TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

The Problem and Promise of Generative AI

What you need to know about generative AI tools and why it is critical to understand the technology quickly.

• By Rob Enderle
• April 24, 2023

Last month, Microsoft launched Copilot for Office after previously announcing it for many of its programing tools, Edge browser, Windows, and Bing. IBM has been advancing AI for some time, and Google and Apple are racing to catch up. This promises to be the most disruptive technology wave so far, and it isn’t even the last one we’ll see this decade. Think: advancements in quantum computing and robotics that are slated to enter the market soon.

For Further Reading: The Importance of Generative AI and How Education Is Getting It Wrong Why Conversational AI Is a Game Changer for Support at Scale How AI Will Advance This Year

Let’s talk about the promise and current problem with generative AI tools such as ChatGPT and Microsoft Copilot and why it is critical to understand this technology quickly.

Generative AI in a Nutshell

Generative AI, also called conversational computing, is the application of increasingly massive natural language models into a new class of common language interface tools. It is the extension and progression of earlier human/machine interfaces that evolved from punch cards to GUIs last century and then stagnated for a bit.

Generative AI is designed to allow users to increasingly interact with their high-tech products as if they were talking to another person, but as with another person, you must be both clear and complete with your directives. Although generative AI is getting smarter very quickly, it is still young and in need of significant training to perform as expected.

It isn’t just limited to producing text. Generative AI is being developed to work with and create images and videos as well. Many of these efforts will become increasingly blended. This blending will allow a future user to make complex requests and have the computing system on their desk automatically launch and direct multiple applications to complete the user’s task.

In short, this product class is on the path to full automation, but it has a long way to go before it gets there.

Generative AI’s Promise

The underlying concept is to create a user interface that doesn’t require training and will execute even complex tasks autonomously. Somewhat ironically, the skillset needed is similar to working with a new employee who understands how to use a calculator and typewriter but doesn’t yet know the unique policies and behaviors that surround the specific tasks you need done.

The promise is a kind of automated subordinate that will take our direction and then, using the tools available to it, complete a complex task we need performed. In a way, it is the classic promise that has yet to be met of a truly smart computer that not only understands what we need but can go out and independently accomplish the related tasks to create a finished, high-quality outcome.

This would mean massive productivity improvements and potentially higher quality, but we aren’t there yet.

The Generative AI Problem

Unfortunately, this class of tool is very immature now, having just been introduced into the market. It often can’t tell the difference between good information and bad information, and mistakes or outright damage can be introduced into the language models that result in less-than-optimal outcomes. In addition, few people know how to work well with human assistants or how to make requests so well defined that an AI process can’t misinterpret them.

This means that users need training so their requests result in the minimal number of corrective iterations. They must focus sharply on quality or that quality will suffer, and their mistakes will mis-train the AI which may then make those same mistakes at a far greater scale. It may well be that we’ll end up with yet another AI class focused more on correcting errors than on creating content, one that will be isolated from users in terms of advancing their back-end language models and focused on improving the quality of the result, not creating that result. Grammarly has announced an editing-focused generative AI offering focused specifically on finding and correcting the mistakes that a productivity-focused generative AI creates.

Final Thoughts

Generative AI tools such as ChatGPT are becoming the new interface into productivity tools such as Office, operating systems (including Windows), and web services (think: Bing). It is a game changer with the potential to cause more disruption and impact more people than Windows or the iPhone ever did.

Making sure you learn how to work with, integrate, and ensure the quality of the results from this tool should go a long way to securing your value to your company and help keep you from being functionally obsolete as this class of tool becomes more common and proficiency in it becomes a requirement for employment.

As this capability expands into automotive, robotics, entertainment, and medicine, disruption should reach a level we haven’t seen before. Making sure you both understand the related disruptive threat and how to embrace and benefit from it will be a critical path to both survival and success as the market pivots to generative AI.