When it comes to completing projects, the pareto principle often works in overdrive.
As someone once said to me that once you think you’re 90% done on a project, you’ve got another 50% to go.
This observation is somewhat analogous to the “last mile” challenge, which arises in various fields, such as supply chain management, transportation planning, and telecommunications infrastructure.
Someone can drop a package off at a post office, and getting that package to a distribution centre, to a port, to another country, and to one or more distribution centres, is a relatively efficient process. My understanding is that it’s getting the package to its final destination that stumps people trying to improve efficiencies. In fact, some sources suggest that in many cases, the last mile delivery cost is more than half of all delivery operational costs.
I’ve been playing around with various AI services lately, and the concept of the last mile has particularly struck me.
As an example, I used ChatGPT and Midjourney to generate the following image. Midjourney generated the image (and various iterations before I settled on this one). ChatGPT helped me to come up with prompts to give to Midjourney.
I started by asking ChatGPT to help me come up with prompts, by sharing the “About” page of this blog. I asked ChatGPT to come up with suggestions for an image or mascot or logo for the NZ Wealth & Risk blog based on this information. I then told ChatGPT the ideas I liked, and asked it to generate some more. I did this again a few times, and eventually settled on a few ideas I liked. I then asked ChatGPT to suggest some prompts, which I tested in Midjourney.
With Midjourney, I used these prompts to see what I liked. I then used a “cheat sheet” I’d created to make further tweaks to the prompts, before coming up with a prompt that generated something similar to what I’ve shown above – an overflowing treasure chest, with gold hearts instead of coins. (The prompt itself includes the words “representing harmony between financial goals, values, and life priorities”.)
I found an image I liked, and created a number of variations of that image. As you can see above.
Which I really like. I think it looks great, and I feel it resonates pretty well with this blog. Maybe I’d have come up with this concept if I’d spent enough time brainstorming and sketching, or if I’d used a design professional to come up with it. But I suspect it would have taken a long time, and probably been an expensive process.
The thing is, the image has some shortcomings. If you look at it for more than a few seconds, you’ll notice some weird quirks. Some of the hearts are weird. Some even look like they’re broken hearts. The “V” on the front of the chest looks pretty cool, but it doesn’t match with the trunk’s lid. I could go on and on.
It didn’t take me long to come up with this image. Probably 20 to 30 minutes. However, perfecting the image would take me a lot longer. Whether that involved using Midjourney or DALL-E 2 and additional prompting, using Photoshop, or engaging someone to “perfect” the image, it’d take a lot to get it where I want.
In the case of this image, I’m probably going to use it for this blog, and perhaps use variants of this over time, and make tweaks as and when I can be bothered. For my purposes, near enough is good enough.
But that’s not the case in many situations.
An AI might provide a lot of great suggestions and direction, in lots of domains. But getting from 90%, or 95%, or 99% to getting “all the way” will be a final jump that’s hard. A lot of the time, 90% or 95% or 99% might be fine.
But in other domains, it might not be enough.
Which is a long-winded way of saying: my hypothesis with a lot of AI tools is that they will help people in a lot of ways. However, people using these tools will often experience a modern version of the last mile problem.
In lots of cases, AI will speed up a lot of the process. But in speeding up the initial 90% to 99% of the project, it might mean that an even bigger proportion of the time and resources will be needed to get projects fully completed.