By Maxwell Chalkin, Chief Digital Officer and Vice President

The MKP Thought Leadership working group has had a handful of contentious debates about the role of artificial intelligence in our content creation efforts. The basic question that has emerged is: Is it ethical to use a large language model (LLM) chatbot (like Chat GPT) to help create our thought leadership content? When you unpack this question, there are a multitude of smaller questions that kind of unfold out of it and the answer it turns out is a resounding, “it depends.”

The first question that might be worth addressing is, “Why would we want to use a chatbot like Chat GPT to write thought leadership content?” The answer is simple and perhaps predictable—and, dare I say, representative of many use cases: we’re busy. We do not have a dedicated team of copywriters standing by to develop content for our website or article submissions to industry publications. Our team of content specialists, copywriters and project managers have client deliverables they’re working on day in and day out, and, while we understand the need to produce original think pieces, share our smarts and keep the content on our website and social media feeds fresh, we simply cannot prioritize developing thought leadership content over revenue-generating client work.

And so, since LLM chatbots have become widely accessible, the question has arisen time and time again: if our team doesn’t have a lot of time, why not ask this incredible technology to do some of the work for us? As it turns out, there are several reasons why having Chat GPT write on your behalf can be problematic.

Quality, Originality and Creativity

The way LLM chatbots work is that they are trained on some data set. In the case of Chat GPT, it was essentially given the whole internet as training data (to my understanding, it was largely trained on Wikipedia), meaning that its neural network has been developed on the basis of an astounding amount of information.

As a dataset, the internet is, in a lot of ways, a mile (or several millions of miles) wide and only an inch deep. That is, there is relatively a whole lot more surface-level information on a wide variety of topics than there is deep, rigorous and well-researched information. It is true that Chat GPT has ingested countless books, but so much of the truly substantive writing “out there” is protected by copyright and/or paywall. All in all, Chat GPT’s focus has not been in any single area of expertise. Rather, it has been trained on vast troves of data that may not have been fact-checked. When examining outputs, it can be difficult to determine what is fact and what is hallucination—and the chatbot itself has no built-in way to check.

Our company has built a reputation for being original, creative thinkers and subject-matter experts. These are two areas where Chat GPT leaves much to be desired. When you ask it to be creative, it can often be hokey, awkward or impersonal, much in the same way a non-native speaker can sound when trying too hard to speak naturally in a foreign language. And when it comes to cogently and compellingly writing about our area of expertise, financial services marketing, it can regurgitate platitudes and best practices, but there is a palpable distance between the words that come out and their practicability. After all, the chatbot has no actual experience developing creative or building or executing marketing campaigns. It can only repeat and reformulate what it has ingested. On the other hand, our team boasts decades “in the trenches” solving marketing and communications problems and talking about financial products.

Tipping the Balance

At this stage of the debate in our Thought Leadership meetings, someone might chime in, “We don’t need to use Chat GPT to write our hardest hitting, most prominent and thought-provoking content, but why not use it to build a repertoire of blogs and increase our SEO?” And that may be one of the most compelling and macro-problematic use cases around.

There’s nothing inherently sinister about using Chat GPT to turbocharge content creation. The fact of the matter is, chatbots can create an absurd amount of content in a ridiculously short period of time. You basically just flip a switch and words come pouring out. And so, especially if the priority is quantity over quality, chatbots are a powerful tool.

Are there unintended consequences that will result when lots of people start using chatbots this way? Might there be an artificial content arms race, where more and more of the internet is machine-generated and the quality and originality of content spirals infinitely downwards?

A chatbot might do an excellent job generating keyword-rich content, but is artificial content en masse going to bury quality human-generated content? Might it be increasingly difficult to find high-quality content, or might we the users have to sift through endless machine-generated content to find the real thing?

