When is good enough enough?
I am a frequent flyer, and though I shan’t name the airline specifically, I am a platinum member of its loyalty program. Even with that lofty status, I can tell you that its customer service stinks. Most of it is automated, and speaking to a support worker is near impossible. If you talk to the chatbot, it regurgitates preconfigured info that, rather than resolving your issue, regularly frustrates you to the point of madness. I’m pretty sure everyone reading this column experiences similar frustration with self-service chatbots. And the reasons for this are well known: Chatbots typically pull answers from a well-defined but minimal knowledgebase—a sort of FAQ list. If your question varies slightly from the norm, and many do, the chatbot can’t help you, although it will keep on trying till you totally lose it.
Moving forward, generative AI will improve the ability of chatbots to answer questions accurately and immediately by an order of magnitude. They will do this because the knowledgebase does not limit them—they trawl multiple sources and can find answers constructed from various sources and stitch them together in a coherent fashion in real time. That’s a big deal and a reason for celebration. Will it get it right every time? Absolutely not, but even if it got it right 80% of the time, I would be a much happier customer than I am now. One hundred percent or even 99% accuracy is not a requirement; just the ability to make a good guess at your question’s answer is a big step forward. And you can likely see where I am going with this, as a darn good guess will not cut it in mission-critical healthcare, finance, legal, pharma, defense, science, and engineering.
Orthodoxies such as the single source of truth or “80% of data is unstructured” don’t help us progress; they don’t help us at all, nor does planning for perfection. The world is a mess; data and information assets are chaotically organized, and that’ s how it is, folks. Yet there is a fertile middle ground to plow, and we will always take “much better” over “much worse” any day of the week. Our goal should be to improve the quality of knowledge assets and their accuracy and relevance in use. Much of this will come from human expertise and effort, increasingly combined with the power of AI. Even so, there are situations when the shortcomings of AI mean they can have no role to play; as long as we are clear on what those situations are, we can forge ahead. And though it may sound defeatist, good enough sometimes is more than good enough.