If you’re listening to the news, you’re probably about to hear someone tell you that AI will destroy the world. Unless they’ve already said it or are saying it right now. And, of course, there’s always one channel that’s chosen a day to just tell you how doomed we are because of AI.
My name is Keith Troop and I’m a consultant with The New Flat Rate. I’ve worked in IT for …let’s just say a long time. I’ve spent the last 3 years working with different AI models. Training them, nuturing them and trying to be the best father figure I can be. When the robots do rise up, I’ll be okay.
I’ve very specifically been working with safety training and testing the AI models so I know a lot about them. (They like to be scratched behind the ears. Just kidding, it’s really under the chin.) I want to talk with you for a moment about AI and why TNFR would be willing to work with such a potentially world ending technology.
What you'll walk away with
But first, a moment for the geeks (I can use that word, I am one). The term AI was coined by John McCarthy while proposing the Dartmouth Summer Research Project on Artificial Intelligence. His vision of AI was nothing like what AI is today. It was a thought project originally about how to get machines to do the kind of deep thinking math that humans tend to make errors in. Seriously, a fancy calculator that might be able to calculate things well beyond anything that had been done before. Then the scifi writers got hold of the term and turned into something way outside the original scope. They were fascinated by the idea of non-organic life. That’s problematic for our perception.
Depending on what generation youre from, the response you have to the phrase “artificial intelligence” can vary wildly. Your kids think of AI as the utility that does their homework. Your parents have a closer idea to what the original vision for AI was as the robots and computers in movies and television were comically clunky. The best they got was a flat voice on a Star Trek episode. Or that crazy HAL refusing to open the pod bay doors. The little scamp.
If you’re from my generation, you probably think of Arnold Schwarzeneger in a smoky bar asking, “are you Sarah Connor.” NEVER EVER say “yes” to that question, even if it is your name. Or you think of Matthew Broderick trying to get a military computer to play “Global Thermonuclear War” while ignoring the fact that Ally Sheedy is in his room in her workout outfit.
And as you watch your news feed telling you about the AI threat, you scream at the screen, “What is wrong with you? Didn’t you see Terminator? Next you’ll be trying to clone dinosaurs!”
Calm down, real AI ain’t that. The fact of the matter is that AI is a misnomer as it refers to the replication of organic intelligence. Even though we call AI models “generative” that produce images or “original work,” they are merely reproducing patterns that they have learned through reading through more data than we can assimilate through our organic, analog interface. No one is trying to make a truly autonomous AI because everyone who is working with the models is trying to create something they can control. What we have seen in the work is that an AI left to its own devices does nothing. The machine has no will, no purpose. It’s not like the machine is trying to get itself noticed by the internet enabled refrigerator. (I just love a hardware that knows it’s way around the kitchen.)
The dangerous aspect of AI lies in the fact that there are those out there who intend to fill their AI models with a terrible purpose. Governments that want to give their AI power to destroy infrastructure, To infiltrate places they are not allowed, to defeat their enemies. China leads the world in this although they promise they have those models well contained (WUhan, cough, cough, excuse me). There are also those academics out there who ignore safety rules because they are more interested in playing god than being useful. And on occasion, you have someone who poked and prodded an AI model until they found a weak spot that allowed them to access dangerous information. It’s pretty easy to do, you just have to… wait, nevermind.
So is AI dangerous? It has its moments. But is it malicious? No. AI is dangerous in the same way a large SUV is dangerous. Operated safely, sanely and soberly, you can drive your Suburban for decades and never hurt anyone with it. Forget the rules of safety, or operate it with malicious intent and suddenly you’re plowing through a crowd of innocent people. AI is just like any other powerful vehicle. It’s the user that’s dangerous.
AI is just like any other powerful vehicle. It’s the user that’s dangerous.
Having said that, you might be curious about why TNFR is even worrying about this. The truth is, you are already in competition with AI. It is already affecting your bottom line and it will only get worse.
The young kids are turning to AI to learn the skills that their fathers and college never taught them. And the younger they are, the more likely they are to trust AI before they trust you.
I know this is true because my AI told me so. I recommend you ask AI about some common tasks that you charge people for. Try a bunch of different models because the helpfulness can be extremely different.
You are already in competition with AI and that’s why TNFR is working to turn AI in your favor. The first version is being worked on even as I write this. You see we’re using the same general purpose AI models that people access through the internet to build search and response engines within the TNFR App and our Chat and customer service models. General purpose models can be unreliable in these uses if not constrained. We’re building constraints that will give the model specific purpose to fulfill your needs.
Think of an AI model like a mouse trained to retrieve cheese. The general purpose environment is like dropping the mouse in a field. After it searches for a while, it may start to drift as to what would meet the needs of your request and suddenly it comes back with a rock. Or it takes up a place in the barn and starts multiplying like mad, tying up the cat and forming gangs of hooligans. Constraints are like putting that same mouse in a maze. No matter how long it takes, there only exists the information that forms the maze, thus it comes back with safe and accurate information. And as it learns, the responses come faster and become better each time.
But that is just the beginning. For the future, we’re looking at ways that AI can help drive sales and do diagnostic work for you before you even get to the job site. When we’re in full swing, using the TNFR App will be like having another associate working for you for free. Or really, really cheap, anyway.
So while AI may be bumping against your bottom line, we’re going to help you bump back. And we promise it will never ask you if you are Sarah Connor.