Every week Surevine runs an internal unconference called “Techcellence” where we present and discuss things we have learnt from the previous week. Don’t let the title fool you, we cover a whole range of topics to ensure everyone in the company can get involved.
Last week we had our first external speaker join us. Caroline Jarrett, the legendary form designer, presented her new talk “How I became an AI feasibility investigator”. At Surevine, we’re excited about the application of new technology but experienced enough to know that a problem needs to be understood before picking a solution so the title was intriguing.
Caroline is still working on the talk so we don’t want to give too much away, but here is what we learnt.
10 minutes of fun, 30 minutes of problems.
In the tech world, it can be tempting to dash headlong into cool technological solutions, but this inevitably leads to problems if you haven’t done some groundwork to understand the problem to be solved. As the old adage goes “If you have 5 mins to chop down a tree, spend the first 3 mins sharpening the axe”. Caroline walked through four examples of her personal experience where AI had been used effectively or not, and why.
The common themes for success were:
When making the case for AI, context is everything.
Be specific about the problem to be solved. Demonstrate the problem rather than talk about it.
Understanding the whole system.
Introducing improvements to one part of a system without understanding the whole inevitably leads to bigger issues. Local optimisation rarely has an impact on the whole and often introduces new problems. “There is nothing so useless as doing efficiently that which should not be done at all.” – Peter Drucker.
Understand the roles in a system, and where AI can benefit.
Is this the most effective place to apply AI? AI is great at large processing of well-understood data so is amazing for removing constraints from a system, but if the data is inconsistent or flows through the system simply it’s not going to be beneficial. Not every problem can be solved with AI.
Understand the quality of the data that the AI will use.
Tightly constrained data with clearly defined outputs is still the best use case for AI systems. Don’t be afraid to handoff results to humans for validation.
People will always enter what they want regardless of the rules.
If your AI can’t handle edge cases it will fail, so make sure you take the time to understand them. Do user research and prototyping to validate behaviour.
Data accessibility is a challenge.
Could your AI budget have more impact if it was funnelled into improving access to data so that people can do their jobs simply and effectively?
The talk was interesting and thought-provoking allowing for great discussion with Caroline at the end. And it’s done what it set out to do – make our engineers start thinking about how AI could help in an AI feasibility study! Big thanks to Caroline for opening our minds to something new.
If you’d like to hear the new talk, please get in touch with Caroline at Effortmark.co.uk.
Or if you’d like to be a guest speaker at Techcellence, drop us an email at firstname.lastname@example.org.