Introduction: David George, partner of a16z Growth, and Chris Dixon, partner of a16z crypto, had a conversation to explore their vision for the new Internet, including decentralized AI infrastructure with cryptocurrency; initiating network effects, AI will become the native media form of this era, etc. This conversation also explored why the original business model of the Internet is decomposing, and how the new Internet can introduce a completely new business model for creators.
How technology has evolved
David George: You now spend most of your time focusing on the encryption field. What do you think about the relationship between encryption technology and AI?
Chris Dixon: My macro view is that technology waves tend to come in pairs or triplets. Fifteen years ago, mobile Internet, social networking, and cloud computing were the three major trends. Mobile Internet allowed the number of users with computing devices to grow from hundreds of millions to billions; social networking was the killer application that attracted users; and cloud computing was the infrastructure that supported it all. These three are interdependent and are indispensable. At the time, people argued about which one was better, but it turned out that they are all important.
David George: Yes, they are all necessary.
Chris Dixon: I think AI, cryptography, and new devices (like robots, self-driving cars, and VR) are the three most interesting trends right now. They complement each other and grow together. Cryptography is something new (and that’s what my book is about) and it’s a whole new way to architect the Internet and build networks. It has some unique properties that make things possible that weren’t possible before. When a lot of people think of crypto, they think of Bitcoin or meme coins. But for me and a lot of professionals who really understand crypto, crypto is much more than that. It has a lot of intersections with AI. One of the most basic ways to combine it is to use crypto architecture to build AI systems. We’ve invested a lot in this direction.
We have a core discussion internally about whether AI will be controlled by a few large companies or managed by the broader community. The first question is: Is AI open source? I am really shocked by how closed the AI field has become. Ten years ago, all AI research was open and published in papers. But then, the industry suddenly became closed. They claim that this is for security reasons, but I think it is for their own competitive advantage. Fortunately, there are still some open source projects, such as Llama, Flux, and Mistral. But I am a little worried that this open source model is a bit fragile because many projects do not disclose their model weights. Is this really open source? Some models are open source, but their data pipeline is not. Is it really freely reproducible? They may change the model tomorrow, and there is nothing you can do. These AI models are getting better every month, but if they are no longer cutting-edge, I don’t know what to do.
David George: At least for now, AI is very dependent on large companies.
How Cryptocurrency and AI Interact
Chris Dixon: Some of the projects we invest in focus on building a decentralized Internet service architecture suitable for the AI ecosystem. For example, there is a project called Jensen that is building a decentralized computing resource network. Its model is similar to Airbnb, allowing users to submit computing tasks and allocate them to idle computing resources around the world, thereby optimizing the supply and demand of computing power. This network is like an economic ledger that manages the supply and demand of computing resources.
Another example is Story Protocol, which is a new way to register intellectual property. Lets say youre a creator, you can register an image, video, or music onto the blockchain, and the blockchain will record the media and all the rights to it. It uses existing copyright law to clearly define who owns the copyright. In this way, anyone can use this content under the premise of complying with the agreement, and anyone can come and you might say, You can use this remix, you can create derivative works, but you have to pay me 10% of the revenue.
David George: ...or any ratio.
Chris Dixon: In the blockchain you can set the terms and create an open market. But in the current market, you can only contact the company yourself and try to negotiate. This leads to people either stealing content or simply not using it, or only large companies can make copyright deals. For example, OpenAI paid Shutterstock $100 million. The blockchain creates a broad democratic resource where small creators can set their own terms.
A core advantage of cryptography is composability. Open source software is successful in large part because it allows developers to combine and innovate on existing modules. Linux is a good example, which has grown from almost 0% market share in the 1990s to more than 90% of the server market today because of its composability. People contribute to the system (even if it is small) to make it better. This is also like Wikipedia as a knowledge integration system.
Back to Story Protocol, it also allows creative content to be freely combined like Lego bricks. For example, someone creates a character, another person writes a story, and another person uses AI to generate animation. You can create a new superhero universe, and as long as the funds flow back, everyone can get a share in the end.
