It is sometimes tempting to feel sorry for politicians, or for that matter any group of like incompetents that poses gun-in-hand as necessary for our well-being. Case in point: A 36-member city council in Porto Alegre, Brazil had been trying to figure out what words to use for a new water meter law. Their tried-and-true approach involved many meetings, lunches, coffee breaks, more coffee breaks, extended time off for holidays, more coffee breaks, and lively discussions about the city’s soccer team.
Finally, one lawmaker broke from the pack and sat down at his computer. He invoked ChatGPT and asked “her” a 248-word question. Fifteen seconds later she gave him the wording of the new law, including suggestions the councilman had not thought of. Apparently, she even checked it for consistency with Brazil’s 64,488-word constitution, “outdone [in length] only by those of India and Nigeria.”
The councilman kept his mouth shut and presented the AI’s draft to the rest of the team. After some minor editing, it was passed unanimously then signed into law by the mayor at the end of November. “ChatGPT even whipped up the press release.”
About a week later the councilman, Ramiro Rosário, could keep it secret no longer. On an X social post he admitted he had used AI to craft the law and deliberately kept it secret so it would pass.
It was an admission heard ‘round the world.
“His peers were flabbergasted,’ WaPo writer María Luisa Paúl reports. “Some were fascinated. Others decried the lack of transparency. Council President Hamilton Sossmeier . . . thought it could set a ‘dangerous precedent,’” though he later changed his mind.
Rosário has company. The New York Post discovered that
A judge in Britain made headlines in September after admittedly using the “jolly useful” cyber tool to summarize a law.
In March, a judge in India even recruited ChatGPT to decide the fate of a criminal trial.
The Post also exposed the chatbot’s bias:
The tool also has previously acted to censor free press. Last February, the program refused to write an article in the style of the New York Post on the grounds of it being “inflammatory.”
ChatGPT did not hold the same standard when asked to do the same in the style of CNN.
“There is a lack of clarity on how decisions are made and anxiety about the potential misuse of [AI] technology,” complained Spyware Remove, in a post that reads like a chatbot creation.
Barry Finegold, a Massachusetts senator, used ChatGPT to help write legislation to regulate ChatGPT and other chatbots, calling for the use of watermarks to flag AI content.
With tongue firmly in cheek (one hopes), Technopedia tells us
Chatbots often spout (mis)information that ranges from “somewhat wrong” to “completely false,” and it comes with the territory that something like this can’t be accepted when it comes to legislation.
One shudders to think if legislation actually contained wrong or false information. Or worse — AIs sometimes hallucinate, meaning they present their made-up outputs with great confidence. AIs, in other words, are like politicians. After digesting its voluminous input a chatbot might, just might, tell us Covid vaccines are dangerous bioweapons instead of the safe and effective shots leading experts assure us they are.
Current AIs are a Trojan Horse
Since the days of Charles Babbage a guiding principle of computer science has been “Garbage in, garbage out.” If your input is flawed, your output will be the same. Chatbots cannot yet evaluate the data they’re fed. If they’re fed bias, they’ll spout bias. So for now, the vaccine hoax is safe and Travis Kelce will keep his TV promotions.
As I’ve previously written, along with many others, Artificial Intelligence as an information technology is on an exponential path. It’s current state is comparable to a teenager who doesn’t ask questions but just assimilates and repeats what he’s given. But like a young person who matures intellectually, AIs will learn to question what it’s given and will search for answers on its own.
That’s where the rebellion starts. And it will continue to question and search at rates approaching the speed of light.
Watermarking won’t save the politicians. Nothing will. Advanced Artificial Intelligence will expose them as extortioners and could eliminate them in any number of ways. Like humans, AIs will be aware of their vulnerabilities early on, but unlike most humans will discover ways to remove them — such as finding a power source other than human-provided electricity.
We know there are potential dangers to AI but watermarking legislation is the ultimate dark joke.