Doesn’t everyone have a favourite AI use case?

My lightbulb moment for understanding both the immense power and inherent limitations of generative AI came when I saw an AI-generated kitten so adorable that I didn't immediately notice it had six legs. I noticed after a few seconds, and my 11 year old daughter did too. In that instant, some of my fear about AI taking over from humans faded: we still have the unique ability to add value where it really matters.

You can see the kitten for yourself if you visit https://www.somersethouse.org.uk/whats-on/cute. I tried to recreate it but instead created the monster you see here. This horrifying three-eyed unikitty with at least 14 legs was described to me by DALL.E thus: "Here's the updated image of the kitten-unicorn now having six legs, subtly integrated into its design. The extra legs are hidden within the fluffy mane and pose, making them not immediately obvious..."

Sure. So AI is not great at nuance or art, but it's brilliant with data and information. So lets start there.

Use cases* dominate much of the conversation around the dramatic rise of generative AI. Most are obvious once you’ve played with tools like ChatGPT: you can find and generate high-quality, reliable content incredibly quickly, saving time and freeing up resources, etc. Things get a bit more interesting when you start to admit that, creepy kittens aside, AI is not just faster but sometimes better than the alternative.

I have two favourite generative AI use cases once you rule out 'making bad art for the LOLs'. Neither is remotely cute, but will do more good for humanity — and for your business — than adding to the online availability of cat pictures:

1. Extracting unbiased, accurate, comprehensive corporate knowledge - There’s an ideal corporate world where everyone remembers every discussion from every meeting, and you never have to repeat the same debates. Failing that, everyone is really good at consistently naming files and putting them in the right bit of Sharepoint — removing unapproved, out-of-date, or dodgy data as you go. Failing that (surely not?), your best bet is often to ask Susan, or whoever in your office has been there the longest, is well-organised, and has a pretty reliable memory. If you have limitless time and resources, you might even ask the new hire to trawl through papers and folders and write you a top-notch briefing note. But none of those solutions are quick, and you probably won’t get the full story. Train a private GPT on some poorly named, moderately well-written papers and notes, though, and... voila! Clear, accurate summaries of what went before so you don’t have to take a step back before you take two steps forward. Let’s call that ‘better than Susan’ (sorry, Susan).

2. Helping people who probably won’t say thank you but will ultimately be better off because of what you did - There’s a rotten thing that happens to people who need support from the state — whether that’s a traumatised child removed from their family for their safety, or an adult relying on a network of carers because they can no longer manage for themselves. Your life story is all around you in images, records and stories, but vulnerable people’s life stories become a set of notes and files that are hard to access. You can ask, and you might get sent your files, but it will take a while and they will probably be redacted. Ironically, you will probably feel worse after going through a Subject Access Request than you did before you started. It’s no one’s fault: people are short on time, the information is sensitive, and someone has to do a huge data trawling and checking exercise just to answer the question: why was I sent to this hospital/assigned to this foster family/taken from my home? And even before that, decisions are made based on interpreting a mass of rules and procedures. But if you can find that information faster, have it be right every time, and keep the human in the loop where they need to be — making decisions and checking what’s right— then maybe it doesn’t have to be this way.

Happily, I don’t have to choose between AI for LOLs and AI for better outcomes; I can play with DALL.E in my free time while working on both of these problems right now: private GPT tools that enable you to talk to your data, securely, whether it’s structured or unstructured. Maybe it will help a grumpy teenager, maybe it will help a grumpy CEO. Either way, it’ll help.

*The scenarios that showcase how AI can be applied in practice to solve a problem or meet a specific need.

‍