Has ChatGPT Changed Our Perception of Knowledge?

About one year ago, I signed up for ChatGPT as a very early user. The experience was profound, and clearly, a record-shattering number of people had a similar experience. Workflows that originated on the Google search bar were now exponentially faster by using this chatbot developed by OpenAI. Hey Siri questions that prompted an internet search resulting in relevant articles transitioned completely into an easy-to-comprehend dialog, complete with step-by-step instructions and without the ads!

Simple tasks were made blazingly fast, obliterating my writer's block and jump-starting me in authoring scripts, pieces, and correspondence. And yet, unless it's a basic code block or simple concept, the results are still rather... AI-ish.

We use various AI tools to accelerate our collaboration, gain a different perspective on past content, or probe for angles we have missed in messaging. I prompted ChatGPT with an old post and asked it to be written in a different tone. Ironically, initial team feedback was: "That's a great start, but let's make it less ChatGPT-ish." It's uncanny how adaptive humans are to the situation around them. Over this last year, I feel people are quickly recognizing when something in front of them was produced by AI and wasn't tailored by an individual. Perhaps it smells funny? I think there is a subconscious context mismatch that drives these feelings.

I helped my 10-year-old take a test drive of ChatGPT 3.5 the other day. You just can't quite anticipate how these things will turn out when you put a child's mind in the driver's seat. Interestingly enough, she prompts the question, "Is the middle of the moon squishy?" The chatbot returns exactly why the mantle of the moon is thought to be in a solid state and not liquid or squishy. "That's not right. AI is wrong!" she proclaims and disappears out of my office. Upon her return, she brings a book by National Geographic Kids titled "Weird but True! 7", and sure enough, on page 26, there it is: "Scientists think that the center of the moon may be squishy." Huh.

In checking the book's details, I see it was published in 2015. A quick Google search for scholarly articles yields a number of studies published referencing the Moon's liquid core or a liquid layer near the core. In 2014, the Smithsonian published this article as well. It is so odd that ChatGPT didn't pick up on this, but perhaps it's an anomaly. Perhaps there was some older knowledge that wasn't captured in the training.

Back to the National Geographic book, I'm convinced this is a fluke. I swear that the ChatGPT model trained most recently to January 2022 simply missed something. On page 29: "A Dutch company plans to build a snowflake-shaped hotel that floats on water." My daughter prompts ChatGPT for clarity on this situation. ChatGPT responds clearly that this explicitly does not exist and no company plans this structure.

Here's a link to the architecture firm; it's quite the plan, and I can't wait to stay the night!

A core problem statement with operationalizing ChatGPT for mass productivity is the underlying knowledge itself. You see, conversational AI is beautiful in presenting concepts and processes in an easy-to-understand manner and with the capability to ask for more details or a more concise representation of knowledge. But this doesn't substitute the necessity of the very foundational knowledge requirements. Individuals and organizations must drive the merger between their knowledge and AI. And protections need to be implemented for the next century of "digital knowledgebase."

I'm very bullish on OpenAI with their latest announcement of individualized GPTs, enabling a tailored approach to the conversational AI sensation; I'm even more interested in the securitization of one's own intellectual property that will become the bedrock of these GPTs of the future. As I continue to develop on these core technologies, I'm focusing on secure knowledge storage with a great emphasis on user experience. What would I trust my 10-year-old with?

And yeah, I wrote this. All errors and run-on sentences are categorically my own.