Looking Back and Looking Forward: AI, Trust but Verify

(Looking Back and Looking Forward takes a look at the articles and posts I found interesting from the previous week, along with reflections about how the trends they point to might shape my thinking about education, technology, and culture.)

My wife asked me to help her with some research on gaining followers for her social media account and it seemed like another good opportunity to use the latest version of ChatGPT. I typed in my prompt and within a few seconds has a nice list of recommended tips/strategies. Based on that information, I created a follow-up prompt asking for assistance with creating good captions.

Both answers were helpful and easy to understand. But were they accurate? Easy enough to check in this case, and two Google searches later, along with scanning a few articles (total of 10 minutes), I was satisfied that ChatGPT had provided solid information.

To be clear, this “fact-checking” is not something reserved for questionable or suspect Internet sources. It’s pretty much SOP for me when it comes to researching information and resources. The practice goes back decades, to an English professor who was fond of saying, “Just because it’s printed doesn’t mean it’s true!”

In other words, trust but verify.

I suppose that’s why I wasn’t concerned when Wikipedia came on the scene and the debates raged about its validity as an information source.

I treated it like any other information source. Trust but verify.

I take the same approach to media outlets that are not aligned with my thinking, as well as to those that are. Trust but verify.

So, when news of ChatGPT-4 broke this week, nothing really changed for me. I have used it previously for certain types of information gathering but always do a quick fact check to ensure the information is accurate. And, by the way, even allowing for fact-checking, using ChapGPT saves me quite a bit of time.

There is plenty that has changed about ChatGPT with the new update, however. For example, it now “exhibits human-level performance” on professional benchmarks, such as Bar Exams and the GRE. It has proven reliable enough to be incorporated as a learning tool in such high-profile platforms as Khan Academy (see here for information on Khanmigo) and Duolingo (see here for more information about Duolingo Max). The same generative AI technology is also now part of Grammarly’s new offering, GrammarlyGo. Heck, even scholarly publishers are finding ways to use ChatGPT in their work.

Does this mean ChatGPT and other generative AI platforms are always correct and/or free from mistakes or misinformation? As I’ve said before, trust but verify.

There are, after all, some challenges inherent in large language models, including the use of latent spaces and large data sets. that cause them to provide inaccurate data or to “hallucinate.”

As Harold Jarche points out in his post on Microfoft’s Co-Pilot for Office 365, it “can be a powerful performance support tool. But as the demo shows, there will be mistakes so that any Copilot output has to be checked by a human.” Co-pilots, he points out, require a pilot, and generative AI is no substitute for real human experience. In other words, trust but verify.

If you haven’t already been trying our different generative AI tools, now is the time to start. If nothing else, it will help you understand why Sam Altman, co-founder of ChatGPT parent OpenAI, is both scared about the jobs AI will likely eliminate and excited about the new employment opportunities it could create. opportunities that maximize human experience and insights.

From an educational perspective, there is growing enthusiasm for a new generation of tutor bots and personalized learning facilitated by AI.

Also on the education front, it’s worth noting that undergraduate degree completion falls for first time in a decade. In fact, undergraduate credentials earned hit four-year low in 2021-22 academic year. While some will be quick to point out that this is related to a growing misalignment of higher education design and market/student needs (jobs), others continue to argue that higher education institutions need to be about more than professional preparation (jobs).

Finally, while everything AI seems to be going gangbusters, it seems that other technologies aren’t doing so well. To wit, tablets and PCs to are predicted to see double-digit declines in 2023, and AR/VR headsets are also in a downward spiral.

Have a great week, everybody!

Looking Back and Looking Forward: AI, Trust but Verify

Further Reading

Higher Education

K-12

Workforce

Online Learning, Learning Design, and Education Technology

Technology and Culture