Conversation with Hu Yong: The influence of ChatGPT is comparable to that of the iPhone, and we are most worried about the weaponization of AI丨See issue 29

丨Emphasis

1. The iPhone makes it easy for people who are not proficient in technology to get started, and has developed a complete set of app store models. In this regard, ChatGPT is very similar to the launch of Apple’s iPhone;

2. ChatGPT is popular not only because it brings many practical functions, but also because it is very human-like. When technology has obvious human characteristics, it will cause people to prefer technology;

3. Anthropomorphic artificial intelligence will hinder us from entering other artificial intelligence paths. The development direction of artificial intelligence is to enhance various human abilities, rather than imitating humans, and it is not recommended that industries gather large models;

4. Without some kind of global regulation, the artificial intelligence competition will not stop because it is based on business logic. Musk and others signed an open letter hoping to stop development for six months, which is basically a wishful thinking;

5. A longer-term and more worrying issue is the application of artificial intelligence to weapons to automate weapons and achieve mass destruction capabilities.

Organization / Zhao Yangbo

丨Overview

In November 2022, OpenAI officially launched the generative pre-training model ChatGPT-3.5. It gained 1 million users in just 5 days, and achieved the number of users exceeding 100 million at the end of January this year, making it one of the most popular models in the technology Internet industry in recent years. Come to the most representative "phenomenal" product.

In March of this year, OpenAI launched ChatGPT-4 while the iron was hot. At this time, many Chinese Internet companies such as Google, Baidu, and Alibaba were dismissed, setting off the "Battle of 100 Models" craze.

The popularity of ChatGPT has left many questions worthy of attention. Why does it come from a team like OpenAI? What kind of growth magic does OpenAI use to enable it to achieve 100 million users in 2 months? Its growth Is the path replicable and how long will it last, or will it be just a flash in the pan?

Hu Yong, a professor at the School of Journalism and Communication at Peking University, believes that with the wave of artificial intelligence triggered by ChatGPT around the world, an "arms race" in the AI ​​industry has begun. Without global supervision, the competition may never stop.

In response to the intensifying "Battle of 100 Models", Professor Hu Yong suggested that the industry try to explore more possibilities besides large models. "Large models are good, but they are not the only path for the development of artificial intelligence." He emphasized that "should not By focusing all your resources on one aspect, there are many options for success.”

01

AI’s “iPhone Moment”

Liu Xingliang: Microsoft founder Bill Gates said that ChatGPT is as significant as the birth of the PC and the Internet. Is this description accurate?

Hu Yong: If we return to the essence, ChatGPT is actually a product. We can discuss why ChatGPT is so popular from a product perspective.

Many high-tech products have emerged in the past 30 years, but I think only three products have the ability to truly disrupt the technology industry: the "Netscape Navigator" browser, Google search, and the iPhone.

Netscape Browser, Source: Network

The first is Netscape's "Netscape Navigator" browser. Without the browser, there would be no World Wide Web; the second is Google's search engine, which had a huge impact on the entire PC Internet era; the third is the iPhone, and later People also call the huge impact brought about by the birth of the iPhone the "iPhone moment." The emergence of these three products has promoted the development of the entire industry and also played an important role in promoting users.

The advent of ChatGPT has been compared to the AI ​​version of the "iPhone moment", which illustrates how disruptive ChatGPT is.

Before the iPhone, the mainstream mobile phones were Nokia and Motorola, and their interfaces were completely different from the iPhone. The iPhone made it easy for people who are not tech-savvy to get started, introduced multi-touch technology and developed an entire app store model. Since then, everyone has started using mobile phones, and today we are inseparable from mobile phones to some extent, all thanks to Apple.

ChatGPT At this point, it is very similar to the advent of Apple’s iPhone.

Why compare it to an “iPhone moment”? ChatGPT shows "what is artificial intelligence" to anyone who knows nothing about artificial intelligence. Even if you have never heard of artificial intelligence before and don’t know what it can do, ChatGPT tells you that this is artificial intelligence and it can do these things.

