AI has become an inescapable topic, with everyone holding an opinion, many of which are contradictory. The AI hype touches all aspects of our lives, further confusing the subject. So, to simplify, allow me to clarify the world of AI regarding enterprise communications. Let’s start with the conclusion: Generative AI changes everything, and yet Generative AI won’t last.
If that didn’t help, let me clear it up even further.
Unraveling the Paradox
The contradiction stems from the premise that because the term AI is thrown around so loosely, many simple statements are bound to create conflicts. We need to start with some basic definitions to make sense of it.
Let’s start by categorizing all AI into three groups: Generative AI is the current darling of the tech world, captivating every product manager and investor for nearly two years. However, AI's history stretches back much further, encompassing breakthroughs in machine learning (ML) and neural networks. I group all those breakthroughs before generative AI into a single container called ‘Legacy AI.’
The third category of AI has yet to arrive, so I’ll refer to it as ‘Nexgen AI.’ Generative AI and large language models (LLMs) are creating a lot of buzz because many believe they could lead to Artificial General Intelligence (AGI)—the controversial idea that computers might one day match or surpass human cognitive abilities across a wide range of tasks.
Shifting AIs
Despite its achievements, Legacy AI is being sidelined by generative AI. Many examples illustrate this out-with-the-old shift. Legacy AI-powered assistants like Alexa and Siri are making way for new generative AI-driven replacements. The transition adds to the confusion because Legacy AI is just as effective as it was two years ago, and often better than generative AI. Legacy AI has enabled significant recent advancements in enterprise communications. These include blurred and virtual video backgrounds, improved noise reduction, virtual agents, transcription and translation services, and more.
Legacy AI’s strengths lie in its consistency and reliability. It executes trained processes flawlessly and consistently. Legacy AI conquered Jeopardy! and outplayed human champions in Chess and Go. Generative AI isn’t quite there yet and finds Sudoku challenging.
Generative AI shines in human-like interaction and requires less training. Straight out of the box—or more accurately, out of the cloud—it can answer many questions naturally. It boasts a vast, multilingual vocabulary and can comprehend most requests and conversations. Generative AI is not confined by knowledge boundaries or training limits. Legacy AI, however, sticks to its training and scope.
In comedy terms, you want Legacy AI for sketch/scripted comedy and generative AI for improv.
Today’s Decision
If you're an enterprise communications product manager, should you stick with the tried-and-true or embrace the new? It depends on the application. The answer appears to be a little of both: legacy AI remains useful, but generative AI, despite its limitations, is young and scrappy, but may turn into something.
Generative AI is very good when given boundaries, like summarizing meetings, conversations, emails, and reports. The output is relatively accurate as long as the source content is part of the request. It’s not perfect, but it’s a new form of automation that is proving quite useful. Summarization is one of the best uses of generative AI in enterprise communications. However, generative AI can falter when boundaries are vague. For instance, asking for suggested pizza toppings could get you the suggestion of "glue" to get the cheese to stick.
Some vendors are front-ending their legacy AI solutions with generative AI. This combines the human-like conversational capabilities of generative AI with the time-proven, ML-assisted workflows of Legacy AI. Legacy AI handles defined workflows better, like rebooking a flight, while generative AI provides superior, human-like interactions. It can effectively understand the inquiry and relay the result. Product managers are attempting to extract the best that the old and new each have to offer.
This explains another apparent contradiction. Several vendors tout years of experience with legacy AI will lead to success with generative AI. Yet, we keep seeing evidence that Legacy AI is being pushed out. Legacy AI use cases still apply in enterprise communications. The future of enterprise communications, especially customer-facing use cases, will involve legacy and generative AI.
The Future
Now, let’s talk about the third AI—the future AI. LLMs have brightened AI's future. Generative AI has opened a new chapter in AI research, and there’s a widespread belief that we’re on the brink of major breakthroughs, including AGI. For investors, there’s no legacy versus LLM question; the future (most likely) will come from LLMs. There’s no more reason to invest in legacy AI technologies as they suddenly become mature.
Legacy AI will likely improve, but legacy and and generative AI are at opposite ends of innovation S curves. An S-curve shows the innovation is slow at the beginning and ending of a lifecycle with an accelerated phase in the middle. Big Tech and Wall Street are racing toward that acceleration. Even though legacy AI is superior today, the expectation is that generative AI will soon surpass it.
Generative AI has also captured the attention of businesses of all sizes across many industries. But, as we are finding, the technology has significant limitations today. Its propensity to hallucinate limits its enterprise suitability. However, it can enhance employee productivity. The costs are also uncertain, but the opportunity cost of ignoring it may be higher. We are seeing some spectacular failures with generative AI, as some companies are very eager to deploy and learn. Apple appears to be much more cautious, and intends to enable features and geographies cautiously.
As LLMs evolve, their influence on enterprise communications will increase. Legacy AI's proven reliability still has significant value but isn’t likely to dramatically improve. The smart play isn't to choose one AI type over the other but to leverage both.
Dave Michels is a contributing editor and Analyst at TalkingPointz.
