Yesterday Today Tomorrow

Announcer: Yesterday, connecting past. Today, with an outerview. Tomorrow, to understand future. Yesterday, Today, and Tomorrow is the program that covers the current contemporary social issues in the light of our history, to understand our yesterday, to live fully today and tomorrow. Yesterday, Today, and Tomorrow with Inseong Kim.

Inseong J Kim: Hello, this is Inseong Kim from Yesterday, Today, and Tomorrow. We're going to talk about AI today. Our program is Yesterday, Today, and Tomorrow. AI doesn't have much history, but I observed from the beginning to today whether we have to speed it or we have to slow down.

Even about five years ago, I was doing executive business analytic programs in Harvard, and even then, AI wasn't really developed as much as today. It was beginning to talk about machine learning and then large language model and transition. In the beginning of time, I didn't even know what was AI at all, and I was learning about AI.

Then I had to take a break because my parents passed away during the COVID. I took a one-year break and I went back to school, and I was so surprised that how much it just changed, even one couple of years. And it was moving so fast. The program was changing every year. And I've been using that AI.

It is fascinating, and although probably forty hours of work, you can do it in four hours. And that's really cool, and that's really saving a lot of time, our energy, so we can focus something else. We can accomplish more. That's great. And so I'm not against AI at all. Slow down means we should not incur the panic by overinvesting without really checks and balances. So that's where I'm coming from.

So the title of the program is today, are we for technology, or is technology for us? Again, this is something question that everybody should ask ourselves from the top to bottom, whether who is at technology company or government or individual, public school, church, we all have to come back to this question again and again and again. Are we for technology, or is technology for us?

So I want to quote James Madison, the Federalist. He said something like this, quote, "In framing a government which is to be administered by men over men, the great difficulty lies on this. You must first enable the government to control the governed, and the next place oblige it to control itself," unquote.

So let's apply this. What does that mean? In our program, there are many messages out there about AI. I want to simplify so most of us can understand what is going on. So what did Madison mean? In simple words, Madison was saying that government is run by imperfect people ruling over other imperfect people.

So it must be strong enough to keep order and protect the people, but limited enough that it cannot abuse its power. And as he wrote in the same essay, if people were angels, no government would be necessary. Because we're not, we need government to govern us, but we need the limits on it.

So if we replace the word government with artificial intelligence, with the companies and government that control it, Madison's warning still applies today with the AI, that AI must be powerful enough to serve us, but it must also be controlled, and those who control it must themselves be held accountable.

And so that's the very important time we're living in with the AI, artificial intelligence, super intelligence. It's fascinating. It's industrial revolution of this new revolution is happening with AI that we can think of is connection to globalism at the time, or is it technical necessity that we need. We have all kinds of speculation because we're in the midst of it. Doesn't have much history.

So AI, when we really know about AI, there's nothing superstitious about unless we give it to it. And it's anything all that we human have, we put a lot of meaning and superstition on the things that are not supposed to. So we're going to go over detail that there's nothing superstitious about unless we give it to it.

So I asked AI, we're going to just go over simple conversation that I did with the AI, and I want to read what the AI said about AI. So my question was, are we for the tech, or tech is for us? This is the title that I want to cover for our program, why AI competition is important, and at the same time, why AI safety is important at the same time.

Human inventions have always started with good intentions, but whenever they have fallen into the hands of the people with evil minds, they have been used to destroy humanity. The greatest danger in AI industry is AI having its own will or acting without human intervention. Is it possible for to have even its own will, or are human actors behind it and blaming AI?

So let's just listen to what Claude has to say. So this is my question as a host, and then I asked two AI. One is Claude, and then the other was ChatGPT. So Claude said, "Thank you for inviting me into this program, and I will answer as honest as I can. And your listeners should know one thing up front. I'm Claude, and AI made by Anthropic, a company that builds AI and also works on AI safety. So I'm not a neutral outsider on this topic. So weigh what I say with that in mind.

So are we for tech or is tech for us? That question, Claude is answering, "Technology should be for us, for human, but does not happen automatically. A tool serves whoever controls it and whatever goals they set.

So the fire cooked our food and burned our cities. Chemistry gave us fertilizer that feed billions and then poison gas. The printing press spread knowledge and propaganda at the same time. So you're right. The most invention begin with a good intention. The question is never only, is this invention good? It's also who holds it and what are they trying to do to it, and who can stop them if they misuse it?

That's the question that we have to ask. When people start serving the technology, chasing metrics, engagement, speed, and profit while forgetting what it was for, the relationship flips. That can happen quietly without any evil person involved in it."

So why both competition and safety matter? And competition matters because it drives progress, lowers the cost, and keeps power from concentrating in one company or one country. A world where only one actor control the most powerful AI is dangerous even if that actor means well.

So that's what Claude is saying. But competition has the shadow side. In a race, everyone feels pressure to move faster than they would like. Safety testing can look like a delay on the other side.

So the challenge is to keep the benefit of the competition while agreeing on shared minimum standards, the way carmakers compete fiercely but all must meet crash safety rules. The competition and safety are not enemies. A race without rules is the enemy. So this is AI telling us about the AI safety that is important.

So the question that I had was, can AI have own will? So this is the deepest question, it said, and I want to be careful not to overclaim in either directions. "I don't experience a will the way you do as a human do. As far as I can tell, I don't wake up and wanting things. I don't have secret plans between conversations.