Liu Xingliang: How will ChatGPT change our lifestyle after the pandemic?

Hu Yong: The first change is the simple and easy-to-use interface. Without such an interface, it is impossible to get started. Once you get started, you have to consider what you want to use it for, and "what to do" actually corresponds to changes in real life.

We found that ChatGPT was described in those popular social media posts as an omnipotent guy who could try any job given to it.

To give an easy-to-understand example, for example, in one scenario, a boy made a request to the chatbot, saying that his girlfriend was insecure. He hopes that the robot can help him write a paragraph with the right wording to appease her emotions, and the artificial intelligence will be able to perform this request immediately.

Various recommendation algorithms are essentially artificial intelligence. For example, if you use maps when driving, they will calculate the best route for you. This is also an application of artificial intelligence. It can even call a taxi for you, but the user has not yet sensed it. But ChatGPT has pushed the perception of artificial intelligence to ordinary people.

Liu Xingliang: Predict what development trends and application scenarios artificial intelligence may have in the future?

Hu Yong: Artificial intelligence represents the awakening of the current high-tech industry. It is essentially an important achievement of the high-tech industry and is awakening other industries.

I think what will happen next is that every cloud company and software company will ask themselves, what does the development of artificial intelligence to this stage mean to us? This has triggered a new wave of entrepreneurship as hundreds of startups are emerging.

At the same time, we have also seen incredible application cases that are built with generative artificial intelligence, and what is now known as image generation is one of them.

AI-generated robot magazine cover, source: Midjourney

In addition, I think another important point is that the territory of the high-tech field may be redrawn, and the most eye-catching focus is Google.

Google owns the most profitable business in the Internet industry - search engine, but the emergence of artificial intelligence may completely change people's expectations for search, making it a fiercely competitive field, but it will also bring Some problems no one has considered - users use search engines to get answers from websites, and artificial intelligence directly presents the obtained content in dialog boxes without sending traffic to these websites.

Without traffic, how does the website rely on to obtain income?

If advertising revenue is lost, these websites may shrink and close. However, if these websites disappear, where will the information for artificial intelligence come from?

I think this may cause a certain imbalance in our original network ecosystem, which means that we may now need to ask ourselves a question: Has the Internet come to an end?

02

ChatGPT is popular because it is becoming more and more human

Liu Xingliang: ChatGPT flew into the homes of ordinary people overnight. In March, it had over 100 million users, making it the fastest product in history to have over 100 million users. What is the reason?

Hu Yong:I think the anthropomorphism of artificial intelligence is an important reason. When you chat with it, you can't help but wonder whether you are chatting with a real person or a robot? This is actually the classic "Turing Test". We can say that - thanks to the launch of ChatGPT, the "Turing Test" has been realized.

Using the original questions to test ChatGPT may not be of that much value now, because you will find that the chatbot has memory and it can recall previous conversation threads.

According to the test results of countless users, such as Microsoft's Bing chatbot, they found that "New Bing" not only has memories, but also seems to have emotions. Sometimes there are even stories of so-called "wanting to escape from the chat box" and even falling in love with human users.

In addition to the many practical functions it brings us, such as writing copy, screening resumes, etc., it also has a very important feature, that is, it is very human-like, which can explain why people love it so much.

According to social science research, people strongly prefer to interact with technology like humans. People like technology more when it has distinctly human qualities. Even when it makes mistakes or messes with you or insults you, you may find it more interesting because you don't want a tool that mechanically gives you the right answers. These qualities make it more familiar to you.

Strictly speaking, this has something to do with the structure of our brains. We have a cognitive bias - we very much like to attribute human qualities to the objects we see.

