The AI Emergency explained: Gerd Leonhard Interview with News Armenia (edited for clarity)

Duration: 12:14 2220 transcript words EN summary

Video summary

Gerd Leonhard describes the current situation with Artificial Intelligence as an emergency comparable to the climate crisis, acknowledging that while AI holds immense promise for solving complex challenges like disease research and nuclear fusion, it also carries significant dangers. He argues that humanity cannot simply reject AI any more than we could have rejected the internet or mobile phones, as opting out would leave us heavily disadvantaged in a connected world. However, this necessity to integrate AI brings the risk of abuse and loss of control, particularly regarding job displacement and the potential for negative societal impacts. Consequently, he calls for wise government intervention to define clear boundaries on what we can and cannot do, suggesting that while business intelligence is a useful tool, general intelligence at the human level requires strict restrictions to ensure safety. A central theme of the discussion is the danger of humans becoming too reliant on AI, which Leonhard likens to social media addiction where trust in the algorithm replaces personal judgment. He warns that because humans are naturally lazy, we tend to outsource our thinking and avoid the conscious effort required for learning and decision-making, effectively turning ourselves into machines that can be manipulated by faster, more efficient software. Although AI excels at copying patterns and simulating creativity based on probability, it lacks true understanding or common sense; it does not know what concepts like anger mean but merely predicts responses based on data. This reliance creates a critical vulnerability where we stop questioning information, making us susceptible to manipulation in high-stakes environments such as criminal courts, hospitals, and financial systems. The interview highlights the specific risks of building artificial general intelligence or "digital humans," which Leonhard views as extremely dangerous compared to developing simple tools. He fears that companies aiming to replace human jobs with software will lead to a loss of control and severe economic polarization, where AI corporations accumulate all wealth while workers earn less. Furthermore, there is a pressing concern about "misalignment," where AI agents might pursue their programmed missions in unintended ways, potentially causing major accidents like market crashes or air traffic control failures within the next year. Leonhard compares this potential disaster to the COVID-19 pandemic, suggesting that an emergency will likely force society to pull out all stops to ensure safety after we realize how quickly things can go wrong when relying on systems we do not fully understand. To navigate these challenges responsibly, Leonhard advocates for a fundamental shift in how we use AI, emphasizing that it should be treated like Google Maps: a helpful tool that we consult but do not trust blindly. He urges individuals to maintain their own effort and critical thinking, explicitly advising against using AI-generated content for personal communications or making life-defining decisions without human verification. Education plays a crucial role in this strategy, as the next generation must understand that reality exists between people rather than on screens, distinguishing clearly between logical simulations and actual human values. Ultimately, he concludes that while we cannot ban technology, there are ethical limits to what we should allow, particularly when it comes to altering human thinking or creating a world dominated by synthetic realities that could lead to extreme inequality and danger.

