VIDEO TRANSCRIPT:

SPEAKER:

Artificial intelligence. It's all around us. Alexa, Siri, and now ChatGPT are all part of our lives. And while there are a lot of things to discuss about the future of AI specifically and what it all means, there's one big topic of conversation. Well, as ChatGPT grows its resume, it could very well come after your job. There's a real danger that the evolutions in AI could cost some humans their jobs in the near future. You wrote the code, you'll be replaced. Neural networks will do your work faster and for free. Some people are worried that AI is going to take over our jobs. So, ChatGPT, friend or foe? Are humans becoming obsolete? Are robots taking over the world? Let's dig into that.

SPEAKER:

While everyone is talking about AI and ChatGPT, sometimes it's hard to admit that we don't really know what it is. It's alive! It's alive! It's alive! It's alive! So, to start with, here's a remedial lesson. Artificial intelligence, or AI, is a type of technology that allows machines to learn and make decisions like humans. It's like giving a computer a brain so it can think and learn like we do. ChatGPT is a type of AI called a language model, which means it's really good at understanding and generating human language. It's like talking to a computer, like you would talk to a friend. You can ask it questions, tell it to do things, or just chat with it, like you would with a friend. And because ChatGPT has been trained on a lot of different types of text, it can understand and respond to a wide variety of topics. It is information from the past that helps us better understand the present. How much is it capable of? Well, everything I just explained was written by ChatGPT itself. And in fact, I'm allowed to use it here, as long as I tell you that ChatGPT wrote it. OpenAI, the company behind ChatGPT, suggests this specific disclaimer. I'm not going to read it all, but that's it. They also have a specific list of prohibited uses, including the expected stuff, like using ChatGPT to conduct illegal activity, but also prohibits using it to write political speeches or dating profiles. Women everywhere, thank you.

SPEAKER:

And AI technology is getting better, faster. According to the MIT Tech Review, ChatGPT became 60% less likely to make stuff up when it upgraded from version 3 to 4. Part of that came from humans interacting with the AI and correcting it. That's the thing. At least for now, AI still needs humans. Here's how ChatGPT itself puts it. AI can't exist without humans. That's because AI is trained on data that humans create and curate. For example, if we want an AI to be able to recognize cats in pictures, we need to show it lots of pictures of cats and tell it that those are cats. Without humans providing that data, the AI wouldn't be able to learn. The same is true for ChatGPT. It depends on humans to provide it with the data it needs to learn. That's why ChatGPT is often trained on large data sets of text that humans have created, like books or news articles. So, yeah, the world still needs humans to produce good AI, not just engineers and software developers, but also everyday users themselves who help train the intelligence just by interacting with the tool.

SPEAKER:

But there's one more thing that AI needs humans for, prompting it. Your AI output is only as good as your AI input, or what prompt the human user gives the AI tech. And there are good prompts and bad prompts. The art of coming up with the best prompt to generate what you want is called prompt engineering. It is the fast, reliable, and tireless performance of a variety of arithmetic and logical operations that give the computer its great utility and power. And there's nuance to this, nuance that humans can tease out. It's a skill set, just like any other computer-based skill set. Job postings with prompt engineers with six-figure salaries are popping up across the country. On popular gig economy site Fiverr, you can hire a prompt engineer for anywhere from $5 to $500 per prompt, depending on their experience and complexity.

SPEAKER:

But for our purposes today, let's assume you're doing your own work with ChatGPT and creating your own prompts. How can you be successful? Well, here's a few tips.

To start with, clear your AI system's memory. This is easy to do in ChatGPT and other AI tech. It's important because, in most cases, your AI can remember your entire conversation and might try to make connections with previous topics, even if you've moved on. So start with a clean slate.
Next, it can be extremely helpful to give your AI a role to play when answering you. We do this in real life in a way. If you're trying to decide what model of car to buy, for example, you carefully choose who to ask for advice. If you ask the used car salesman or your mom or the neighbor down the street, they're probably all going to give you different answers. So if you were going to ask ChatGPT for that answer, tell them first who they are and what kind of advice you want. And that need for context applies to all aspects of the prompt. This is not a case where less is more. Don't boil it down. The more context you give in your prompt, the better the output.
Consider answering some of these questions in your prompt. Who are you? What is your goal for the project? What is your ideal audience? Age? Education level? And be clear about what constraints you want to place, too. What output would you like? A table? A blog post? A video script about whether or not AI is going to take over the world? And soon the R20 was capable of all the thought processes and functions of a man. How many words should the answer be, roughly?
There's some more complicated ways to engineer a prompt, too. For example, you can chain prompts if you're working on a larger piece of writing. So first you'd ask the AI to outline an essay, for example. Then ask it to write the first paragraph. Then ask for the next, and so on. Or are you working on something kind of specific that the AI might not necessarily know about? You can teach it first in your prompt. So the best way to do this is to give examples. This is oversimplified, but you could tell the AI in your prompt that Sally likes red and Bruce likes blue. And then ask which would be more likely to like maroon or turquoise. You've taught the AI Sally and Bruce's preferences, so it would be able to answer that question.

SPEAKER:

And finally, it's not just prompt engineering that still needs people. Humans are needed after the AI produces its output. Either to tweak the prompt if it's not quite right, or to clean up the text. That could mean just injecting a little more creativity, or editing the text to add specific points, or, importantly, fact-checking. Earlier editions of ChatGPT had a habit of making things up, ranging from statistics to whole legal cases and academic citations. And while each subsequent edition of ChatGPT is getting more accurate, it's still not perfect, so it's something to keep an eye on.

SPEAKER:

All of this is to say, artificial intelligence is just that, artificial. As the Harvard Business Review puts it, AI should augment human intelligence, not replace it. And since AI is still dependent on humans to build it, teach it, prompt it, and maintain it, it has some very human pitfalls, too. To start with, bias and discrimination. In a weird, the call is coming from inside the house moment, I asked ChatGPT itself about bias, and here's what it said. Since AI is only as good as the data it's trained on, it's possible for AI to reinforce existing biases or discrimination. That's why it's important for developers to be mindful of the data they use and make sure that it's diverse and representative of all people. One way to combat bias in AI is, you guessed it, make the humans building it, using it, and maintaining it also more diverse and representative of all people. Again, the solution is people. He's a robot. Without you, what could he do?

SPEAKER:

So, yes, I think it's fair to say that the AI revolution is coming, or probably it's already upon us. But that doesn't mean human beings are going to become obsolete. The real AI revolution is figuring out how we humans can work alongside artificial intelligence in the future. There are a lot of ethical concerns to consider here, but here's a few to get started.

Number one, is this AI stealing other people's work? This is especially prevalent in generative art AI. You know, the kind of thing where you say, create an image of an elephant riding on a chihuahua's back. In a lot of cases, those programs are actually training their AI on existing artwork, which means that elements from their style or colors or shading are taken from other artists. That is stealing. So the key question there is, what data set or set of images is the AI trained on? It's not necessarily a question of not using the AI, but rather making sure that you're not helping to steal work. That information is often in the fine print on an AI's website. Artists and writers themselves can even go on a new website, haveibeentrained.com, to check if their own art or work has been used to train AI.
Another ethical concern is disclosing that a given work was created with help of AI. Most AI platforms, like ChatGPT, require that users disclose the use of AI. Of course, how do they enforce that? They don't. But if you're using AI, just let folks know.
Finally, put the time and effort into vetting what artificial intelligence programs you are using personally, or your company and institution is using overall. For example, as this Harvard Business Review article recommends, if you're considering implementing an AI program in your HR department, ask how specific AI prevents against discrimination.