Recent developments in AI have challenged innovation-minded R&D and business leaders to carefully consider how to use the technology in all areas of their businesses and in new products. One area of adoption we’ve noticed is the use of AI as a shortcut to develop management presentations, investment pitches and even in developing product specifications and project plans. And this got me thinking about a model I’ve often used for crafting important stuff like this called, “Madman, Architect, Carpenter, Judge”. It’s advocated by Bryan Garner, author of the HBR Guide to Better Business Writing – I even used it to write this blog!
It got me wondering – is AI really good at all of these roles, or are some of these better left to a human touch? Is AI a better “Madman”, or a better “Judge”?
First, let’s unpack the model…
- Madman– this is the creative stage, where you let your imagination run riot and capture every idea as it comes to you as quickly as possible. There’s no editing or filtering; no crafting of sentences. You just want to get everything down on paper that might be relevant to your purpose. Here you’re thinking with lists, spider diagrams, mind maps and post-it notes.
- Architect – now you look at everything you have gathered, group your ideas and come up with an order. You consider what’s most important and what should come first. You organise your content into a basic outline and helping to structure the narrative and flow of information.
- Carpenter– this is the building stage where you write everything out in full, following your outline and adding the details. It’ll take time but you try to do this without thinking too much about how things are phrased, there’s time for that later.
- Judge– get a coffee… then go back and look critically at what you’ve written. “Could I say this better, clearer, more concisely?” becomes the mantra. Is the tone and emphasis correct for your audience. You remove repetition or clutter that risks muddying the message. Crucially, this is also the stage when you assess the accuracy of your content – can you really support what you’re saying and how you’re saying it?
Don’t let AI be your Judge!
In my experience and that of the colleagues I’m learning from about AI, there are some important dangers to highlight of relying on AI to evaluate your work:
- Hallucinating: a problem where the AI model will produce confidently sounding but factually incorrect outputs. (To attempt to counter this, Microsoft have recently linked Bing to ChatGPT in order produce accurate references to accompany the AI-generated text, but from our experience this is still a work in progress.)
- Changing personas: Large language models are trained to predict the next most plausible token given a context window. In case of ChatGPT, there’s a small prompt hidden from the user that enables the AI to take the persona of a helpful chatbot and respond in such a style. There have been reports that in longer conversations the chatbot can occasionally take on a different persona (e.g. suddenly responding as a child would) – this is the sort of inconsistency a human will spot instantly.
- Bias: The AI training process consists of both supervised learning and reinforcement learning using human feedback. The model is therefore shaped by the views of the people providing feedback, who will inevitably have their own biases. So the creation of the resulting dataset can introduce unintentional biases to the model. For example, there have been some reports that ChatGPT expresses political bias when subjected to political compass tests. Of course we all have bias (that’s a human characteristic too as it happens) – but you need to “judge” carefully you agree with any bias your pitch is implying.
This is why I think AI shouldn’t be used as a Judge. Instead, I’d suggest analysing and interrogating your pitch/presentation/project with other experts from your industry and beyond who can bring complementary perspectives and “judge” your work in a fresh light. An AI’s response is also only as good as the questions you ask it, often the most valuable feedback in innovation and communication is – are we asking the right questions, or solving the right problem? So for now, many fresh (human) perspectives are still better than one AI!
But AI does make a great Madman!
Conversely, AI is superbly positioned to extend and accelerate your work as a “Madman”. AI tools can be great for:
- Gathering targeted and tailored research that would traditionally have been too time consuming and/or costly to do manually.
- Quickly exploring and suggesting ideas, options and sources based on your prompts.
- Refining and replicating findings across other problem areas once you’re confident in sources, processes and content.
However be warned – while AI tools may speed up any of the research or content generation, the need for rigorous fact and phrase checking becomes even more vital, and potentially time consuming.
And it seems that AI agrees! When a colleague asked it how best to use ChatGPT to write a business proposal it replied “Seek human input and feedback… always seek input from human experts, colleagues, or mentors who can provide valuable insights and help refine your proposal further.” Or perhaps its just telling us what we want to hear 🙂
Of course I’m still learning how and when to use AI best to complement and accelerate our work in innovation and business strategy – and no doubt in parallel ChatGPT is continuing to evolve even more quickly and there will be new and more capable judgement tools out soon.
In the meantime, what’s been your experience? Do you use AI as a Madman, Architect, Carpenter or a Judge of your work?
For a more in-depth conversation please get in touch.