The machines are taking over! AI is going to replace us all!
If I only had a dollar for every time I heard some version of that. Even in its early stages, artificial intelligence, or AI, feels like an imminent threat. After all, intelligence is what has truly set human beings apart from the rest of the animal kingdom. The ability to solve unique, complicated problems is our niche.
What makes the AI panic feel even weirder is that it tends to freak out really smart people like Elon Musk and the late Stephen Hawking. Rationalists like Eliezer Yudkowsky warn that AI is a global risk, as do science communicators like Josh Clark from Stuff You Should Know.
Political figures on both sides are afraid that advanced AI could destroy jobs, which is maybe the only thing Democrats and Republicans can still agree on. Others still, myself included, are concerned about how automated AI-driven algorithms feed people increasingly extreme and misinformed content. Still others are giddy at the chance to play with new software, waiting for the next internet-level revolution in technology.
But all these fears and dreams are abstract, global, and technocratic. On Weird Marketing Tales, we’re interested in edge cases and the unique oddities of each individual person and company. And one thing I’m bothered by when it comes to all this AI doomer talk is the complete lack of any practical tips on what to do about it!
In other words, what the hell does AI mean for your business, really?
All we know for sure right now is that AI is becoming very sophisticated. Much like social media in the late 2000s, it’s on the brink of dramatically changing our media landscape. It logically follows that companies will have to adapt yet again to new rules.
In this article, I’m going to lay out some practical tips on how you can carve out a uniquely human niche in an ever-more automated world.
But first, let’s talk about…
What AI can do right now
If you want to see the most impactful way AI is used right now, look at Facebook, Instagram, Twitter, YouTube, and TikTok. Any social network you enjoy using is unceremoniously running on AI right now. The algorithm is designed to keep you scrolling and clicking so that you can see ads and generate revenue for the company. That’s why social media is freakishly good at adjusting what you see around your tastes, for better or worse.
If you want more information on how social media works, check out the back half of my post about Gawker Media.
But in addition to that, I want to talk about GPT-3/GPT-J and Midjourney, both tools with which I’ve become absolutely enamored. These tools could very well be the future of content creation on the internet. I want to talk about them before they go completely mainstream.
GPT-3, and its open source cousin, GPT-J, are capable of creating long stretches of text based on any prompt you provide. Midjourney focuses on creating great works of art based on the instructions you provide. I’ll show you what that looks like.
GPT-3 and GPT-J can generate long stretches of coherent text faster than any human could ever manage
Born in the heady days of June 2020, Generative Pre-trained Transformer 3 (GPT-3) is a “language prediction model.” If you give it a prompt, it can create long stretches of text for you like autocorrect on steroids. GPT-3 itself is not widely available to the public right now because it has tremendous power to be used for both good and evil, hence the existence of an open-source alternative, GPT-J.
You can find a simple version of GPT-J online via EleutherAI. You can type in a short prompt and receive a longer stretch of text as a response. For me, I typed in “I think marketing is” and received the following as a response:
I think marketing is more like magic than science. It’s an art and a science. It’s like a superpower.
Marketing has always been my dream job. I’ve always wanted to be the person that does the marketing for the cool companies. I wanted to do the marketing for these big companies that are way bigger than the average person has ever heard of. I wanted to be the person that creates marketing campaigns for these companies. I want to be the person that makes the customers feel special, and I want to be the person that makes the customers want to come back. I want to be the person that is the CEO of marketing.Written by the GPT-J-6B language prediction model
Bear in mind that no human being wrote that, and yet it sounds like it could be a short section of the transcript from a podcast. I ran it again two times and within forty-five seconds, I received:
I think marketing is a team sport. The thing I find most challenging is the constant changing of what people want and what people need. I think it’s a skill that most marketers learn, but I’m always learning.
Given about an hour, I could coax GPT-J into giving me a truly great stretch of text to share. But I’ve given you two of the first three excerpts I received as a representative sample. If you want to see text generation at its best, watch the short video below (it’s good even if you have to mute it).