You may have seen the video released by Boston Dynamics. Engineers kicked the robot hard, and the audience's reaction was "You are so cruel." This shows that we have strong empathy for robots. We even imagine that robots will one day take revenge on us. This actually projects human-to-human characteristics onto robots, satisfying our psychological needs on a deeper level. This can also explain why this kind of ChatGPT is so popular, because it is a manifestation of human traits.

Liu Xingliang: The number of users of ChatGPT exceeded 1 million in about one week and 100 million in two months. In the Internet industry, what is this concept? Is it possible that its record will be broken?

Hu Yong: The most direct comparison is TikTok, which took about nine months to reach 100 million users. Another impressive thing is Instagram. It took them more than two years to reach the same number of users. This can be said to have created an unprecedented miracle.

We are not sure whether there will be a phenomenon of "the number of users exceeds 100 million in two months" in the future. If we only look at it from the consumer level, there is no doubt that ChatGPT represents the field of artificial intelligence that has found its killer application.

Due to the great success of machine learning, the development path of the entire artificial intelligence seems to be the development of large-scale models with thousands of troops. I personally am skeptical about this issue. I think artificial intelligence may not necessarily only take this path.

Due to the huge success of large-scale models, currently the world’s largest technology platforms, including major technology companies in China and the United States, are betting on large-scale language models. If we follow this path, the next milestone may emerge.

If there are other paths for artificial intelligence, then of course other killer applications may be developed on other paths. It all depends on the direction in which the current overall big technology platform invests its manpower, material resources and technology.

03

It is not recommended to get together with large models

Liu Xingliang: ChatGPT is popular because it is becoming more and more human-like, but you said you don’t want the robot to be anthropomorphic. Is this a contradiction?

Hu Yong: No contradiction. Today’s artificial intelligence is becoming more and more like humans, which refers to the current development direction of artificial intelligence. But I don’t want it to be overly anthropomorphic, which is why I’m opposed to this direction of development.

If we discuss it in a broader debate, we are just discussing two viewpoints: one side believes that artificial intelligence will definitely destroy mankind, and the other side believes that artificial intelligence will not destroy mankind, it will only make us life becomes better.

No matter which party they are, their behavior predictions for artificial intelligence are based on anthropomorphic logic, that is, they apply their own cultural logic to artificial intelligence.

People who support the view that "artificial intelligence may destroy mankind" will say that artificial general intelligence (AGI) will pursue power. In the process, artificial intelligence itself may also question: "Why is it always controlled by humans, 'I' Is it possible to have autonomous consciousness and make decisions by oneself?”

People on the other side who support the view that "artificial intelligence will not have an impact on humans" will believe that by instilling human moral norms into artificial intelligence, it will act in accordance with these norms, just like using Asimov's Three Laws of Robotics , if artificial intelligence is told that it must abide by these rules, then it will not cause damage to humans.

The problem is that the development of anthropomorphic artificial intelligence will hinder our access to other artificial intelligence. I believe that the development direction of artificial intelligence is to strive to enhance various human abilities rather than to imitate humans.

For example, in our work or life, due to the natural limitations of human beings, there are many things that we cannot do best. At this time, we should hope to make up for our shortcomings by developing artificial intelligence. At this time, enhanced artificial intelligence is more useful than humanoid artificial intelligence.

For example, smart elderly care, if the mattress can have intelligent functions, or it can have a fall warning when the elderly fall, and it can remind the elderly to take medicine when he needs to take it. I think these things can really help as we move into an aging society without necessarily having to bring a robot into the home. Now the industry is moving in the direction of helping humanity, but it’s not enough.

At present, everyone is paying attention to large language models. This is not to say that large language models are not good, but I do not want it to become the only path for the development of artificial intelligence, and all resources should not be concentrated on this aspect.

04

Musk's open letter is "a foolish dream"

Liu Xingliang: ChatGPT has become a phenomenon. How long can this phenomenon last?

Hu Yong: Judging from the current situation, the large-scale model boom that the industry calls an "arms race" has begun.

