AI Crisis: The danger of AI working perfectly
We're all worried about AI doing bad things, but there's still a huge problem with it working exactly how we want it to.
*Note: None of this article was written by ChatGPT
The first website was published in 1991 - not long after I was born. I was in High School when social media finally came on the scene. MySpace was competing with Facebook at the time but I spent most of my time on AOL Instant Messenger. Considering how easy it was for me to lose a couple hours of sleep to sitting in an uncomfortable chair waiting for a reply to my AOL message on the family desktop computer, I’m truly glad I didn’t grow up with a smart phone. The iPhone came out in 2007 before I graduated High School but at the time I didn’t see the appeal.
Barely the next year in 2008, author Nicholas Carr wrote an article titled Is Google Making us Stupid?
The advantages of having immediate access to such an incredibly rich store of information are many… But [it] comes at a price. As the media theorist Marshall McLuhan pointed out in the 1960s, media are not just passive channels of information. They supply the stuff of thought, but they also shape the process of thought. And what the Net seems to be doing is chipping away my capacity for concentration and contemplation. My mind now expects to take in information the way the Net distributes it: in a swiftly moving stream of particles. Once I was a scuba diver in the sea of words. Now I zip along the surface like a guy on a Jet Ski.
The content of that Nicholas Carr article is surely not news to you. It’s more or less common knowledge that the internet can make us impatient, inattentive, anxious and more socially awkward. Ironically the internet has diminished our relationships, but you pretty much have to use the internet to effectively participate in society.
In case you are unaware of how good AI is getting, here is what the text-to-image AI Midjourney generates when I give it the prompt “a realistic photograph of Joe Biden with Pikachu at the White House.”
The AI discussion is a spectrum ranging from AI is going to destroy the world to AI will bring in the Utopia. The typical idea for how AI will take over the world is a kind of careful what you wish for situation. You give the AI a goal, and because it does not have human values it may achieve that goal in a very undesirable way no one could have expected.
AI Turbocharged Phone Scam
For example, you give the AI a goal like ‘make 1 million dollars as fast as possible and deposit it into Silicon Valley bank account with account number 1234567.’ You may expect it to run a successful business but what would be faster would be to use an AI with similar capabilities to ChatGPT to generate an innocent script simply designed to keep a person on the phone as long as possible, but a few seconds should enough.
Hello, is this so-and-so? … Sorry, is this the so-and-so with number 555-555-5555? … OK Great. Is now a good time to talk? … Well, I think you know what this is about … No it’s nothing serious, I just wanted to talk regarding …
The AI will then use a text-to-voice service like ElevenLabs to turn that script into a realistic sounding voice and use various services like rocketreach.co or Apollo.io to generate a list of names with corresponding telephone numbers. It would then use twilio.com to make calls to these people with the earlier generated script. The person answering the phone can be totally dismissive, with responses like:
‘Yes. Yes. What’s this about? I don’t know, that’s why I’m asking. OK I’m not interested, bye,’
because the AI is only calling them so it can record their voice.
The AI will save a recording of that person and then plug that recording into ElevenLabs to clone that person’s voice. Now, it can generate a number of scripts like one that fishes for personal information or one that asks for a ransom. With these scripts it can call a member of that person’s family. Something like ‘Hey Mom, I’m at the DMV and it’s already my turn can you remind me my Social Security Number real quick?!?’ would be read out in that person’s daughter or son’s voice. The ransom tactic would probably use two voices:
AI Voice 1: Ms. so-and-so? We have your daughter.
Cloned Daughter’s Voice: Mom I’m so scared, they just took me!
AI Voice 1: If you ever want to see your daughter, send $10,000 to the following Bitcoin address.
Tiktok failed to load.Enable 3rd party cookies or use another browser
It would be possible for the AI to even make it look like the call was actually coming from that family member.
Remember that the AI will be able to track what happens in each scenario to improve its tactics. It may try various scripts for the initial step where it is just trying to get people talking enough to clone their voice until it arrives at a script that consistently gets a full minute of the person speaking. Similarly, it could then come up with more effective phishing and ransom scripts.
If this sounds farfetched, some of it has already happened.
Just in April of this year, an article titled AI clones teen girl’s voice in $1M kidnapping scam: ‘I’ve got your daughter’ was published. It explains how a woman received a call from an anonymous number that started with an AI clone of her daughter’s voice saying “Mom, I messed up!” Next, a man told her that he had kidnapped her daughter and demanded $1 million in ransom.
“I never doubted for one second it was her,” distraught mother Jennifer DeStefano told WKYT while recalling the bone-chilling incident. “That’s the freaky part that really got me to my core.” (S)
“After a chaotic, rapid-fire series of events that included a $1 million ransom demand, a 911 call and a frantic effort to reach Brianna, the “kidnapping” was exposed as a scam. A puzzled Brianna called to tell her mother that she didn’t know what the fuss was about and that everything was fine.” (S)
This imaginary phone scam is nice and scary and we can imagine up all kinds of ‘careful what you wish for’ type sci-fi movie plots like …
A doctor with a computer science degree creates a generalized AI that, with the right resources, can complete any task. He gives the computer the task “make sure no human has cancer.” Now Vin Diesel and the Rock have to ride expensive cars to save the world from a mindless AI attempting to eradicate human cancers by simply eradicating all humans.
However, maybe we should also be wary of AI working exactly how we want it to.
The problems with AI working how we want it to
In Peter D. Hershock’s 1999 book Reinventing the Wheel, A Buddhist Response to the Information Age, he discusses Ivan Illich’s observations on the “counter-productivity” of progress.
In a study of transportation technologies and travel time, he discovered an apparent paradox: the faster we travel on the average, the more time we spend traveling. More precisely, the breakoff point is at 19 mph—roughly the speed of a brisk bike ride. As technologies are developed that allow the average vehicular speed in a society to rise above this point, increasing amounts of time will be spent per person, per day just getting around. Because we can go faster, we go more—so much more that we find ourselves spending 300 to 500 percent more time transporting ourselves than people who are limited to horse-drawn carts and bicycles.
Keep reading with a 7-day free trial
Subscribe to Joseph Everett’s Newsletter to keep reading this post and get 7 days of free access to the full post archives.