Is ChatGPT ready to crawl, walk, or run?

ChapGPT is an Artificial Intelligence (AI) program that was launched in November, 2022, and it was created by OpenAI, a research laboratory in California. While ChatGPT is an AI program, it is based on a Large Language Model (LLM). With the LLM approach, an AI program is fed information from many sources including, but not limited to, web content, news articles, and books. However, ChatGPT has taken this one step further and included questions and responses from humans in the learning process. All this information is used to train a statistical model to predict words and sentences based on the context and the preceding text. As a result, many people view ChatGPT as the next iteration of predictive text on mobile phones.

Why are we talking about it?

ChatGPT is much more than just predictive text: ?since the model was trained using a very large amount of data, it can provide responses to a broad range of questions in a very confident manner using sentences that follow the syntax of the content consumed. As we all know from our roles in IT, the quality of output is determined by the quality of the input, and (at times) the information we read and consume is not (completely) correct and if published may not be redacted. As such, any model trained using incorrect information may provide incorrect responses.?Because this type of information that may not completely represent the complete truth is a small portion of the available content, ChatGPT is able to provide answers to a very broad range of questions in a confident manner. Since large amounts of data are available and used as the training ground, ChatGPT relies upon a treasure trove of information that is available at everyone’s fingertips. Additionally, ChatGPT uses complete sentences to provide the answers, and that is one of the reasons that this AI has gotten so much attention over the last 3 months since its inception.

What type of AI models are there?

Humanity’s contemporary quest began in 1960 as Computer Science began to take a foothold. Having said that, Humanity has flirted with AI ideas for the last two millennia. In general, AI models are classified in three major categories from an AI evolution perspective: Narrow AI, Broad AI, and General AI.

In the Narrow AI model, AI is trained to perform a very narrow and specific task. For example, IBM’s Watson AI played the game “Jeopardy!” in 2011 against humans. Several years down the road, Google’s DeepMind played the game of “Go” against humans. These cases illustrate Narrow AI, where the AI engine is developed and trained to play a specific game. The same AI model would not be able to go and play other games, for example. In other cases, AI was used to be a physician’s assistant for leukemia – but this instance of AI would not be able to become an expert neurosurgeon assistant.?These AI models can only deal with a very narrowly defined task, hence the name.

That brings us to Broad AI - around 2018, AI models evolved into Broad AI, which meant AI could now perform more broadly defined tasks which simply means AI was now capable of performing narrowly-defined tasks as well as tasks adjacent to those tasks. An AI performing a task of physician’s assistant for leukemia could now perform tasks for diagnosing skin cancer. This advancement helped solve multiple tasks under a single domain such as cancer identification and treatment.

Evolving even further, AI is evolving into General AI, whereby AI models begin showing human-like characteristics and capabilities. In this instance, AI can tackle multiple tasks under multiple domains just like a human would.

While Open AI’s ChatGPT shows human-like characteristics in its behavior, it is not quite human-like (but comes very close) and is far more capable than its predecessors.?You can ask it to write an essay on the US Civil War in less than 500 words, and it is able to do just that. IBM’s Watson AI can watch a movie and create a trailer for that movie just like a human would be able to. ?As AI continues to evolve, it will have human-like capabilities falling under the umbrella of General AI.

Does ChatGPT mean human skills are irrelevant and obsolete?

In recent days, it was discovered that CNET started using ChatGPT to create much of their content;?in about 2.5 months, starting in November, 2022, ChatGPT helped CNET author more than 70 articles. No readers nor the industry at large was able to figure out that the content appearing in these CNET articles was authored by ChatGPT and not by humans. That begs the question, “Are human skills obsolete as AI becomes more human like?”?The answer is no, not quite. Throughout technological advancements in history, this question has always been asked;?so far human skills have not become obsolete. While some human skills have been eliminated by technological advancements over the years, we do not feel that human skills will become completely obsolete. Humans have many more complex characteristics. For example, humans can ask very insightful questions. ChatGPT or other AI models are not capable of such capabilities.

Summary

While many of us are familiar with chatbots, the introduction of ChatGPT in November, 2022, has accelerated the interest in this next generation of AI that interacts in a conversational way. With the introduction of ChatGPT, AI has the potential to become a mainstream tool not just in the workforce, but also in everyone’s personal lives. With ChatGPT just three months old, we can only imagine what it will accomplish as a toddler!

Please note that the views expressed in here are that of the authors ( René J. Aerdts, Ph.D. and Hemang Davé ) and do not represent Kyndryl’s official positions.

As always, we would love to hear your feedback on this topic as it helps us all think outside of the box.