Does it even matter (especially if the quality of artificial content continues to improve)? What if the next generation of humans encountered artificially-generated content every time it sought information? How does the notion of our children being taught how to think by computers sit with you? What about replacing teachers with computers? We are sitting on the brink of a revolution in artificial intelligence and there are certainly a number of big questions with which we will need to grapple.

What’s Ethical? What’s Plagiarism?

The line can be blurry. One thing is clear: It is unethical to represent artificially generated content as your own.

In recent weeks, a major publisher was caught pawning off AI-generated content as original journalism. To those who caught wind of the story, the revelations were shocking. A reputable and storied publication used fake personas thereby inventing journalists out of thin air to disguise the fact that stories were authored by a chatbot. To return to the point above about “filler content,” the technique was not used for top stories, but rather to populate the website’s less-trafficked corners and to click-bait and optimize. It’s not an uncommon tactic on the web generally, but it is unheard of among well-known publications.

There is no room for debate here: this is a deceitful practice with consequences that reverberate outwards for any publication that deigns to engage in it. Readers are, of course, harmed, but the collateral damage incurred extends beyond the publication at fault to hardworking journalists at the offending publication and generally those in the field who uphold high ethical and quality stardards of reporting. Journalism, an industry already facing an historic assault by market forces, now has to contend with a parallel assault by unscrupulous, profit-seeking fatcats attempting to pass off AI content as news. If not addressed swiftly, public trust in the institution of the press will only continue to erode.

Appropriate Use

In the evolving landscape that is AI content generation, there are a lot of questions around what is acceptable and what is not—and perhaps just as many questions when it comes to predicting and mitigating unintended consequences. But, putting those questions aside for a moment, it may be helpful to enumerate a few use cases that are “in bounds” and worthwhile. I will do so in order from plainly ethical to approaching-the-gray-area:

1. Fun creative exercise: Have a chatbot write a poem or original lyrics set to a song for a colleague’s or a friend’s birthday, for instance. It may not be good, but it will be sure to entertain. Just don’t take credit for the product (not sure why one would, but it bears mentioning).
    
2. Plan a trip: Ask a chatbot to put together an itinerary for an upcoming trip. In conjuction with travel books and blogs—and advice from friends or contacts who have visited your destination prior—a chatbot can be helpful in terms of suggesting destinations, neighborhoods and activities, or at least providing some ideas. Keep in mind that, in all likelihood, the chatbot will not suggest many specific destinations like restaurants (nor should you be considering, in my opinion, restaurant reviews from something that has never eaten). It is also worth mentioning that the chatbot may be working off of outdated data and, therefore, you should always cross-reference recommendations with up-to-date sources.
    
3. Research: As discussed, large language models have, for the most part, been trained on large volumes of data. They can, for certain applications, be used like search engines. Type in your question and out comes information about the subject you’ve asked about. Beware, though, of the junk in/junk out principle… if the chatbot has been trained on inaccurate data, you may get an inaccurate response. Chatbots have also been known to “hallucinate,” presenting something as fact that is spurious. This can happen when the underlying algorithms misunderstand or inappropriately combine data. One additional pitfall to be aware of is that chatbots do not typically cite their sources.
    
4. Summarizing/synthesizing: Maybe you have notes from 10 different meeting participants and you’d like help paring down redundancies or pulling out key facts or action items. Ask a chatbot to perform that task for you. Just make sure to check its work! Again, a word of caution… it may be considered ethically problematic to pass off AI-generated work as your own.
    
5. Social media posts: If you have a topic that you’d like to post about, ask AI to draft post caption ideas. In all likelihood, you are not going to attribute the writing to AI, and that may or may not be problematic for decision-makers (if you’re posting on behalf of a corporate account or brand) and followers. Keep that risk in mind if you are going to use AI to draft post copy. Lastly, be forewarned that, even if you try to train a chatbot how to speak in your personal or corporate brand voice, there may be a distinctly robotic and formulaic quality to these artificially-generated post captions. Therefore, it is probably best to farm the options that are output for ideas and then redraft in an authentic and human-sounding version of your corporate or personal brand voice.

