world that is.
The real threat from AI is not revolution; it’s entropy.
A Dumb Feedback Loop
The content used to train AI is the key to successfully mimicking people. However, not all creative or critical thought is equally valid. Education, experience, and talent mean that some opinions are more balanced, long-sighted, and grounded in fact. To quote Agent K: a person is smart, people are dumb. If AI was allowed to learn indiscriminately, it would adopt a keto diet, get a face tattoo, and vote to leave the EU.
How did ChatGPT know to list WordPress as the top CMS? As part of its training, it scraped the web and found thousands of blog posts stating that WordPress is the top CMS.
Most of those blog posts aren’t critical assessments, they’re just repeating a known fact. Now imagine all the companies padding blogs with AI-generated posts, repeating the known fact that WordPress is the number one CMS. Now imagine the next-generation language model scraping those posts and outputting the same known fact that WordPress is the number one CMS. A feedback loop emerges in which WordPress can only ever be the number one CMS. The true thing will continue to be repeated, which will continue to make it ‘true.’ The status quo is maintained, and there is no innovation.
A few weeks ago, OpenAI released a tool for identifying AI-generated content. Sadly, it doesn’t work terribly well (yet). But what is interesting is that OpenAI allocated development resources to it. I suspect OpenAI’s interest in dynamically detecting AI-generated content is to avoid the feedback loop by ensuring that it trains its AI language models on high-quality, original, human-authored content.
The quality of what you put in determines the quality of what you get out. The paradox of AI is that its ability to mimic human creativity relies on human creativity.
The Real AI Revolution
Humans are more creative than AI is capable of being — it all comes down to complex psychological nuances like trying to hook up with the sexy neanderthal in the cave next door.
But humans are also terrible at some things — I sincerely hope someone is working on an AI-powered plugin to name my layers for me. AI is great at checking details. For example, I ran this post through a grammar and spelling checker before publishing it. In these ways, AI can serve as a valuable tool.
We know from the history of the last couple of decades that when technology is hyped as a design killer, it either slots into our workflows or falls by the wayside. AI is no different: it can’t hold an opinion, it can’t innovate, it can’t experiment.
Instead of focusing on AI and what it will do next, our attention should be directed at ourselves. Because all of those things that AI can’t do, we can.
Once we deduct what AI can do, the designers that remain over the next few years will be opinionated, innovative, and experimental. That’s the real AI revolution.
Ben Moss has designed and coded work for award-winning startups, and global names including IBM, UBS, and the FBI. When he’s not in front of a screen he’s probably out trail-running.