David George: The key to this model is that the flow of funds is transparent and fair.
Chris Dixon: This way creators can use AI tools to improve efficiency while getting financial rewards instead of being used for free. This is a great vision - it incentivizes people to use these new tools while providing an economic model. In our investment, we often think about how to find new economic models for creative workers in an AI-driven world. This is the area that excites me most at the intersection of AI+Crypto.
David George: In the past, social platforms received 100% of advertising revenue, and creators could only rely on traffic to monetize. What we hope to see is a new system where creators can set prices and trade freely. This will drive more innovation.
David George: Because the economic incentives are aligned.
Chris Dixon: Building on this, we are seeing more of this crowdsourcing approach to AI. From a data perspective, AI needs more data. The breakthrough in encryption technology is that it can design new incentive systems. The key is how do we use these systems to collect more AI training data? Data can be used as input for AI, for model evaluation, or for other purposes. This is similar to what Scale AI does, but the difference is that we want to do it in a decentralized way, rather than having a centralized company control the entire process.
One of the projects we invested in is WorldCoin, which was co-founded by Sam Altman. The core idea is that in a world where AI can forge human identities and content, we need a way to prove that a person is real, and the best way to do that is through blockchain, using cryptographic technology to complete identity verification. WorldCoin designed an incentive mechanism for users to register and obtain identity verification, such as an orb scanner to scan the iris, but this practice has caused some controversy. Now they offer other ways, such as identity verification through passports. Once you complete the identity verification, you can get a cryptographic credential on the blockchain, which can be used in various services.
A simple application scenario is verification (CAPTCHA). Current verification codes have become so complex that even humans may not be able to pass them easily. Compared with these cumbersome anti-fraud systems, we can use encrypted verification. Users can receive an encrypted code to prove that they are human, and then add an additional layer of verification on this basis. This is another interesting intersection.
There are still many opportunities for decentralized AI at the infrastructure level, such as disassembling centralized AI systems to make them decentralized at both the code and service levels. There are also some new possibilities, such as machine-to-machine payments. And so on.
I think the most exciting part is exploring new business models in the AI era, especially business models for creators.
Breaking the Internet’s Economic Contract
David George: You pointed out to me right after the ChatGPT moment, “Hey, we might be breaking the contract of the Internet,” and I thought that was a very interesting question.
Chris Dixon: Theres a chapter about this in the book, near the end. I call it the New Contract. If you think about incentive systems, one of the main reasons the internet was successful is that it has a very clever incentive system. How do you get 5 billion people to join a system without a central authority? Its because of the incentives of the internet.
ChatGPT shows people that the economic contract of the Internet may be broken. Over the past 20 years, the Internet has formed an implicit economic contract: search engines and social platforms obtain the rights to content, and in return, creators can get traffic. For example, travel websites, food websites, illustrations, etc. will let Google crawl content in exchange for search traffic. This model supports the development of the Internet. But now AI directly generates content, users don’t even need to click on links, and Google no longer needs to distribute traffic to websites. In this way, the creator’s source of income is cut off, and the original economic model of the Internet has also collapsed.
In the past, Google would also distribute part of the traffic. For example, when users searched for questions, Google would display a summary, but would still guide users to visit the website for more information. But later, Google began to intercept traffic. For example, for StackOverflow content, Google directly displayed the answer in the search results instead of asking users to visit the original website. This caused the traffic of many websites to drop and the monetization ability was affected. Google is also doing similar things in the travel, catering and other industries (such as Yelp), and will even give priority to displaying its own content instead of independent creators content. Although these problems have existed for a long time, the AI era has made this problem more serious.
But if AI can directly generate illustrations, recipes, and travel advice, users don’t need to visit those content websites at all. This may be a better experience for users, but it is a devastating blow to content creators. In the future, we may only have a few AI giants left, and the original independent websites and creators will lose their living space.