In the spirit of illustrating the power of AI, I “reached out” to ChatGPT to ask it to contribute a conclusion to this blog post. Here is the exchange I had with ChatGPT to make this request:

You

Hi ChatGPT! I'm writing a blog article on the advantages and pitfalls of using LLMs and chatbots like ChatGPT in our business writing. I thought it could be interesting to share the blog with you and ask you to "guest author" the conclusion. How does that sound?

ChatGPT

Hello! I'm glad to hear that you're writing an article on the advantages and pitfalls of using large language models (LLMs) and chatbots in business writing. I'm certainly interested in contributing to the conclusion of your blog. Feel free to share the main points you've covered in your article so far, and any specific themes or perspectives you'd like me to emphasize in the conclusion. I'll do my best to provide a fitting and insightful conclusion for your readers.

You

Thanks very much. I will post the blog in its entirety here shortly.

ChatGPT

Certainly! I'll be here when you're ready to share the blog. Feel free to provide any context or specific points you'd like me to consider while crafting the conclusion. I'm looking forward to contributing to your article!

You

I would ask that you please simply read the article and provide an original conclusion in the form of 2-to-3 paragraphs.

ChatGPT

Of course! I'll keep an eye out for your blog post, and once you've shared it, I'll provide an original conclusion for your article in the form of 2-to-3 paragraphs. Take your time, and whenever you're ready, feel free to share the content.

You

Thanks!

ChatGPT

You're welcome! Whenever you're ready, go ahead and share the blog post, and I'll be happy to provide the conclusion.

Here is the conclusion that it provided:

In the ever-evolving landscape of artificial intelligence (AI) and content creation, the ethical considerations surrounding the use of large language models (LLMs) and chatbots like ChatGPT are multifaceted. The question of whether it is ethical to employ such technologies in thought leadership content creation reveals a complex interplay of practicality, quality, and the potential consequences of widespread adoption.

While the allure of leveraging LLMs to ease the burden of content creation is evident, the pitfalls come into focus when assessing aspects like quality, originality, and creativity. Large language models, despite their vast training datasets, often struggle to match the depth and practicality of content crafted by human experts. The nuanced understanding and hands-on experience that humans bring to areas of expertise, such as financial services marketing, remain challenging for AI to replicate convincingly.

The debate extends to the macro-level implications of using chatbots to amass large quantities of content quickly. As organizations seek to boost their online presence and SEO through automated content generation, concerns arise about a potential decline in content quality and the dominance of machine-generated content on the internet. Unintended consequences, such as an artificial content arms race and challenges in discerning high-quality content, loom large on the horizon.

In navigating this ethical terrain, a key consideration is appropriate use. From fun creative exercises to aiding in research and summarization tasks, there are instances where AI can be a valuable tool. However, a clear delineation must be maintained, emphasizing that AI-generated content should not be misrepresented as human-authored. As we stand at the crossroads of this AI revolution, grappling with questions of ethics, plagiarism, and the future of information dissemination, it becomes imperative to tread carefully and thoughtfully integrate AI into our content creation endeavors.

P.S. In reading the chatbot’s submission, I noticed that it provided one more paragraph than the maximum number I requested. I couldn’t help but point that out to GPT and so ensued the following exchange:

You

Thank you for providing this cogent and topical conclusion to my blog post. I noticed you provided four paragraphs instead of the 2-to-3 I requested. You really are an overachiever :)

ChatGPT

You're very welcome! I'm glad you found the conclusion helpful. Apologies for the extra paragraph—I might have gotten carried away with the complexity of the topic. Feel free to trim it down as needed to fit your desired length. If you have any further adjustments or specific requests, please let me know, and I'd be happy to refine it for you. Thanks for the opportunity to contribute to your blog post!

MKP communications inc. is a New-York based marketing communications agency specializing in merger/change communications for the financial services industry.