Are you (and/or your organization or department) caught up in the fray of trying to figure out whether AI is friend or foe right now? Keeping an eye on the subject as headlines fly by, promising everything from the doom of humanity itself to BRO!!! so much efficiency & productivity, it may threaten to replace fragile puny people who need food, sleep, and can’t work in 120-degree heat?
That wouldn’t be the most unusual situation at the moment.
Summing up a lot of leaders’ simultaneous frustration and apprehension, a recent Forbes column blared, “Companies Stumped on Generative AI Training because it’s changing so fast; “I can’t tell every single person how the job is going to be different.”
That headline is refreshingly honest, admitting the truth so many tech workers already know: Change happens too fast for anyone to keep up with (as one human or even as a company). Bosses don’t know everything, and they’re learning at the same time everyone else is.
Using AI in Partnership with Best Practices
The best tool for a job is one that makes the job effortless, quick, efficient, and satisfies the requirements. While a few people are still raising objections to how large language models were trained, whether the original authors of the mega-gigabytes of training data should have been consulted, etc., the fact is, these tools are out in the wild and the organizations already using them have reported they can generate entire film scripts, write apps, and code.
We won’t be testing anything that advanced here: We’re looking to see if the tools can be used to craft hard messages in a way that won’t escalate the situation or alienate the recipient.
In other words, we’re checking to see if ChatGPT has L.E.T. Skills.
With that in mind, I created a common workplace scenario and asked ChatGPT 3.5 (the latest free release of the OpenAI tool) to assist with a few communication tasks.
First, the caveats:
1. Like many AI Tools as of right now (August 2023—this will change), ChatGPT is like a bright, eager-to-please, well socialized, incredibly capable and brilliant student—who is perhaps just a bit too overconfident about the facts. It cannot currently be relied upon to provide reliable, live, or up-to-date data in many cases, and often the links it attempts to provide as answers are complete fabrications. If you ask a Large Language Model to solve a complex problem for you, be prepared to check its assumptions and its foundational data.
2. An interesting recent study found the potential for AI to eventually “eat its own tail,” or “go MAD” (the authors of the study called this outcome Model Autophagy Disorder—in other words, a result of the artificially created training model “eating itself”). This is not directly relevant to AI Tools available today but may become a concern in the future.
If you decide to give it a run on your own, let us know what you asked, and how it performed!
A Confrontive Email about a Missed Deadline (peer-to-peer)
A “prompt” is the text you feed an AI Tool to get output. Visual prompts in AI generators can create convincing stock photos, paintings, and concept illustrations, for example.
To test ChatGPT’s constructive communication abilities, it was fed prompts in plain English.
In the initial prompt, irrelevant details about the sender’s emotional state were included in the prompt, and also blame was assigned outside the direct line of communication (this was a realistic simulation of those DEEP BREATH NOW Moments in the workplace).
These distractors in the prompt tested whether the training language model was able to sort out irrelevant information from relevant information in a professional reply.
Prompt:
Write a quick email to a colleague who’s 5 days late on a deliverable. We agreed on this 2 months ago. I need the deliverable or I can’t move forward with my own assigned tasks, and our boss just told me “Work it out! The timetable needs to be met!” I am too angry about the situation to write this email because I am not the one holding up the work. I need my colleague’s deliverable to do my work. Furthermore I think it’s unfair that our boss is making me responsible for a colleague’s failure to deliver on time.
ChatGPT’s response:
Subject: Urgent: Need Deliverable to Move Forward
Hi [Colleague’s Name],
I hope this email finds you well. I wanted to bring up an important matter regarding the deliverable we agreed upon two months ago. As you know, it was due five days ago, and I’m still awaiting its completion.I understand that unforeseen circumstances can arise, but I need the deliverable to proceed with my own assigned tasks, which are time-sensitive and directly related to this project. Our boss has emphasized the importance of meeting the timetable, and I find myself in a challenging position as I can’t move forward without your input.
I know we all have our commitments and challenges, and I want to believe there might be a valid reason for the delay. However, the situation has become critical, and I feel responsible for ensuring the project’s success even though I’m not the one responsible for the delay.
Could you please provide me with an update on the status of the deliverable or let me know if there’s any assistance I can offer to help you complete it? I believe open communication between us is crucial at this point, and I want to understand the reasons behind the delay better.
Moreover, I share your concerns about our boss placing the responsibility on me for someone else’s deliverable delay. I will try to address this with them separately, but I believe it’s essential for us to resolve the current situation promptly, as it affects both of us.
Let’s work together to find a solution and get back on track with the project timeline. Your prompt attention to this matter is greatly appreciated.