This is the question we need to think about: Can the Internet in the AI era still support innovation and entrepreneurship? If we don’t solve this problem, the Internet may become like the television industry in the 1970s, with only a few giants controlling all content. This is not the Internet future we want.
So how do new websites emerge? How do new things get created? We haven’t really thought this through.
I don’t think I have the only answer, and the solution to this problem doesn’t necessarily have to rely on cryptography. But we need to recognize that this is destroying the original incentive mechanism of the Internet. Secondly, we need to think: Is this a good thing? I think not. We need to find the right solution - should we create new incentive mechanisms?
This is why I have been focusing on investing and thinking about new incentive systems, such as projects like Story Protocol. We need to explore new ways to superimpose new economic structures on top of existing systems to ensure that the Internet can continue to innovate and develop.
From mobile internet, social networking and cloud computing to encryption, AI and hardware
David George: One thing you talked about is the emergence of three technology products at the same time - generative AI, cryptocurrency, and new hardware platforms. How do you view the combination of these three?
Chris Dixon: The analogy is of course mobile, social and cloud computing. In the last wave, they promoted each other and jointly promoted the development of the Internet. We are already seeing some of this combination today.
Now, we are in the midst of another wave of technology, this time with AI, cryptography, and new hardware at the core, such as robotics, self-driving cars, and VR. These technologies are not independent of each other, but complement each other to form a new ecosystem. New hardware devices, such as AR and VR glasses, rely on AI to provide better interactive experiences, such as smart assistants like the movie Her. Self-driving cars, Teslas robotics technology, and various humanoid robot projects are also deploying AI technology in physical environments to apply it to the real world. And cryptography provides a new way for decentralized networks to support these AI applications. So one area I am interested in is DPIN - decentralized physical infrastructure. The most prominent example is Helium, a community-owned, crowdsourced telecommunications network project that is competing with traditional operators such as Verizon and ATT. Helium has designed an incentive mechanism so that anyone can build a node at home to support the network. These nodes act like wireless signal transmitters, and hundreds of thousands of people have installed them across the United States.
Now, Helium has also launched network services, and the price is much cheaper than Verizon-only $20 per month, while Verizon charges $70. This is mainly because Heliums network is built by the community and does not need to invest tens of billions of dollars to build infrastructure like traditional telecommunications companies.
How to use encryption technology to trigger network effects
Chris Dixon: Encryption technology is very advantageous in solving the cold start problem.
Many network effect projects face a challenge in the early stages: how to attract enough users to make the network really work?
For example, Helium is built and operated by the community. But if there are only 10 nodes, it obviously wont work. The establishment of network effects is a chicken-and-egg problem. If a new social network has only 10 people, it will not be attractive to new users. But if it already has 1 million users, the value of new users joining will increase significantly.
The unique thing about crypto is that it can drive the formation of network effects by incentivizing early adopters through token economics. Helium is just one example, and other areas, such as climate data, autonomous driving data, electric vehicle charging stations, decentralized maps, and even scientific research, can all be built in a similar way.
Is AI the frosting or the sugar?
David George: Marc gave me a metaphor that I really like: Is AI icing or sugar? If AI is just icing, then the existing industry giants will win because they can simply add an AI chatbot to their existing products and continue to dominate the market using their existing distribution channels, sales capabilities, and customer relationships. But if AI is sugar, that is, it is the core ingredient, then you cant just add it in, but need to build the entire product from scratch. In this case, the AI field is more likely to be dominated by emerging companies.
At present, we have not seen a clear answer. The more a product follows the traditional model (such as simply using AI to enhance the original business), the more it will benefit industry giants rather than startups.
Chris Dixon: You can look at this from the perspective of Clayton Christensen. He proposed the concepts of disruptive innovation and sustaining innovation. Many people misunderstand the meaning of disruptive innovation. It is just new technology and means that this innovation does not fit the business model of existing companies. This is why even the largest companies have a hard time dealing with real disruptive innovation because their core customers do not need it.
