Within the world today the term AI seems to be being used everywhere. it’s cool, a mother told me she uses GenAI to generate new PAwPatrol stories every night – and that it won’t let you create a story where one of the characters dies. Pretty cool eh!
It’s the new data (which was the new oil).
Last year (2023) I worked on a paper for the UK Cabinet Office. https://www.gov.uk/government/publications/generative-ai-framework-for-hmg
We published after much research on the subject Guidance on use of Gen AI.
The research included visiting Amazon, Goggle, IBM, MS and other big techie players.
The biggest questions that arose were not about the ethics, the legality, the use, the trust but about how will businesses manage the AI and LLMs over time: what skills are required? What roles will be needed?
Lots of Data Scientists, Data Engineers, Data people have been soaked up into organisations but are those the right kind of people to manage LLMs over time? How do we know?
At IBM a subject called ‘drift’ appeared. Drift is the concept: “Semantic drift” is changes in the meaning of data while the structure does not change.
Who will manage that in your LLM? A data scientist or someone with a feel for integrity, brand values, pattern seekers, philosophers?
Bad actors to your corporate LLM will and can deliberately change your LLM through interactions. How will you manage that? What will you do to prevent bots hijacking your open interface filling it full of nonsense to just disrupt you and gain competitive advantage.
Another case study in the MoJ was using GenAI to support users to seek help and advice from and AI based on the reams of info on the law. A question that arose was :”How will your GenAI bot communicate with users who might be talking about sensitive subjects without re-triggering the user in their responses? (e.g. Sexual abuse) ”
From a UI/UX perspective : It’s ok to have a vidi-type prompt putting text on the screen to make you feel as if the computer is talking only to you, to trick you into feeling wanted, needed by it. But other interfaces are required to be accessible, hoe could you embrace a brand such as rthe Michelin Man to tell you what pressure your car tyres should be?
There more to deploying AI than meets the eye and plenty of opportunities to personalise.
Time to think is now, design thinking is vital, mitigate the risk of deploying the wrong usable thing and maybe you’ll get an advantage.
Or we’ll discover a new oil!
