What ChatGPT is Showing Us About the AI Revolution

If you’ve been online recently (and if you’re reading this, you obviously have), you’ve heard about the artificial intelligence chatbot called ChatGPT. It’s the miracle AI tool that can do your kids homework and write all your emails. You also may have heard that this disruptive technology is coming for everyone’s jobs. While this existential threat is most certainly exaggerated, ChatGPT is providing us with a tangible view of how AI will affect our daily lives both at home and at work. As I’ve seen the hype engine spin around ChatGPT, here are some things I’ve observed that may tell us more about how the AI revolution will unfold.???

What seems to have caught most people by surprise with ChatGPT is how capable it is of supplying direct, complete, coherent, and usable responses to user-provided prompts. The baseline comparison is with a standard search like Google, and the efficiency of ChatGPT’s responses are jarring. AI has quietly become incredibly capable in areas many thought it might never be proficient. Just months ago, the prevailing public sentiment was that AI would first revolutionize and replace menial, repeatable, physical tasks like those found on the factory floor. Furthermore, most people probably would have told you that creative cognitive capabilities would be the last set of skills to be conquered. ChatGPT is showing us that the exact opposite is most likely true. We are seeing an explosion of AI tools that turn text and speech into copy, audio, images, and even video, but we are not yet surrounded by autonomous robots in our everyday lives.

I think one of the primary drivers for this is that cognitive endeavors must tackle tasks only in the intellectual realm while performance of physical tasks requires capabilities in both the intellectual and physical realms—the latter being a complex engineering challenge. Another explanation is that there is not always a definitive right or wrong output in creative cognitive endeavors, while with process automation, outcomes are much more black and white. Since AI models are probability-based by nature, meaning their outputs are based on what the model sees as most probable, it seems easier to appear successful in performing creative cognitive tasks judged on a sliding scale of relativity than in autonomously performing physical tasks which either work or don’t. Much more leeway is given to the response of an open-ended question versus a close-ended question. What this means is that we likely may see the progression and adoption of AI happen in the exact opposite sequence than what we originally anticipated, disrupting knowledge-based professions before manual labor.

Another interesting observation from the success of the ChatGPT launch is that even with this revolutionary technology, adoption success ultimately boils down to user experience. ChatGPT is certainly not the first powerful, large-language model to exist. Similar models, and even earlier versions of this specific model, have been around and accessible for at least a few years. What makes ChatGPT different is the simplicity of the user interaction combined with the efficiency and quality of the response. Whereas earlier models required working through an API or using specific prompt formats to produce a result, ChatGPT has a simple browser-based user interface and can produce very impressive results with even the most informal of prompts. It was this ease of use that separated it from the pack and achieved over 1 million users in just five days! As we see the AI race accelerate, I think we can be certain that one factor separating the winners from the losers will be the ability to simplify the user experience of interacting with these powerful models.

ChatGPT is also reinforcing a fundamental concept of information technology–the importance of abundant, accurate, relevant, and accessible data in producing impactful results. Intelligence is not the default state of an AI model. Models must be trained on datasets to develop their capabilities. ChatGPT underscores this concept by not relying on the internet as a data source for its responses. Everything the model knows and conveys is derived from the datasets on which it was trained (ending the summer of 2021) as well as current interactions with humans. OpenAI, the developer of ChatGPT, readily admits that it can struggle with responses related to current events because of this limitation in its underlying dataset. This dependence on good data means that one of the biggest differentiators between AI models will trace directly back to the quality of the training datasets. This just further cements the importance of data standards and governance within an organization and across industries wanting to use AI. It also reinforces the relatively new concept of data as a marketable asset in and of itself. I expect to see as many companies peddling data as a product to support AI endeavors as there are software companies selling ai-based solutions.

My final observation worth noting from the excitement surrounding ChatGPT is that humans are and will continue to be a key part of the development and use of AI. AI models require validation and valuation of results. This how the actual intelligence of AI is achieved. This validation must come from actual humans. In fact, one of the primary drivers behind the timing of the release of ChatGPT was to widen the size of the human-based model validation sample.

Validation of the model is not the only aspect of AI currently dependent on humans. As an AI model must be trained to build intelligence, this implies that someone must be training it. Or more specifically, someone must be selecting and curating the datasets that are being used for training. The effective use of AI is dependent on humans as well. We are learning that the format and content of a prompt can have a significant impact on the quality and appropriateness of the response received from the model. This is creating a new demand for what are being called prompt engineers who are quickly developing a knowledge base of the most effective ways to achieve specific outputs.

The statement that AI is coming for your job is hyperbole. You should not fear AI as an existential threat to your job—at least not anytime soon. The far more likely and urgent existential threat comes from another human that has harnessed the power of AI to do your job more efficiently and effectively than you.

Regardless of where you stand on the relevance of the ChatGPT release in the grand scheme of things, there can be no denying that the AI revolution has begun and appears to be picking up speed. What exactly that means for industry and the human condition is still to be seen, but it should make for a very interesting ride!