GPT-3/GPT-J are capable of generating ideas, writing the news, creating how-to guides, and even making recipes
Now EleutherAI is pretty great, but it doesn’t give you very long stretches of text. For that, you need a little bit of technical savvy and the ability to use Google Colaboratory. You must run an instance of GPT-J by following the instructions to the letter and be ready to troubleshoot.
However, if you are able to get it to work, you can ask to generate blog ideas, as I have:
1. How The Simpsons Made $1 Billion and Then Lost It All
2. Why The World of Video Games & Esports Can Help You Get Rich
3. This is How I Learned to Stop Worrying And Love SEO
4. Why Apple Is the Ultimate Brand Expansion
5. What Really Causes That Sharp Pain When You Touch Your EyebrowSuggestions are from GPT-J, with links to the posts whose titles I think inspired these fake headlines.
Of course, I had to narrow down from a list of 70 to get these five ideas, and that last one isn’t really a fit!
You can also have it create entire fake news articles, how-to guides, and transcripts for episodes of TV shows that never happened. I even managed to get it to create a recipe, complete with the long preamble that bloggers affix to their recipes in order to rank better in Google.
Though Sylvia had made this cake for about 40 years, she had not eaten it since she was about 30. What she would eat was her husband’s, as he had always been the one to eat it.
“He really likes the fluffy texture of the cake,” she said. “It was good stuff.”
3 tablespoons butter
1 cup sugar
1 cup brown sugar
1 cup milk
2 teaspoons vanilla
1 cup shortening
4 cups cake flour
2 teaspoons baking soda
2 teaspoons cinnamon
2 teaspoons salt
2 teaspoons nutmeg
1 cup dark molasses
I’ll touch on the implications of this technology in the section after the next, but first, I want to talk about something else that another widely available AI is capable of doing right now.
Midjourney can rapidly generate artwork based on any prompt you give it
As it turns out, text is not the only medium which can be created by training a computer on existing data and defining specific paramaters. In fact, in 2021, OpenAI released DALL-E, a computer model which has the ability to create digital images from natural language descriptions. They have since released DALL-E 2, which further improves on the ability of the first iteration.
Now I don’t personally have access to DALL-E 2, but I do have access to Midjourney, which is in public beta and is very, very close in quality. You can access the free beta by going to Midjourney’s Discord server and typing in /imagine [your prompt]. Below, you can see six images created by the Midjourney AI based on prompts I provided, which include:
- A rainbow spaceship fighting in a war
- Blade Runner Walmart
- Abraham Lincoln taking a selfie
- Calico cat smoking a cigarette after a long day at the office
- Van Gogh paints the Atlantic City Boardwalk
- Ninja Turtles having an existential crisis
Notice the extraordinary detail on each of them! (Click any image to enlarge.)
Each of these images, it should be noted, took about 90 seconds to draft and another 90 seconds each to upscale so that I can post them in glorious high-resolution on this blog. By comparison, the artists I have worked with on board game projects would need at least four hours to create any one of these images.
But, of course, just like with GPT-J, Midjourney’s art is just a little…off. It has a tough time getting faces right and sometimes it just doesn’t understand the prompts you provide. Like with my examples above, I went through a lot of prompts to get results worth sharing with you on this blog.
So what’s with the crapshoot odds? Well, it has a lot to do with how AI itself works.
How AI works and how it might be used – for good and for evil – in the future
Now that I’ve shown you some concrete applications of AI and how it can be used today, I think it is now helpful to talk how exactly AI – specifically content generating AI – works.
Take GPT-3 for example. It runs on a neural network, which is a type of computing system that takes notes from animal brains. These networks are made up of nodes which are connected by “neurons.” Neurons are given different weights based on whether or not the network does what its creator wants it to do.
GPT-3 has over 175 billion parameters, all of which are used to shape the text created by the AI in some way or form. Through the process of machine learning, people “train” the system by running prompts through it and telling the system whether the output – as a whole – is good or bad. The network is run over and over again, incrementally changing the weights on its parameters so the results become more consistently good.
I’m oversimplifying a lot, but suffice it to say that this training can be done with or without a person in the loop after a certain point, allowing the AI to “learn” on its own.
