Insights from my chat with ChatGPT
I asked ChatGPT to explain artificial intelligence to me, but I can’t tell you what its response was without first recounting how I got to that point.
I had attended a conference of state legislators where it was clear that the legislators were very skeptical of how technology companies were developing AI products and whether adequate public safety considerations and data privacy protections were reflected in them.
At the same time, many of the legislators conceded that their rudimentary understanding of AI technology impeded their ability to advance appropriate public policies. How are policymakers supposed to propose legislation for something they don’t fully get?
It quickly occurred to me as an employee of the American Association for the Advancement of Science (AAAS), one of the largest multidisciplinary scientific societies, and working for its Science is US program, we could play a role in helping legislators, among others in the scientific community, gain a fundamental understanding of AI.
If only it were so easy. While gathering the latest information to adequately define AI in layperson terms, how it is being used and why regulation is needed, I realized a lot of the research that existed was very technical and filled with jargon. A basic “AI 101” rundown didn’t exist.
I consulted experts within our network who universally said what I didn’t expect—there is no established definition for the emerging and controversial technology. AI is both a technological concept and a category of tools powered by advanced mathematical models and data. It is also more accurate to refer to “AI systems” instead of simply “AI” because the technology does not operate in a vacuum.
A member of our Science is US steering committee at the Institute of Electrical and Electronics Engineers offered an interesting cooking analogy as a starting point to understand the many moving parts of AI systems. We can think of inputs such as data and prompts as ingredients to a recipe; processes such as the different algorithms and programmers equivalent to the corresponding steps and measurements a chef follows; and the outputs of predictions and responses to prompts as the finished culinary delight. And just as chefs taste test their newest creations, adding a pinch of this or that as needed, AI developers must establish a feedback loop of their own to tweak the next generations of their products as necessary.
As I continued to scour the internet for other definitions of basic AI terms, it occurred that I had neglected to go straight to the source—an AI chatbot. Surely, ChatGPT could tell me about itself in understandable terms, right?
ChatGPT (3.5) sees AI as a “very smart assistant” that is capable of continuously learning. But it can only perform its duties if it is fed with the necessary information to work with, making both data and humans indispensable to its processing prowess.
State legislators, however, don’t see AI systems as just helpful, innocuous super-smart computer programs. This disconnect is what we originally set out to resolve with our own AI 101 factsheet.
Past technological advancements in biotechnology, cloning and nuclear energy were met with unease at first. Public acceptance grew as our collective understanding of them increased.
Developers of AI systems must learn a lesson from those earlier achievements. If they don’t, public policy makers will have little to go on except for misperceptions and even fear as they create needed legislative and regulatory frameworks. At that point, creators will not be the chefs in the AI kitchen, they’ll simply be choices on the menu.