AI 101: A Primer For Marketers and CX Teams in the Era of ChatGPT

With all the hype and media coverage around?ChatGPT?and the tech behind it, it’s always good to revisit the foundations for big (and small) data applications and what AI technique works best in which applications. This is especially true for modern marketers looking for smart machines to help them spot trends or automate their daily work. So if you are not already, it’s time for brand strategists, CX data teams, forward-looking customer service execs, and even?creatives?to get on board with today’s AI. This article provides a primer to help all of you to get up to speed.

But first a bit of history. AI is certainly not a new topic. In fact, scientists gathered for the Dartmouth Summer Research Project on AI in 1956, and release of the preliminary version of the AI programming language, Prolog, occurred in 1971. Heck, I was coding neural net pseudo code in 1992. Through its history, AI has been interdisciplinary, with commercial advances funded by both public and private initiatives, and developments in all corners of the globe.

In this piece I am focusing on practical, applied AI (for personal or business tasks) not artificial general intelligence - even though some see the appearance of ChatGPT as a sign that Skynet has arrived! A good overview of what it will take to get closer to real general intelligence is found this executive primer from McKinsey.

THE TWO MAIN BRANCHES OF AI

For marketers and their friends, the best way to envision the potential of AI is thinking about your last online shopping experience. Odds are you encountered a virtual assistant powered by AI, saw tailored recommendations, and received personalized offers after making a purchase, raving about it on your social channels, or filling out a post-purchase survey (small data!). These functions are commonly powered by a collection of techniques from the two primary branches of AI: reasoning systems and learning systems.

At a high level, reasoning systems offer guidance and help to automate manual processes that teams conduct every day. They are a type of AI that is programmed or guided, follow a logical process, and map conditions to actions. They may operate autonomously (like a factory robot picking an order) or interactively with humans to gather needs, ask clarifying questions, etc. Examples include chatbots, shopping recommendation engines, and guided “helper” tools in everything from your CRM to your creative design tools.

In contrast, learning systems help teams spot patterns in a consumer or other data set, make predictions, or even suggest a design or answer to a question. Learning systems apply algorithms that are trained vs programmed, and can be supervised or unsupervised. They can be focused on two types of trend detection tasks: predicting something that has happened before like seasonal shopping behavior, or spotting an anomaly that hasn’t happened before, like a new consumer fashion trend.

At the same time, some of the most popular AI applications like natural language processing (and conversational AI) apply both reasoning and learning approaches together, to help marketers or CX teams better understand what their audience is saying, or get feedback on product or service experiences by automatically processing emails or open-ended surveys. Another use case is tracking consumer behaviors and trends, and predicting what they want or need using AI and social data to boost CX.

WHY HYBRID AI IS THE FUTURE

Most learning systems cannot explain their outcome without a human helper adding context or sharing details on how the system was trained, what performance it achieved, etc. Meanwhile, reasoning systems need to be programmed, and tested, and updated. As ChatGPT shows us, techniques like?RLHF?blur the lines as direct human feedback or examples are used during the training process, and once deployed, the output can appear remarkably human-like.

For these reasons a growing number of practitioners and advisors like my friends at?Forrester?are promoting the benefits of a “hybrid” approach in domains like NLP. I have also written about the benefits of?humans and machines?working together in consumer intelligence applications, even when advanced AI or ML does the heavy lifting behind the scenes. This approach is especially valuable for mission critical tasks or consumer facing applications where there is data to understand, a structured process to follow, and feedback that can be used to tune system performance.

So just as marketers need their data friends, machines need their human friends, especially when translating big data-derived insights into small, consumable bits of advice or actions for a specific task. And anyone using learning systems needs experts to look over algorithms and outputs to watch for biases, and then adjust or communicate to stakeholders as appropriate.

Check out many other articles on AI, NLP, and the role of small data sets on Small Data Group (my side project for almost a decade), and a great new piece from TreviPay's product team on the TreviPay blog. Happy reading!