If that still feels a bit fuzzy, I’ll use an example from social media instead. Every time you watch a YouTube video for a certain amount of time, the system takes note. It shows you videos that have similar properties to what you spend more time watching in the hope that you will keep watching more videos. Over a considerable amount of time and sheer trial and error, it will eventually adjust your YouTube recommendations to perfectly suit your tastes. The same principle applies to Facebook, TikTok, and all the rest.
Good and evil uses of content-creating AI
Having computers that can create AI is honestly amazing. For advertisers, computers can come up with serviceable copywriting options in a matter of seconds, giving the advertiser a smorgasbord of options to choose from. This could be a massive time-saver in the long run, and AI may very well end up being better than human copywriters soon.
Similarly, once AI gets to a certain point, I may not have to write explanations of how machine learning works. I can have the AI write sections that I don’t feel confident about, allowing me to focus on things I do well.
Coming up with images for content on the internet right now requires either graphic design skills or access to a stock photo library. This may change as AI software such as Midjourney becomes more accessible and inexpensive, allowing people to focus on written or audio content instead of finding visuals to compete with the demands of search engines or social media networks.
In short, AI could end up saving many marketers and business owners time, effort, and money by eliminating tasks they don’t want to do, just like how spell check has made it to where we don’t have to spend quite as much time proofreading.
Of course, you can likely see how AI can be used to do harm too. It could potentially destroy jobs, putting many writers and artists out of business. On top of that, AI has a bad habit of reflecting the worst qualities of humanity too, such as our racial biases. Don’t believe me? Talk politics with Cleverbot. That AI, like any other, is trained on all kinds of human speech.
What AI can’t do right now
Discussing how AI works and its potential applications, advantages, and drawbacks is useful, but it’s important not to freak out. The world is changing fast, but it’s also changing painfully slowly at times. The worst pandemic in 100 years did not destroy the office. Internet access and social media have not destroyed television. Self-checkout has not replaced cashiers. Spell check has not replaced proofreaders.
Could we eventually get to a Star Trek universe where money doesn’t exist, clean energy is abundant, and replicators make all our stuff? We might!
But is that going to happen tomorrow? Hell no!
Ultimately, the kind of AI used by GPT-3/GPT-J and Midjourney is pass/fail. That means that it can sometimes make wonderful text or art for the wrong reason. Likewise, these AIs can respectively follow all the rules of grammar or composition and create work that seems soulless or alien.
I’m not talking about GPT-3’s tendency to talk in circles or Midjourney’s 50/50 chance of totally misunderstanding the prompt. Those are bugs, and they will be worked out.
What I mean here is that GPT-3/GPT-J and Midjourney do not understand the process by which text or art is made. It only understands how to generate specific outcomes. GPT-3 can write a book, but it doesn’t understand the Hero’s Journey. Midjourney can create beautiful images, but it can’t quite nail the subtle emotions of the human face. And I don’t think these things are going to be patched in a later update. I think AI will forever seem alien and uncanny in some ways, though it will be increasingly competent in many others.
Or to put it in a different way for philosophy buffs, the computers don’t understand qualia. That is, they don’t understand subjective, conscious experiences. They do not know how beautiful the sunset can be over the ocean, nor how it feels to be in love, though GPT-3 can write poetry to describe it.
I think this is most obvious when you have AI imitate music. Just look at how OpenAI Jukebox just doesn’t quite understand what to do when trying to continue a Queen song part of the way through its completion.
Believe it or not, I’m going somewhere with this. Your unique human experience is a prerequisite for doing the…
7 things you can do to prepare your business for a more AI-friendly future
The key thing here is to remember the all-important marketing lesson. Successful businesses have niches, and this is going to be true even as the world is flooded with ever more powerful AI tools. Humanity has a niche, believe it or not. Oddly enough, it’s kind of a circular niche: humans are the best at knowing what other humans want.
1. Lean into what makes you unique. Don’t get bent out of shape about the AI.
Let’s say that when GPT-4 comes out, it’s really consistent about using proper grammar, syntax, and structure. That would make it, in many ways, a more competent writer than most of us.
But I guarantee you that GPT-4 will not be able to decide what is worth writing about. It won’t know how to read its work when it’s done to make sure it’s useful to its reader. Some of its points will be made in ways that don’t emotionally connect.
