Confluence for 8.25.24
The McNamara Fallacy and AI. Midjourney keeps getting better -- and more accessible. Book recommendation: The Skill Code. How one tech writer moved past AI resistance.
Welcome to Confluence. It’s hard to believe, but we’ve now been publishing Confluence for a full year. It’s been a fantastic experience for us to share what we’re learning about AI, and we’re looking forward to keeping it going. We’ll have a specific post reflecting on what we’ve taken away from the past year soon, but in the meantime it’s business as usual. Here’s what has our attention this week at the intersection of generative AI and corporate communication:
The McNamara Fallacy and AI
Midjourney Keeps Getting Better — and More Accessible
Book Recommendation: The Skill Code
How One Tech Writer Moved Past AI Resistance
The McNamara Fallacy and AI
Don’t let the search for KPIs interfere with how you use AI.
In the rush to quantify the impact of artificial intelligence, are we at risk of falling into a trap that has ensnared leaders for decades? Ethan Mollick recently re-posted on old post of his on X that speaks to a number of conversations we’ve had recently.
As organizations reach to assign metrics related to their use of AI, we believe some are overlooking a crucial step: building a deep understanding of the technology itself. The McNamara Fallacy, named after Robert McNamara, U.S. Secretary of Defense during the Vietnam War, describes a form of cognitive bias that overemphasizes easily quantifiable metrics while ignoring harder-to-measure factors. It unfolds in the four stages Mollick references in his post:
Measure what can be easily measured.
Disregard that which cannot be easily measured.
Presume that what cannot be easily measured isn’t important.
Say that what can’t be easily measured doesn’t exist.
This progression leads to decisions based on incomplete or misleading data, often with negative consequences. In the context of AI, the search for these measures creates a specific challenge. Companies and leaders are focusing their attention on metrics first, rather than grounding themselves and their teams in the technology itself.
Instead of immediately reaching for the yardstick, organizations would be better served by investing in education and experimentation. This means creating spaces for teams to play with AI tools, share learnings and best practices with each other, and discover novel applications. It means fostering a culture of curiosity and continuous learning around AI.
This exploration phase yields benefits that far outweigh any short-term metrics. Teams develop a nuanced understanding of when and how to leverage AI effectively. They learn to ask better questions, to prompt more effectively, and to critically evaluate AI-generated outputs. All of these are skills needed to truly get the most out of AI in the long run.
None of this is to say that metrics are unimportant. They play a crucial role in decision-making and resource allocation. But in the rapidly evolving field of AI, premature quantification can be as misleading as no measurement at all. The key is to strike a balance: foster understanding and exploration first, then develop metrics that make sense in the context of how your team can get the most value out of AI.
We want to resist the urge to fall into the McNamara Fallacy. Instead of asking “How much is AI improving our bottom line?” start instead with “How can AI make me and the work I do better?” Bring AI to the table, engage in meaningful conversations about its potential, and allow your understanding of the technology to grow organically.
Midjourney Keeps Getting Better — and More Accessible
Free access for everyone, improved image editing, and no more Discord.
Longtime readers of Confluence will know that Midjourney is our preferred tool for image generation (and the one we use for our cover image almost every week). In our view, the diversity and quality of Midjourney’s output is superior to other tools we’ve tried, and its images mostly lack the “cartoonish” quality we’ve written about before. Despite these advantages, mainstream adoption of Midjourney has lagged behind OpenAI’s DALL-E, most likely due to two factors: the difficulty of the user interface and the cost ($10/month for the basic plan). Both of those barriers have fallen in recent weeks.
First, Midjourney announced last week that users can generate 25 images for free. And second, the Midjourney web interface is now available to all users. Previously, new users had to access Midjourney through a separate platform, Discord, in a confusing, multi-step process that many of our clients and colleagues found too tedious. Now, users can create an account directly on midjourney.com and begin working in the new interface right away. We’ve had access to the new interface for several months now, and we’ve found it to be a vastly superior experience. In short, Midjourney is much easier to access and use today than it was even just a few months ago. If the interface and challenges have previously held you back from trying it, now would be a great time to revisit.
