Three spark plugs that fire up your generative AI engine

About the Author

Paul Roehrig is the Chief Strategy and Marketing Officer for Ascendion. He is a co-author of multiple award winning and best selling books and a sought after presenter at public, academic, and industry events. He is regularly featured in major publications all over the world. Paul holds a PhD from Syracuse University and was formerly a professional musician. He lives in the Washington, DC, area with his family.

Special thanks to Muthu Chandra, Director of Artificial intelligence & Data Science at Ascendion.

Seemingly overnight, the whole world caught fire with the idea that Generative AI strikingly illustrated by ChatGPT, Bard, DALL·E, etc. will either make our lives hell or create a techno utopia.

Of course, the likely truth is somewhere in the middle, but the groundswell of expectations is forcing business and technology decision makers to rapidly begin exploring options to deploy and even customize Generative AI solutions to drive near term value.

Our new machines are now creating diverse content types like text, images, and responses that eerily mimic our own human dialog. It can be simultaneously exhilarating and confusing. So where should we start?

Based on our experience, we’ve identified three critical steps to take at the beginning of the trip to ensure success of enterprise IT projects to mine value from generative AI today.

1. Focus on the experience. Right now, we’re throwing ChatGPT at … well … everything. That’s fine in the early phases, but AI solutions that deliver real value will start with the experience and engineering from there. This could be someone doing enterprise work (loan processing, insurance underwriting, pharmacovigilance, etc.), or it could be a consumer experience (customer support, ecommerce, gaming, etc.). In each case, considering the audience, the expected outcomes, the ideal process flow, the beauty of the engagement, and more will help drive joy, trust, and value. The AI output needs to be aligned to the specific use case experience. Every word, image, sound (and someday taste, smell, haptic response) should improve the process to provide a terrific and more productive (!) experience. Otherwise, it’s investment money down the drain.
2. Embrace Iteration. Once you’ve targeted a specific moment in a process where generative AI can help decrease friction, keep in mind a hard truth: It won’t be great the first time through. The key here is “generative.” These systems get better the more they are used. The data set you’ve used to create (or “train”) the system will get you started, but there is no substitute for real-world data from live humans. This is all to be expected. (Even Google wouldn’t “work” without 100,00 searches every second!) Plan on questionable initial results, but have confidence that it will improve quickly over time.
3. Mine the right data. There is not a company we work with that isn’t drowning in data. The problems usually (almost always) are related to: it’s not the “right” data; we can’t find the data; we don’t know what the data means; and a regular favorite the data is too expensive to get/keep/secure/analyze. How do you know what you really need? Go back to the experience, the process. Once that is crystal clear, you’ll know more about what data you really need to generate content, recommendations, next best actions, sound, etc. It could be past user behavior, preferences, emotional state, language, tone, aesthetics, health history, geo-location, sensor data, and more. This is nothing new. Companies have always considered these things, but now those decisions are virtual, take nanoseconds, and are “final” (because poor digital experiences are no longer acceptable to any of us). The key is to align the data to improving the experience. Otherwise, generative AI will stay in the realm of parlor tricks, jokes, and Drake deep-fakes.

This is, admittedly, a simplification, but Shakespeare rightly said that, “brevity is the soul of wit.” There’s merit in simplicity, in starting with clear goals, because as we say we should not expect progress with generative AI to be a straight line. Many questions will need to be answered farther down the road.

Expectations for quality and relevance will need to be set. Criteria boundaries that balance leniency and severity will need to be encoded. Subjective elements like tone and aesthetics will need to be negotiated and aligned. Ethics and security considerations must be wrestled to the ground. All of this will take place in a context where the technology is changing daily, or hourly.

This quest is not for the meek of heart. It can be overwhelming to try to wring business value from a new technology that, candidly, we’re all still figuring out. But this is the fun part! Start simply. Kick start your Generative AI engine with the “spark plugs” we’ve outlined here to grab real business impact in the near term.

For more on how to tame AI to create your own enhanced organization, please see: What to Do When Machines Do Everything: How to Get Ahead in a World of AI, Algorithms, Bots, and Big Data by Malcolm Frank, Ben Pring, and Paul Roehrig, PhD. John Wiley & Sons, Inc., 2017.