If you’ve read my earlier articles on AI, copywriting, and automation, you probably know that the future of copywriting has arrived.
And it’s moving pretty fast.
But will robots take your job?
Rather than taking experts’ opinions at face value — many who carry contradictory opinions — I prefer to do my own research and form my own opinion.
Below, we’ll deepen some of the themes I explored two weeks ago.
Namely, we’ll:
- Recap why it’s so difficult to predict the future of copywriting
- Conceptually explain AI, machine learning, and deep learning
- See what some AI experts have to say about deep learning and AI
Hopefully, this will help us form our own conclusions about AI and copywriting.
Let’s get started.
The Future of Copywriting Depends on AI
Trying to understand AI’s potential isn’t easy, for a few reasons:
- Most non-specialists don’t understand how AI works, which makes it impossible to know its possibilities and limitations
- Technology specialists themselves can give in to dystopianism or utopianism — both of which use logic to rationalize emotional conclusions (sound familiar, copywriters?)
- Media outlets and reputable research firms fuel the fire with unsubstantiated sensationalism and fear-mongering
So what should you do when no one knows what they’re talking about?
Evaluate the situation yourself.
Do your own research and form your own opinion.
Critical thinking is already in short supply…
Let’s not shorten it further with groupthink.
AI May Be Deep, But It’s Not Smart
Today, what we call AI mainly consists of two computational approaches, machine learning and deep learning.
Let’s look at each one from a bird’s eye perspective:
Machine Learning in 30 Seconds
Machine learning uses algorithms to parse, create models from, and make predictions about data.
It can be used to:
- Recognize patterns — such as faces
- Make predictions based on those patterns — think of Amazon’s recommended item carousels
- Offer insight into data — for example, identifying investment opportunities in financial data
Machine learning can save and make serious money.
Programs that use machine learning can:
- Make product recommendations more relevant, increasing average shopping cart sizes
- Predict when customers are most likely to open an email, increasing open rates
- Recognize patterns in medical data, increasing the accuracy of diagnoses
Machine learning itself uses a few main techniques, such as supervised learning, unsupervised learning, and reinforcement learning.
These techniques use data science algorithms, such as regression or clustering.
A simplified example:
- You have a spreadsheet of your company’s employee data, with two columns: employee salary and years of employment
- This data is fed into a simple linear regression algorithm
- It finds the simple average slope between length of employment and salary
This relationship, a ratio, represents the relationship between salary and experience.
Using a linear regression algorithm, it would come out as a straight line.
At each point on the line, you would have an average salary and experience level.
The result could be used to provide salary offers to new hires, in line with your company’s existing payment standards.
Deep Learning in 45 Seconds
Deep learning is a subset of machine learning.
It uses “deep” layers of artificial neural networks (ANNs) and requires more data.
Artificial neural networks work like this:
- A neural network is composed of layers
- Each layer is composed of neurons
- Each neuron has its own weight — and each uses an equation to process the input — then passes that result to an appropriate neuron at the next layer
- This process is repeated many, many, many times, until we receive the output
Most people say that machine learning and deep learning software both “learn.” And from a certain perspective, it makes sense.
I’m a bit wary of this word though.
“Learning” has pre-existing associations such as logic, reasoning, and critical thinking.
Using this word can inadvertently personify deep learning, making people think it can do things it can’t.
This is how a neural network “learns” from training data:
- Data is passed into the network
- Each neuron, which has its own weight, transforms and passes results to the next layer
- The final output value is compared to the actual value
- This result is passed back into the network and neuron weights are adjusted
- Rinse and repeat
For instance, think of the training set like a test.
If you pass in 1,000,000 cats, and it got 900,000 correct, it would go back, self-adjust the neuron weights, then try again.
Thousands and millions of times, adjusting and improving each time.
Now imagine how copywriting data could be processed the same way — headlines, etc.
Because we’re talking about algorithms here, not humans, I prefer to use technical terms instead of terms that personify the program.
Self-adjusting, self-adaptive, responsive.
Something like that.
Anyways, there is a point to all of this…
The Point
The point is that AI techniques aren’t intelligent, mysterious, or beyond conceptual understanding.
When we understand what they are and how they work:
- We won’t personify or pedestalize them
- We’ll can evaluate them logically instead of emotionally
- We’ll have a better grasp of their possibilities and limitations
And they do have limitations.
In fact, according to deep learning expert Francois Chollet, many more applications are “completely out of reach for current deep learning techniques.”
The problem he says, is that “anything that requires reasoning—like programming, or applying the scientific method—long-term planning, and algorithmic-like data manipulation, is out of reach for deep learning models.”
Chollet inadvertently points out the human trait that is causing much of the hype around this debate:
One very real risk with contemporary AI is that of misinterpreting what deep learning models do, and overestimating their abilities. A fundamental feature of the human mind is our “theory of mind”, our tendency to project intentions, beliefs and knowledge on the things around us.
This is something that should be kept in mind whenever we attempt to predict the future … or listen to others who do the same.
Deep Learning Needs Deep Data
AI needs lots and lots of data, as mentioned.
Deep learning, in particular, needs big data to shove through its big ANNs.
This “data barrier” could prove problematic for all but the largest businesses.
Fabio Cardenas, CEO of Sundown AI, says:
It’s insane to think that a company, unless you’re an enterprise, is going to have millions of records for you to actually put into your algorithms to generate results … Even when a company has 10,000 emails a day, it may not be enough for deep learning.
Deep learning works best, he says, when there’s no goal in mind.
Sundown AI’s product, Chloe, is a question-and-answer platform that uses AI techniques, including machine learning. But it requires much less data than platforms based on deep learning, such as IBM’s Watson.
And implementations cost about 10% as much as deep learning products.
So for copywriting, it seems, the real threat doesn’t come from deep learning, but from hybrid approaches.
Conclusion: What AI Means for the Future of Copywriting
Kai-Fu Lee has said that deep learning won’t be surpassed for a while. And that he feels artificial general intelligence is not on the horizon.
So don’t expect robo-bosses to be walking among our cubicles any time soon.
Or copywriting software to start spitting out projects that require logic, reasoning, and thinking — like white papers and technical reports.
I expect the real disruption to come from hybrid automation solutions. Ones that use RPA, machine learning, and old-school programming.
Sundown AI uses low-cost AI methods, without the need for extensive data.
And their platform offers a great product at a great price.
If copywriting automation software can accomplish the same goals:
- Work around the data problem
- Outperform human writers
- Charge less
We can expect automated copy to start eating away at market share.
However, as I’ve mentioned before, I’d be surprised if automation pushes the copywriting industry back below its pre-internet levels.