On top of those changes, both the underlying model and the product itself continue to improve. One of the most notable new features is the ability to edit images directly in the web interface. We used this editing feature to add the “Confluence” banner to this week’s cover image, for example. Below we share a lighthearted example that shows this feature in action, from beginning to end.
This weekend, the four-year-old son of one Confluence writer had his first soccer game. His team is called The Badgers, but he had no idea what a badger was (neither writer nor four-year-old live in Wisconsin). So, the writer used Midjourney to create a simple image of a badger playing soccer:
A good start, but the four-year-old wanted to see a boy in the picture as well, so we used the image editor to expand the frame and updated the prompt to mention that the image should include a four-year old boy. The result:
We then decided to zoom out, which takes just one click in the image editor, to show more of the setting:
Finally, when gameday arrived and the coach handed out jerseys, The Badgers were proudly kitted out in purple and white. So, for our last change, we used Midjourney’s “inpainting” feature and edited the prompt to change the jersey’s color to purple:
The result, in this case, was a delighted four-year-old — and an impressed and more capable Midjourney user. It’s a silly example, but it does show just how rapidly these tools are improving and getting easier to use. We don’t expect that trend to slow down any time soon. If you’ve stayed away from Midjourney thus far, we’d encourage you to give it a try and see what you can create.
Book Recommendation: The Skill Code
Matt Beane’s new book provides valuable insights on the human side of AI adoption.
For several months now, we’ve pointed Confluence readers to Matt Beane’s Wild World of Work Substack. Matt is a professor in the Technology Management Program at the University of California, Santa Barbara and is one of the world’s leading experts on the intersection of advanced technology and human skill in organizations. All of his work is worth reading. His new book, The Skill Code, is no exception.
The Skill Code focuses on the implications of advanced technologies — including, but not exclusively, generative AI — on the expert-apprentice relationship, which has been a critical component of skill development for as long as humans have been learning. The book lays out a framework for the elements of a fruitful expert-apprentice relationship (the “three Cs” of challenge, complexity, and connection), outlines the threats to each of these in the age of increasingly advanced technologies, and explores strategies that individuals and organizations can employ to overcome these challenges. It’s certainly a timely topic, and we continue to believe that “human factors” like these are under-discussed in most organizations right now. For anyone looking to begin (or deepen) that conversation in your organization or on your team, reading The Skill Code would be a great place to start.
How One Tech Writer Moved Past AI Resistance
Adopting AI can be challenging, even for experts.
Think AI resistance is just for luddites? Think again. A recent article by Rhea Purohit, AI writer for Every, dispels this notion. In her piece “Why I Avoided AI—And How I Finally Embraced It,” Purohit uses her own experience to provide a candid exploration of the barriers that often impede AI adoption, even among those well-versed in the technology.
Purohit posits that reluctance to integrate AI tools, even in tech-savvy professionals, often stems from a combination of comfort with existing workflows and simple uncertainty. This tendency is further amplified by our natural inclination to prefer familiar methods, especially when faced with time pressures or looming deadlines. Purohit’s journey from avoidance to acceptance offers a blueprint for overcoming these obstacles. She outlines several strategies she’s using to embrace AI:
Take it slow: It doesn’t have to be a full 180. Integrate AI gradually, bit by bit — think of that overused idiom about the journey of a hundred miles beginning with a single step.
Design your own curiosity: Actively nurture interest in AI through regular exploration and experimentation, making time to ask and answer questions to diminish your uncertainty.
Let your environment shape you: Surround yourself with AI enthusiasts and engage with AI-related content. (And since you’re reading Confluence, you’re well on your way on this one.)
Purohit likens the process of gradual AI integration to learning a new language—progress is made through consistent, small steps rather than immediate fluency. The post is a must-read for anyone looking to get more comfortable with regular AI use, or nudge their team in that direction. There's no doubt that it’s a journey that takes time and intention, but we believe that finding your way is well worth the effort.
We’ll leave you with something cool: Curtis Sittenfeld is an author who has mastered the art of the beach read. But, given the same prompt, can ChatGPT give her a run for her money? This New York Times article lets you decide.
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.
The McNamara Fallacy is a great find! Go Badgers!