Human beings are still the best at being able to tell what other human beings want, and it’s going to stay this way for a while. Let the AI do grunt work at the speed of light while you can focus on putting that work to wise use.
2. Think about the next 5 years, but don’t worry far beyond that.
Technology develops absurdly fast, and trying to predict it ahead of time is a fool’s errand. In 1997, nobody would have believed that they would have 1,000 songs in their pocket in five years. In 2002, the idea of a well-made touch-screen smartphone was a fantasy. In 2011, nobody would have thought Russia would use advertising on Facebook to try to swing our elections. In 2017, no one could have predicted that AI would be able to paint Costco in the style of Monet.
Focus on the emerging trends that are going to be relevant in the intermediate future, and trust yourself to figure out novel problems as they arise in the long-term future. To do otherwise is to subject yourself to constant anxiety about bad things that may never happen and to blind yourself to emergent problems while they’re still easy to fix.
3. Remember that process matters, not just outcome.
AI can only consider outcomes when it’s creating something. It doesn’t understand the whole process. Much of the magic of creating a great novel, an award-winning ad campaign, a product people love, or music people listen to comes from the iteration. It comes from long days and late nights, months and years of rewriting, inspiration striking while you’re in the shower, and people giving you feedback you don’t want to hear.
AI can somewhat make up for this by creating a massive quantity of work, but you still have to sift through it to know what’s truly good. This takes time and taste – and taste isn’t something we’re going to have an AI for by 2027.
4. Remember your audiences emotions and motivations.
People are not rational, and this is OK. Our purchasing decisions are driven by our emotions and by various motivations which differ from person to person.
Even as fancy AI tools become more common, take the time to really understand who your target audience is and what they want. When you can see your customers as individuals and not abstract concepts, you will be able to create the products and services they want.
5. Don’t be afraid to use AI to create content, but always keep a human in the loop.
To stay competitive in this big global marketplace, you have to be ready to use new and more efficient tools when they become available. For many people, that may include using AI. Being a fan of the written word, it may very well make sense for me to use Midjourney 2 or 3 to create art to help reinforce my points someday.
But for now and for the foreseeable future, AI has a tendency to create things that seem uncanny and soulless. This is because the AI doesn’t understand qualia – it just knows how to string words and colored pixels together to make something that people generally like or generally dislike. Nothing more, nothing less. As long as you keep a human in the loop, you have a chance to override more questionable decisions made by AI.
6. Embrace AI tools for small parts of your work first.
It might be tempting to be the first person to use AI heavily in your line of work. I think this would be a mistake. We don’t have best practices for using AI tools first, so adopt a spirit of experimental marketing and start slow.
For example, I’m playing around with GPT-J to generate blog ideas. Most of them are not great, but sometimes, seeing 100 crazy ideas in a row gives me one really good one. And after all, one really good post on this blog can bring in hundreds or thousands of readers every week for years, which results in more consult bookings. Forget having the AI write for me – having an inspiration machine at my disposal is already an improvement!
7. Compete on quality and relationships, not speed.
For the last 20 years, competing on speed and price has been a foolish thing to do. Someone overseas or some giant corporation can always do what you do faster and cheaper. It is a plain reality.
AI is going to do knowledge work faster and cheaper than any human on any continent can. But let’s not forget – when the factories were invented in the Industrial Revolution, people started working in them instead of on farms. When those factories become automated, many people moved on to offices or skilled trades. Pretty soon, some of those knowledge jobs are going to be phased out too.
But new jobs will come up, as history has shown us. Corporations will always need to compete with one another and they will have access to the same cheap tools (as will small business). These corporations will need human workers to keep their competitive advantage.
When everybody is making knowledge products cheaply and quickly, then cost and speed aren’t the differentiators anymore. Kindness, craftsmanship, the ability to listen, and individual relationships will be.
Mediocre companies will mill out automated content and cookie-cutter help. They’ll turn customer support into a sea of chatbots and have their copy written by GPT-5 with no supervision.
Smart companies will know your first name, shake your hand, and ask you about your hopes and dreams. They’ll use AI to do the grunt work so they can spend their time focusing on what matters: living, breathing people.