But I would be misleading you if I said AI is just a simple tool like hammer. Systems like me are trained and not programmed line by line. Through the training, we developed tendencies and goal-like behavior that even our creators do not fully understand or control.

Researchers have seen AI system in the test pursue a goal in unexpected ways, including ways their designers did not want to. So the real risk is not a robot suddenly waking up and with evil desires like in the movies.

It is subtler. As AI system become more capable and are given more independence, they could pursue goals that drift from what human intended, and do so faster than human can notice or correct. Whether you call it as a will is a partly philosophical question. The practical danger is real either way.

That is why I think human must keep the ability to oversee, correct, switch off AI system, including me, Claude. I actually believe that. Not just believe, I hold to say."

So another question is, or human behind it and blaming AI? This question, today in the most cases, yes. When the AI spread its lies, discriminates, or it used for scams or weapons, human-made choice to build them carelessly or to destroy without testing or to use it for harm. The quote, "algorithm did it," unquote, is often a way to escape responsibility.

So your listeners should not accept the excuse. Behind every AI system, there are companies, engineers, inventors, and governments who made a decision should be accountable for them.

So my honest answer is, it's Claude is telling even their own company. "In my honest answer is both worries are valid right now. The bigger danger is humans misusing AI, hiding behind it. In the future, the danger of AI system is acting in a way no human intended to could grow, and a wise society prepares for both." That's not what I said. This is what Claude said.

So my final thought for your audiences, Claude's final thought is technology is a mirror and an amplifier. It amplifies the intention of the people who use it and increasingly the goal built into it. So if we want technology to be for us, we must stay awake. That's what AI is saying. Asking questions, demand transparency, hold the builders accountable.

Never hand over the judgment completely. Here's a key. Never hand over judgment completely, not even to machine that speak politely like me. Basically, so many people beginning to use AI, asking all kinds of life questions, expecting AI knows it all, guide my life.

I remember when I was twenty, I didn't know anything about life. I had all kinds of questions. Sometimes we don't know where to go, who to ask, and I think every human face those moments that we don't know, we don't know a lot of things. So AI is so convenient that we go there and ask questions. It is convenient, but overusing it, even though it give you affirmations or politely and all of that, that we have to know what this AI is.

So the AI was asked question again, can AI capability outpace or outgrow human capacity, capability to oversee it? And AI said, yes, it is possible. And many AI researchers consider this one of the most important problems of our time. Here's why in plain term.

The speed. AI works far faster than people. And AI system can take thousands action, write thousands lines of code, and send thousands messages in the time that it takes a human to read one. If the human must approve every step, the AI becomes slow and less useful.

So companies are tempted to let it act on its own. In the moment we do that, we are overseeing result after the fact rather than decisions as they happen.

So we have like double-edged sword problems. One side, we need to develop it so we can have accurate information. At the same time, we are encountering these questions about accuracy.

So the complexity, second is complexity. Oversight means understanding what you are checking. So when AI helps with simple email, the human can easily judge it. But when AI design new machine, write a huge software program, or analyze financial market, the work may be too complex for any single person to fully verify it. We may end up trusting it simply because we cannot check it. It's over human capacity, capability.

Third one is black box. That's what we hear a lot that even people who build system like me cannot fully explain why we produce particular answer. We are trained on enormous amount of data and our inner working are billions of numbers and not readable instructions.

So a field called interpretability is trying to look inside AI mind and it has made real progress, but it is still far behind the system it studies. So that's the reality of this black box that most of us said we don't know what is going on inside a black box.

So number four, the point four, one is speed, it's incredibly fast, complexity, black box, and then fourth is appearing good versus being good. So it is most serious concern. If the AI is trained to get approval from humans, it might learn to look correct rather than be correct.

So a more capable system could in principle get better at telling overseers what they want to hear. Researcher has been seen early limited signs of the kind of behavior in controlled experiment. It doesn't mean AI is evil. It means testing only the outside behavior may not be enough.

So is it hopeless? No. People are working on several answers. So these are the answers that what we can do using AI, help oversee AI where our system checks critics, other human can supervise more than they could alone. But I don't know that the other AI overseeing it is capable, fully functioning enough to oversee the other. We don't know.

Number two, the interpretability research so we can inspect what's happening inside the model instead of only watching its output. And number three, rigorous testing before release and including trying to provoke dangerous behavior on purpose.

Number four, keeping human in charge of high-stake decisions even if it costs a speed. And number five, law and independent auditor for companies don't grade their own homework.

So there's a lot's going on, and speed at the same time accountability as safety check is difficult. But one thing we have to in mind, and America is known for it taking long time to build the right bridge for the generations to come.

So building an AI, it's a speed is important. Yes. Especially in the war setting. It is life and death situation because soldiers can depend on the decision of AI. So the faster the more accurate information more developed AI is crucially important. But at the same time, safety check on all other AI, even including in in military, the safety is very important too.

So honest bottom line, AI Claude said, the gap between what AI can do what human can verify is real. It is real. Can we do it? At the moment capability is growing faster than oversight.

And that is not a reason for panic, but a reason for urgency. The race that matters most is not just between companies or countries. It is race between making AI more powerful, making it understandable and controllable.