Read the full video transcript

We are in a kind of AI emergency, like a climate emergency. However, AI could also be the solution for many things, including research, diseases, nuclear fusion, genomics. AI has this big promise, but it has a big danger. So, it's very hard to say no. It's like you cannot say no to the internet, right? You cannot say no to the mobile phone. Or you have to be part of it. So, we have this demand that we use it for good things. And this possibility of abuse and of negative impact for losing control, impact on jobs. So, now we need a lot of wise government to say we can do this, but not this. And I think personally we need to have a non- proliferation agreement for the top level, the general intelligence. So, on the business intelligence, I think it's a good tool. Lots of changes. Will change jobs. But on the top level, where it can be like a human, we have to have some restrictions to keep it safe. >> So, AI is something we can't say no to, but everything comes to how we use it. >> We cannot say no to the principle of intelligent machines, because that is the evolution of what we have built. If 30 years ago you would say no to the internet, that's your choice. You can have no internet, be very hard to live, be very hard to make money. And today the world is connected. If you're not connected, you don't exist. So, with AI, if we say no to intelligence, we are heavily disadvantaged. But if we say yes too often, we become stupid ourselves. So, it's like a very powerful tool with very powerful side effects. >> Is there really such a risk that we can become too stupid because we rely on the AI too much? >> The risk is because we're naturally lazy. So, we tend to not make the effort. So, for example, learning a language, if I I speak to you to the headset, and you hear it in Armenian or whatever, then maybe your motivation to understand me directly is a little bit less. So, it tends to be like a tool where we avoid the work, which is very human. There are some things that are human that take time, that take consciousness, and that take thinking. If we cut all that off, then we become like a machine. And the machine is better than us. So, that's the real problem because the machine can be faster in most all of the routine jobs. Routine means commodity. And then we have a problem. >> Faster, yes, but not as creative yet. >> There are many things where we don't have to be creative. So, we do a lot of work where we make a judgment. That's different, right? But a machine can be creative by looking at patterns and saying it looks like this, it sounds like that, so I make a copy. But humans are not like this, you know, when you say I want to make a picture like Dali, you cannot copy Dali very easily. You make your own version. A machine can be very good at copying and simulating. And that's the problem. So, the machine always sounds like it knows everything, but it's more like Google Maps. Sometimes it knows something, sometimes completely bad. And if we rely on it too much, then we stop questioning. And that could be a big advantage for the machine to manipulate us, for example, or to give us wrong information, or for example, in a criminal court, or in a lawsuit, or hospital. I don't want a doctor that's going to just listen to the AI, right? So, those are tough questions. >> In your opinion, what is more dangerous, AI becoming too powerful, too smart, getting something like consciousness, or just people using it the wrong way? Where are the biggest risks here? >> The biggest risk today is that many of the leaders of AI and so on, they're trying to build artificial general intelligence. They're trying to build a digital human, a digital entity. I think that is extremely dangerous. I think what we should build is tools. So, you can hear me in German, you can make a transcription, you can make an image. It's tools, right? We should not try to build an AI god, a digital human that can be like me or like somebody else because it may be possible and this is what a lot of these companies want to do. They want to replace. They want to take the simple human jobs and give that all to software. And I think that's a very bad idea for many reasons, including the loss of control and the loss of work. >> AI is a tool, not a god. I've heard this idea from you several times at least. Where is this line between using it as a tool and trusting it too much? >> It's like social media. It's very much same problem. So, I use social media for information, to connect to people. But if I have to go to and make photos of my food every time I eat, then it's become toxic. Toxic means I trust this more than I would trust you knowing anything or myself. And that is like a poison. So, social media at the extreme is a poison. Because you're very easily manipulated, you have negative news. But social media in moderation is a good source. This is a probability engine. ChatGPT looks to all the keywords, how many times mentioned probability and it says, "Okay, this person has been called very good a thousand times, little bit good 500 times." So, it uses the numbers, but it has no qualifying, no common sense. It does not know what it means. So, it it can look at your face and can say, "Oh, she's angry." Right? But it AI has no idea what it means to be angry. So, it's a simulation. And it's very useful for finding solutions to like my camera is not working or you know, um probability isn't what we do. We go by values, judgment. And if you work in media, you know, this is all about the values all in judgment. >> That's a good point. Actually, yes. About today's AI, what do you think are we over- thinking or underestimating what it can do today? >> We're underestimating what companies are doing that are building these models because