Lift Up Your Hearts!

Trust, but verify.

Wikipedia informs me that President Ronald Reagan's trademark phrase was a Russian proverb (доверяй, но проверяй, or doveryai, no proveryai) taught to him by Suzanne Massie, a scholar of Russian history, who advised him, "The Russians like to talk in proverbs. It would be nice of you to know a few. You are an actor—you can learn them very quickly." Reagan eagerly adopted the phrase and used it frequently, especially when talking about American-Russian negotiations during the Cold War.

Given Reagan's success in dealing with a highly intelligent and well-informed entity that had the potential to do either much good or devastating harm—or both—we should probably heed his advice when negotiating the tangled risks and benefits of Artificial Intelligence.

We can't live, much less prosper, without trust. We are born with the instinct to trust our mothers, and if we live in a healthy home that spreads to include fathers, other family, friends, teachers, doctors, pastors, government officials—our need to trust the world around us goes very high, wide, and deep. To buy an egg from the grocery store requires trusting with our health a long chain of other people responsible for its journey from chicken to plate. That's why we have customs, and rules, and laws, and government inspectors tasked with ensuring that the process works. But quis custodiet ipsos custodes? Who will verify that our trust is justified?

We will. We must. The buck stops with us.

Because we cannot verify everything we depend on, and must live as if the world is in the main trustworthy, we must also be alert and aware and never forget that betrayal, whether malicious or unintentional, is always a possibility.

What on earth does this have to do with Artificial Intelligence?

I've been writing frequently about my AI Adventures, and the amazing things I've been able to do with its help. I'm having a happy time on this adventure, but the warning at the bottom of each screen, "Claude is AI and can make mistakes. Please double-check responses" is no joke. I cannot do anything about the mid-level worries ("Is AI going to take over my work and eliminate my job?") and certainly the high-level worries ("Is AI going to take over the world and eliminate humanity?") are above my pay grade. But periodically checking on the accuracy of an answer is within my reach.

Here's a small, recent example.

I've been using Gemini to help me identify places and pictures from our recent trip, and it's no secret I've been impressed. But Gemini indeed "can make mistakes."

I uploaded a photo of Raphael's Portrait of Agnolo Doni that I had taken at the Pitti Palace. Gemini correctly identified it, and provided numerous details, which ironically time and knowledge constraints keep me from verifying. But when it told me that the portrait was hanging in the Uffizi, alarm bells rang. I was certain I'd taken that photo at the Pitti, and by uploading the photos just before and after the painting in question, with their timestamps, I convinced Gemini that the painting was indeed where I said.

How did that happen? Gemini's information was not so much wrong as out of date. For a long time the painting had hung in the Palatine Gallery of the Palazzo Pitti. But at some point it was given special honor at the Uffizi and displayed there together with Raphael's portrait of Doni's wife and Michelangelo's Doni Tondo, which had been commissioned to celebrate their marriage. However, thanks to the big renovation project, Portrait of Agnolo Doni made its way back home, at least for now.

Trust your own experience, and verify regularly.