The AI Mirage of Perfect Grammar, But Imperfect Communication
DeepSeek was downloaded 16 million times in its first 18 days. ChatGPT reached 100 million users in just 60 days - a feat that took Facebook four-and-a-half years. As Ray Kurzweil argued twenty years ago in "The Singularity is Near," the pace of technological change is exponential. Today, we're on the steep slope of the growth hockey stick.
Our human capacity to learn and adopt new ways of working is the sand slowing the gears of change. This is not always a bad thing. It is important to take the time to ask fundamental questions about how we use AI in communications and PR, and for something as fundamental as writing.
AI-generated writing is a subset of content generation, which is a subset of AI-use cases built on the backbone of Large Language Models (LLMs) and Large Modal Models (LMMs). Andrew Ng, one of the industry’s leading lights, suggests we need to think of AI as electricity. Tools and agents are essentially appliances with specific purposes, that can be applied to almost every human endeavor.
Part of the challenge of using generative artificial intelligence tools is the “wow!” factor it inspires. This can be so dazzling it's hard to see through perfectly written sentences to the imperfect communication underneath. Just like a mirage on a desert highway, what we see at a distance is deceptive.
Similar to the idea of the "uncanny valley," where the closer a robot gets to being human-like, the more the feeling of weirdness grows, there's something slightly off with AI-generated content. Although it's often hard to put a finger on what exactly that is, if we take a closer look, text may not only feel wrong, it often is wrong, with pieces (some invented) put together like a jigsaw puzzle that doesn’t quite fit. Recent BBC research underscores the challenge, noting that four major artificial intelligence (AI) chatbots are inaccurately summarizing news stories and getting facts wrong 19% of the time.
For our industry, this underscores the importance of human intervention. There is no doubt that current tools are getting better. It is also clear that we also need to get better at using these tools if we are to generate better writing. Plugging in a one-line prompt to free ChatGPT won't generate the best results.
We need to look for the best and most focused model for a specific purpose and learn to create the types of highly-structured prompts that generate better results. And, we need to upload examples of our own content to help the AI figure out our style and cadence. But even with these best practices, don’t expect miracles.
Today, the bottom line is human writing and editing skills, especially when reviewing AI-generated content, remain key to producing materials that will drive engagement rather than stymie it. Underscoring this point, at a technology venture capital event I recently attended an investor shared that he could smell AI-generated content that wasn't authentic to the sender – and told the group it was an instant turn-off. Journalists are no different.
This isn't just perception. A recent Stanford study found that AI-generated business pitches were 23% less likely to receive follow-up interest compared to human-written ones, despite being technically perfect. Again, don’t expect a different result with journalists.
It's important to be fearless and do a deep dive into AI. But this does not mean an unquestioning adoption of tools or use of AI-generated writing to simply generate more content.
Writers who aren't AI natives, and who find the tools to write emails or re-write social media posts annoying, may be tempted to stop their AI journeys. Looking at these tools in one dimension as writers, risks missing other aspects of their value in which they far exceed human capabilities. Providing access to the entire repository of world knowledge to help research ideas, source references, and generating drafts is a powerful starting point for the human work of communicating. But even here – these tools are far from perfect.
The AI mirage is real. Having the capacity to see through it, and the skillsets to re-write, edit and build on content in ways that make it authentic and effective, are learned. At a time when it may take seconds to generate an article, email or social post using AI, the slow skills of looking closely at content, editing, re-writing and fact checking have never been more important. Realizing the potential of AI is not about delegating writing or editing - but using it to augment our capacity to write better.