This is consistent with the “icing vs. sugar” concept proposed by Marc - if AI is just the “icing” of existing products, then industry giants will naturally dominate; but if AI completely changes the business model, the situation will be completely different.
For example, todays database market is basically dominated by traditional relational databases (SQL), while AI may bring a completely different computing architecture and even completely subvert the concept of databases. If AI is only used to optimize SQL databases, it is just icing and poses no threat to existing companies. But if AI completely changes the way data is stored and retrieved, making traditional databases meaningless, then it is sugar and will subvert the entire industry.
David George: We havent seen such cases yet. I only see the impact on price (such as cheaper AI services), but this is not enough to bring about industry disruption.
Chris Dixon: Yes, thats the second level of the problem. I usually use a framework to analyze the implementation process of these emerging technologies, but before talking about this, lets talk about consumer AI. At present, I think there are no products in the field of consumer AI that truly have network effects. Although AI chatbots such as Claude and ChatGPT have been successful, they have not formed a strong network effect. Users can change AI tools at any time, with almost no switching costs, which makes them easy to fall into price competition.
David George: We once believed that data network effects would become the moat of AI products.
Chris Dixon: Indeed, data network effect is a concept that exists in theory, but it is often not that strong in practice. Many people believe that the more AI training data there is, the better the model will be, and the more users will rely on it, thus forming a barrier. But the reality is that the data generated by individual users actually makes very little incremental contribution to AI training. In other words, the usage data of a single user will not significantly improve the capabilities of AI, so it is difficult to form a strong network effect. This leads to a major risk for AI companies: market competition will intensify and price wars are inevitable. Although AI products such as ChatGPT currently have strong brand awareness, the question is how to avoid entering into pure price competition?
If the switching cost between different AI tools is low, then the final market competition is likely to evolve into a price war and all companies will be forced to lower their prices to attract users. In this case, these AI companies will not be dominant companies.
David George: So do startups still have a chance?
Chris Dixon: If AI is only used to improve existing products, it will be difficult for startups to compete with large companies. But if AI is used as the core architecture to create a completely new business model, it will be different. At present, many AI consumer applications we see, such as face-changing and image enhancement, although they became popular in a short period of time, were quickly copied by TikTok or Instagram, and eventually the startups lost their competitive advantage. If an AI product does not have a network effect, then once its function can be copied, it will be difficult for it to maintain competitiveness in the long run. This is why, if you want to build a truly successful AI startup, you must find an entry point that can form a network effect, rather than just providing a function.
Come for the tools, stay for the network
Chris Dixon: A classic user growth strategy is: Come because of the tool first, stay because of the network. That is, many users initially use a product because of a certain tool, but the reason they stay is the network effect. For example, early Photoshop users may just want an image editing tool, but later, they find that the Photoshop ecosystem is very strong, so they become long-term users. The rise of social networks is similar. Many users initially joined because of a certain function (such as a friend address book), but ultimately stayed because of social relationship links. AI can also adopt a similar strategy. For example, AI-generated image tools can be used as an entry point, but what should be formed in the end is a complete AI creative community, not just a tool software.
Imitation technology and original technology
Chris Dixon: Before we get into that, it’s important to discuss how major technologies are rolled out in phases. New technologies typically evolve in two phases:
• Imitation stage: New technology imitates old technology to make it easier for users to accept it.
• Native stage: New technologies create completely different new experiences.
Then there is the third phase: broader changes brought about by new technologies. For example, after the invention of the car, we built other infrastructure such as highways, suburbs, and trucks.
For example, early web pages were like an electronic magazine, all content was static and not very different. This imitation stage may exist for ten or even twenty years, such as Mosaic in 1993 to YouTube and Facebook around 2005.
But as the internet grew, we began to see native internet products, such as social media, search engines, and online video platforms, that had no offline corresponding business models.