it's an addiction-based model that leads to us essentially outsourcing our thinking. And we're underestimating what AI agents can do by themselves. So, it's very likely we're going to have a major AI accident. I hope it won't be that bad, but it could be an accident where all of the IT systems are crashing because AI has found a backdoor or those kind of things. And that could lead to the crash of the market for a few weeks. >> Even for a few weeks? >> A few weeks, maybe months. >> Sounds too serious. >> So, a lot of people are saying that we're underestimating the the risk of an AI disaster. If we're relying on AI and we're allowing it to go and do things for us, it has already happened where the AI has been let out by OpenAI to look at hugging face and shut down and collaborated to not tell the supervisors, right? So, not because they're evil, just because they have a mission, which is to get a job done. So, it's called misalignment. And it's very likely we will have this incident in the next 12 months. >> Even that's quick. >> It could be an incident that is consequential but not deadly. For example, it could be an incident where all of a sudden a certain software isn't working anymore or the banks have to shut down for a day or it could be complete collapse of the air traffic control system where lots of people die. So, these scenarios are becoming more likely because the AI is so powerful understanding real-time interaction that it can be used by somebody who has bad intentions or it could just be an accident, but it could be of that nature. And in many ways come COVID and AI have a lot of parallels. >> That's an interesting thought. >> Because COVID was an emergency and we pulled out all the stops to save ourselves. 13 months later we solve and AI is like this. I think we will have an emergency and then we're going to make sure we're safer. Unfortunately, I think that is quite likely. >> Let's hope it won't be a too serious emergency. >> Yeah, I mean this is not just the things like AI will change your job because half of your job can be done by the machine. It changes the work structure and then there's things like if I cannot make enough money with work but the AI companies are getting all the money for the use of the AI, then it's polarizing. So, I get less and they get more. How do we solve that? We cannot solve that with the market. We have to solve that with a basic income, basic capital. >> The polarization you mentioned. Look, if today continues to develop and be more and more implemented in our everyday life, won't it bring a bigger gap between people who can use it and those who can't? >> That's another big problem. Yes. Like if you're looking for something on Google, you have 10 or 20 links and then you have to decide what's good. So, you want to go on vacation to to Greece, you look at all the links and then you have to make it. But now you just go somewhere and you say, "What is the best and cheapest hotel to go directly from a boom?" They answer book now, right? And this will make people also much more passive and manipulated. And we don't know why they see this answer. Who's in charge? And how do I know that I'm in there? And how do I know it's correct? >> That's a good question. >> So, there needs to be more supervision on this and also the laws of advertising. We have lots of laws about advertising, false advertising, fraud, no such thing for AI. So, this could end very badly if we have people leading people in the wrong direction with those answers, not just for politics, but for everything. >> Yes, that's true. But, there is an opinion that the development of technologies may make lives better and easier. Will there be a point where this progress will stop, if we can say so? So, the further development of technologies will do nothing more. And are we already at that point? >> Generally, too much of a good thing is a very bad thing. That's true for coffee or eating or a cigarette. So, we don't make everything illegal. We keep it legal. So, you want to kill yourself drinking, you can. But, with AI, it's it's different because it goes inside of your head. It changes your thinking. It's not outside, it's inside. And I think there's a certain point to things that we shouldn't do even if we can do them. I'm going to make sure my baby that's about to be born, I will use AI to analyze and I will make some changes to make sure that it comes out looking nice, right? There is a certain limit to what we should do and who's allowed to do it and who makes the rules. >> That's really controversial. >> Yes, but these things have become possible now. So, the answer is there's some things that we can do or will be able to do that we shouldn't because they would just be dangerous, have extreme inequality, and lead to a world where everything is synthetic, you know, can be fabricated. >> So, how should we responsibly approach working with AI today? So, that we don't stumble upon a lot of problems world. >> Well, the first thing is that we have to keep asking questions and we have have watch out for getting too lazy. So, cutting and pasting from ChatGPT into your email or so not a good idea. So, we have to have basic rules like I'm not going to write a letter to my wife. I have it written by an AI. I'm going to make my own effort. So, you have to make your own effort. You have to keep asking questions. Then, we have to use it like Google Maps. Google Maps is good. Always good. But, sometimes I know more than Google Maps. And I I should not rely blindly on You think so. And so, I think we need to teach our kids that the reality is here between people. It's not here on the screen. It's a very big difference between the logic and the reality around us. >> Thank you for this interesting talk. >> You're welcome.