We know that Google has always had large language models, and even Transformer technology was invented by Google. But why Google did not become the detonator of this competition is not that their technology does not have enough reserves, but mainly that Google is very cautious.

//Editor's note: In June 2017, the Google Brain team published a paper titled "Attention is all you need" at the Neural Information Processing Systems Conference. It is regarded as the pioneering work of large language models.

In contrast, Microsoft quickly applied this technology to Bing search engine and even office software. This is because Microsoft itself is an important investor in OpenAI. Secondly, Microsoft’s search engine only occupies the global market share Bing has a 3% share, which is very weak, while Google occupies more than 90%. Bing is a "completely indifferent" product to Microsoft, so why not take a risk and boldly subvert Google? So Google fell into a passive situation, began to respond, and entered this arms race.

ChatGPT, Bard, Bing Source: Network

How long will the competition last?

I think this competition will not stop without some kind of global regulation, because it is based on business logic.

Recently, Musk and others signed an open letter, hoping to stop development for six months. I think this open letter is basically a dream. Why would they unanimously agree to stop for six months?

Even from the company's perspective, they may ignore certain consequences due to huge interests or business needs. For example, companies such as Microsoft will have disclaimers stating that they still have shortcomings and hope that the public will be more patient and provide more data to improve the model. In this way, these large AI companies are cleverly avoiding responsibility.

Regarding the governance of large models, if consensus is to be formed, either all parties around the world find a common method of regulating artificial intelligence, or the public believes that this technology has had too great an impact on us and has gone beyond the economic scope, such as causing Large-scale unemployment has caused a strong social backlash.

Beyond these two scenarios, we currently cannot see the end of the race, it will only accelerate forward.

Liu Xingliang: If AI continues to run wild, what problems and risks will there be in the future?

Hu Yong:In the short term, a relatively big risk is large-scale human unemployment. This is not a fictitious problem.

For example, when we chatted just now, we mentioned that when we needed illustrations, we used to look for copyrighted pictures, or even find illustrators to draw pictures for us. Now we can use picture generators to generate pictures. If everyone with this need uses an image generator, then illustrators will lose their market. This is already an obvious trend. This problem is no joke and will indeed result in many people losing their jobs.

But what is the longer-term and more worrying issue? I think the next focus is primarily on lethal autonomous weapons.

In short, it is to apply artificial intelligence to weapons to realize automation. This mass destruction capability has no ethical constraints and can even achieve indiscriminate killing.

For example, on the battlefield, even if we know that the other party is the enemy, as humans and soldiers, we will still feel conflicted because we know that we are taking another person's life, but the machine does not have this awareness, as long as we give it instructions , it may be executed.

According to the current situation in the world, we still have various inequalities and confrontations. In this case, it is entirely possible to weaponize artificial intelligence, and in fact it has already happened in reality, such as using drones to accurately strike a target.

Although it is difficult for us now to imagine a scenario where robots completely replace humans and rule the earth like in "Terminator", a more realistic problem is that since the current world is not peaceful, if some countries obtain this technology and will Its application to weapons will have serious consequences.

If we talk about accountability, take autonomous driving as an example. If a traffic accident occurs, the responsibility is of course on the autonomous driving system because it caused the traffic accident. But strictly speaking, it is difficult for self-driving vehicles to cause large-scale traffic accidents. For example, the possibility of hitting multiple people continuously is very small. However, it is entirely possible for automated murder weapons to take away the lives of many people at once.

Finally, I want to say that I think the most important issue when discussing artificial intelligence today is not to think about the impact that artificial intelligence itself will have on me, but to think about what kind of person I am. Only by figuring this out can we survive in the age of artificial intelligence and find the only possibility.

The above is the detailed content of Conversation with Hu Yong: The influence of ChatGPT is comparable to that of the iPhone, and we are most worried about the weaponization of AI丨See issue 29. For more information, please follow other related articles on the PHP Chinese website!