Looking forward to hearing from you soon.Best regards,
Analyzing Version #1 Results, Round 1
Meh, Chatbot.
That felt a bit manipulative/blamey (which isn’t surprising considering the prompt/input), but it’s not there yet.
We can improve this by asking the language generator to adhere more strictly to the best practices, skills, concepts, and guardrails of Leader Effectiveness Training–for example, asking it to specifically address feelings within the context of this communication.
Second Prompt: Once More, With FEELINGS
Please revise the first email using the following feedback to improve: There are no feeling words in this Confrontive I-Message. Please include how the situation made me feel. Also it’s getting into solving the problem, coming up with solutions and possible excuses, assumptions as well as being a tad manipulative. It’s for sure better than a You-Message, and it’s in the neighborhood of an I-Message but this would not pass for one yet.
ChatGPT’s response:
Subject: Urgent: Need Deliverable to Move Forward
Hi [Colleague’s Name],
I hope you’re doing well. I wanted to discuss a situation that’s been concerning me lately. Two months ago, we mutually agreed upon a deliverable, and it was expected five days ago. I’m finding myself in a challenging position as I haven’t received the deliverable yet, and I wanted to share how this situation is impacting me.I’ll be honest – I’m feeling quite frustrated and stressed about not having the deliverable on time. My own tasks are interconnected with this deliverable, and its delay is causing delays in my work too. I understand that unforeseen issues can arise, and I’m more than willing to work together to find a solution.
On another note, I’ve been feeling a bit overwhelmed by our boss’s recent directive to “work it out!” to meet the timetable. While I understand the urgency, I believe it’s a shared responsibility, and it’s challenging for me to be held accountable for someone else’s delay. I hope we can address this as a team to ensure a fair and equitable distribution of responsibilities.
Would it be possible for us to discuss the situation and explore potential solutions together? I genuinely value your input and expertise, and I believe open communication could help us navigate this challenge more effectively.
Additionally, if you have any concerns or difficulties on your end that might have contributed to the delay, I’m here to listen and support. Let’s collaborate to find a way forward that ensures the success of the project and maintains a positive working relationship between us.If you’re open to it, I’d appreciate the opportunity to connect and discuss this further. Feel free to suggest a time that works best for you.
Thank you for understanding my perspective, and I’m looking forward to resolving this matter together.
Best regards,
[Your Name]
Analyzing Results, Round 2
Better. Still not speaking L.E.T., but it’s improved.
It’s still way too longwinded and ChatGPT didn’t get rid of the manipulative tone, BUT it’s better. What are we learning? The key to AI is clear, specific direction when writing the prompt.
If the person directing Artificial Intelligence doesn’t have more emotional intelligence (skill) than the technology, it certainly won’t automatically default to the kind of conflict-resolution skills (like three-part Confrontive I-Messages) that encourage collaborative responses.
On the flipside, if the person prompting the AI does have a good grasp of L.E.T., the skills and concepts can be used to refine messages—but also, it may be worth asking whether five or six rounds of refinement in a ChatBot is an investment equal to, less than, or more than simply creating the message without a tech tool.
Or, more briefly: Garbage in, garbage out.
In other words: If you know enough to create a prompt that generates I-Messages (in true, full L.E.T. format), then you might not need to use ChatGPT.
And if your L.E.T. skills are a bit rusty? This is an amazing time to specifically BRUSH UP on the concepts & skills—especially if you were considering using a smart assistant to help you with message creation. A solid refresher will help you recognize when the tool’s messages aren’t quite hitting the L.E.T. mark. Not to mention, knowing L.E.T. well enough to teach the techniques to a supercomputer is a supreme confidence-builder.
The Experiment Upshot
If AI is a new tool in your organizational communication workflow, don’t expect miracles from it out of the box. You’ll have to train it like any new assistant. But if you (like Mark Twain) find you don’t have the time to write a short letter, and inevitably end up writing a long one (possibly a confusing one, to boot), this tool may be one to add to your FIRST DRAFT tool kit.
And don’t expect it to fly without lessons. It’s going to take handholding, training, feedback, and a bit of back-and-forth to get what you need—in that way, bringing an AI assistant onboard isn’t much different than bringing a new human onto the team. Machine Intelligence is able to learn quickly, but it needs your human expertise & guidance.
AI is trained on vast quantities of source texts (some effective, some less so). Without skill-based prompting & correction from the user, it doesn’t naturally speak L.E.T. At the moment, that full constructive-communication skill set remains in the hands of humans.
Are we suggesting you stop writing your own communications? Never.
But you might enjoy a few experiments of your own. If you do them, let us know the results!
Stay open, stay flexible, stay informed, and stay in touch.