For the first time in history, the answer to the following questions of “what item would you bring to a deserted island” and “what item would you keep to start civilization over from scratch” is now the same thing. You would bring ChatGPT, or put more generally a large language model (llm) or generative AI model. At its core, these AI models condense the knowledge of the data it was trained on. It can acquire knowledge and know how to apply it, which is the definition of intelligence. This proves that the most important thing in our world is access to knowledge and how to use it, aka intelligence, and AI models are the best container for knowledge. Let’s dive deeper into why ChatGPT is the best item to bring to a deserted island. Before everyone gets their panties in a bunch about needing internet access for ChatGPT, let’s assume it’s an offline model running locally inside an indestructible cube that is powered by a hand crank and solar panels. Ok? Good. Let’s continue.
When you crash on the island, the AI would tell you what you need to do first to survive. Most likely how to build a shelter with the elements you have around you. Teach you how to get fresh water. Show you which mushrooms are poisonous and which make for a tasty treat. It could entertain you with stories. Even act as a friend for companionship in a lonely place. Having a physical tool like a hatchet would be nice, but it’s not going to help administer first aid to you when you accidentally cut yourself while chopping wood.
Imagine trying to bring other forms of knowledge to the island. Maybe a waterproof copy of a survival manual. That would definitely help you survive the first few nights and eventually thrive on the island. But only for your most basic needs. What if you found other people on the island? Maybe a potential partner you want to settle down with and start a family? The survival book is not going to help you deliver a child.
Instead of a survival book you could bring an entire encyclopedia. Either one of those old school physical books, or a print out of everything on Wikipedia. That could help you with a lot of things. But now your problem is knowing where to find the right knowledge to help you with new problems. Does that rash on your arm match with the description of a rare disease in the encyclopedia? Hard to say. You then find another tribe on the island that speaks a different language as you and they seem pretty pissed off. Scrolling through Wikipedia for 30 minutes can’t help you there. So it’s not enough to have the knowledge itself at hand, but easy ways to access exactly what you need, when you need it.
Easy access to knowledge used to be driven by search engines like Google and Bing. They could search through various information for you automatically. In a sense they condensed the entire internet into a text box to ask questions. But what came out on the other side are things that might be able to help, but not the exact knowledge you might need immediately or how to apply it to solve your problem. New AI services like ChatGPT can do just that. It knows most of what a search engine might know, but it can find the right answer immediately. It’s also able to create new information never thought possible, because it wasn’t in the original training data. Google search may not know what to do when a completely new event like a meteorite hits your island. But ChatGPT can reason and give you the knowledge and implementation to deal with the incident. It not only knows what, it also knows how. Knowledge is power. But being able to apply knowledge correctly, aka intelligence, is even more powerful.
The cost of intelligence with AI is quickly going to zero. What happens then? It feels like the two most important jobs going forward in an AI future is either building smarter AI models. Or finding the right data to feed them. By building smarter models you are improving the how (intelligence), and by getting better data you are improving the what (knowledge). The scary part is that AI models can now create their own synthetic training data, and techniques like reinforcement learning allow them to essential train themselves. So manual human intervention in these models might go away too. What’s left to do if the AI can train itself and get better every day?
The answer is to use the AI to make humans better. To increase their knowledge of the what and the how. Imagine a 30 year old accountant who has never camped outside in their life lands on the island with ChatGPT. They would most likely be able to get by with the help of the AI, but they would still struggle initially, maybe even perish from plain bad luck. Now replace that person with Bear Grylles, the famous British survival expert. I mean, the dudes name is literally “Bear”, how could he not survive in the wild? Combine him with ChatGPT and now he turns into a superhuman. He already knows how to survive, and paired with ChatGPT he can now build an island empire.
Having a smart AI is good, but having a smart AI combined with a smart human is even better. As these AI models get smarter, they should also enable humans to get smarter as well. If you want to start learning how to code today, you shouldn’t read a textbook on python. Instead have ChatGPT start teaching you. It’s a world class tutor available 24/7, tailored to your learning style, and is infinitely patient. It can write code for you, explain how it works, and then quiz you about it. Using AI to increase the collective knowledge and intelligence of every person in the world should be a top priority.
So many problems in our world today can partly be due to a lack of education. When using a college degree as a proxy for education, the problems in our society stick out. People who don’t have a college degree have more problems.
- Poverty: In 2019 people with a college degree earned $25,000 more than someone with only a high school diploma.
- Crime: Less than 4% of formerly incarcerated people have a college degree.
- Suicide: Men without a college degree are twice as likely to die from suicide compared to a college graduate.
Having a college degree is not a perfect measure of who has knowledge and who does not. But the above stats paint a bad picture for those without a degree. Soon everyone with an internet connection will have access to the best tutor in the world, essentially for free. AI tutors can teach billions of people at once. The only limiting factors will become tailoring the AI tutor to each persons wants and needs, and a person’s thirst for knowledge. There’s an old saying that goes: “The cure for boredom is curiosity. There is no cure for curiosity.” We can tailor the AI tutors to the learner, but creating the want inside of people to learn is the hard part. In a world of video games and TikTok, getting someone excited to learn about chemistry or to simply read a book will be a challenge.
This is where the multimodal nature of AI models can save the day. The AI can print out text, talk to you via voice, even create photos and videos. This unlocks limitless ways for an AI to teach you something. Someone may prefer a book like experience with an AI using chat, whereas another may learn best through a superhero movie that teaches you about greek mythology. Creating an AI tutor that a person can’t turn away from is one of the most important problems to work on. Imagine your 10 year old staying up all night because they don’t want to stop learning about the migratory patterns of birds, or how nuclear fusion works, instead of watching TikTok dance trends. This is a possible future. AI is up to the task.
AI models are not only positioned to make the world better, they are also well positioned to make humans better. Let’s go build one.