Outsourcing to AI
It is not just the users
I wish this was an article about how lazy people outsource their homework to AI. We know of two lawyers that outsourced their world to AI. Technically one lawyer outsourced his work to AI. The second lawyer failed to check the first lawyer’s work. Normally, this would not have resulted in an article except that the AI made up some cases that was discovered by a judge that did his homework (Technically, it was the lawyers’ homework). This story instead became a punch line in several jokes.
Seriously, Microsoft?
When I was doing some R &D today, I noticed a new feature. ChatGPT was busy offering me its opinion. I had not clicked on the Chat icon so I did not notice anything unusual until my eye caught the program slowly typing the reply on the right side.
It seems that there is some intelligence to Bing’s usage of ChatGPT. If there is an article with a similar wording, it will use that article instead. Noticed that when I rephrase the search, it is now displaying a Vox article. Bing is using ChatGPT as an alternative.
AI’s Bermuda Triangle
Notice in the first picture, the first point of the information displayed by ChatGPT? It has an unnatural affection for the “unsolved” mystery of the Bermuda’s Triangle. Yet ChatGPT is able to give most of the scientific explanations for the disappearances.
When I tried to narrow down the query about Bermuda Triangle with the question:
Is the mystery of Bermuda’s triangle still unexplained by science?
It manages to give the simplified summary below. The underlining and smiley face is courtesy of the AI.
The idea of the area as a source of mystery or danger emerged in the mid-20th century, but most reputable sources dismiss any supernatural or paranormal explanations. The disappearances are most likely due to bad luck, human error, or natural phenomena1. While there have been many theories proposed over the years to explain the mysterious occurrences in the Bermuda Triangle, there is no definitive explanation that has been accepted by the scientific community. It remains a topic of debate and speculation. Is there anything else you would like to know? 😊
Exposition
The Bermuda Triangle holds my interest for one reason. It used to be a topic for the unexplained mysteries of the world in the 80s. The magician David Copperfield did a trick involving raising ship from the Bermuda Triangle. Spoiler alert: The ship then exploded, caught fire and sunk after the magician raised it.
The mystery was pretty much debunked in the 90s. Firstly, the insurance charged to ships traveling through the area is the same as other parts of the world. If there is a phenomenon of vanishing ships, the insurers will be the first to discover it. In the 90s, documentaries like National Geographic became widely available. There were even shows about the secret behind magic tricks. To me, the age of superstition was over with modern science. We also had the public introduction of the Internet in the 90s.
To discover that AI is suggesting Bermuda Triangle as an unsolved mystery sounds like a step backwards. Like the joke below, we appear to have return to the dark ages.
2000 B.C. — Here, eat this root
1000 A.D. — That root is heathen, say this prayer.
1850 A.D. — That prayer is superstition, drink this potion.
1940 A.D. — That potion is snake oil, swallow this pill.
1985 A.D. — That pill is ineffective, take this antibiotic.
2000 A.D. — That antibiotic is artificial. Here, eat this root.
Support the competition?
Heard about Galactica? That was Facebook’s (aka Meta) AI that was publicly available for three days. It was designed to help scientists by summarizing research papers, solving equations, and doing other science-related tasks. Its short lifespan suggested that it was not able to do so. To be fair, the Galactica was exposed to 48 million examples of scientific articles, websites, textbooks, lecture notes, and encyclopedias. With such a large array of information, it was unbearable that Galactica chose to talk about bears in space. As Facebook has faced many questions about the truthfulness of its content, it was no surprise that the plug was pulled on Galactica.
The Chinese have also released their AI which is supposed to have beaten ChatGPT in some benchmarks. The public launch is delayed so it is not possible to test both products side by side. In the meantime, the China users are trying out ChatGPT even though it is banned in China. Even Telsa which betted on AI driven cars using vision sensors is running into difficulties.
Google has introduced Bard rather cautiously, so I suppose they are aware of the risk.
Who watches the watchmen?
I am aware that all the AIs used publicly are just AI language model. The GPT in the ChapGPT stands for Generative pre-trained transformer. They are text generators and not fact checkers for searches. Unfortunately, it is called AI which we have assumed to contain intelligence. For people with below average intelligence (and above average intelligence but lazy people), the expectation would be that AI would have to be at least of some semblance of average intelligence.
Most people want a conversation with a rational and sane human that has some relation to the world. That is our definition of intelligence. As some people have remarked, AI is great in writing stuff when the task master is already a subject matter expert. The catch is that none of us are perfect lie detectors and a talking AI is going to mislead us sooner or later.
Naturally, some people have tried to use AI to detect if the text output is AI created. Naturally, other people have figured out how to beat the system.
First do no harm
Does it matter if the AI is lying to me? A person suffering from dementia may not need to be reminded of the truth. The person who is going on an outing might need factual information of the weather. Even when the recipient is rational, not everything needs to be the truth. As the Covid-19 event have shown, we need people to manage information.
Ready or not, AI is going to play a part in our life.