AI is still in the skeuomorphic stage. The AI applications we see are mainly replacing human labor, such as AI customer service, AI writing assistants, etc. But the real AI revolution will appear in AI-native products, such as AI-generated game worlds, AI-generated interactive content, etc. Its like when photography first appeared, cultural critics worried about its impact on art. Walter Benjamins famous article The Work of Art in the Age of Mechanical Reproduction asked what would happen to artists when anyone can take a photo.
Today, similar questions exist with generative AI. If AI can create entire movies, what will happen to traditional filmmaking?
David George: Weve already seen this in the images.
AI as a creative foundation
Chris Dixon: Yes, it has already started with images, and video may soon follow. When photography first emerged, people worried that it would replace painting, but in the end, photography and painting each developed their own unique artistic style. Fine art moved towards abstraction and away from photography. On the other hand, photography technology led to the rise of film. People realized that while machines could replace photography, they could also create a new art form that had never existed before.
The same is true for generative AI. Negative views believe that AI will replace human creation, but in fact, AI may give rise to entirely new art forms, providing a new canvas for human creativity, perhaps virtual worlds, games, or new types of movies. In addition to the creative industry, the same logic can be applied to other areas such as consumption and social networks.
When you create something new, broader changes follow. Social networking is a great example. It took off in the 2000s and reached a peak in 2008 and 2012 with the Obama election. News articles at the time also noted that social media moved from a secondary position to a primary position. Then we started to see unexpected social changes. These changes may unfold over the next 20 to 30 years.
Balancing supply and demand in AI
David George: The technical stages you mentioned are very interesting. The development of the Internet took a long time, one of the reasons being the need to build a huge network. This involves supply and demand issues - the development of the Internet requires the laying of wireless infrastructure such as optical fiber and cables. AI requires computing resources, such as large-scale GPU clusters. However, the main limiting factor for AI to move from the imitation stage to the innovation stage may not be technical capabilities, but human creativity and ideas.
Chris Dixon: I think so too. The bottleneck of AI development is probably not in technology, but in the speed of human adaptation and the impact of policies and regulations. The two are closely related.
David George: In other words, the problems of AI development include both the supply side (computing power) and the demand side (user acceptance). But the key may still be the demand side?
Chris Dixon: Yes, the challenge on the supply side is to develop sufficiently powerful AI models and have sufficient computing power to support them. But the real challenge is how to make users accept AI and integrate it into their daily lives.
We now see that many entrepreneurs are exploring how to use AI to solve practical problems. But unlike 20 years ago, the current entrepreneurial ecosystem has matured a lot. A dozen years ago, most smart people would not choose to start a business, but to work for a large company. But now, the entrepreneurial ecosystem is more complete, and financing, talent, and market are more mature than before.
But there is still a big problem with AI, which is how peoples working methods will change and how the industry will adapt to AI.
How AI is changing industries
David George: For example, how quickly will Hollywood adopt AI?
Chris Dixon: This is exactly what I was thinking about. When I was writing my book, I wanted to use AI to generate my own audiobook, but both the publisher and Audible explicitly prohibited the use of AI. Part of the reason is that the industry unions are resisting AI, but there are deeper reasons.
David George: So, the ability of AI to generate content exists, but the industry is not ready to accept it. We can see that many potential applications of AI face regulatory barriers. For example, in the medical industry, the technical capabilities of AI diagnosis are already strong enough, but regulations still restrict its widespread application.
Chris Dixon: In the next five years, US judges may rule on whether AI training data is fair use, or Congress may pass laws to regulate AI training data. Currently, the legality of AI training data is still controversial. AI companies believe that AI training data is learning of information, not copying. But copyright holders believe that AI uses their content without permission, which constitutes infringement.
David George: This is a question that almost all AI-related industries are debating.
Chris Dixon: Yes, ultimately laws may be needed to determine the reasonableness of AI training, otherwise the issue will remain unresolved.
David George: In regulated industries such as healthcare and finance, when will AI be truly implemented?
Chris Dixon: At present, these industries are subject to extremely strict regulation, and it may take a long time for AI to enter these fields. But in some areas, such as autonomous driving, we have seen significant progress.
David George: Waymo is an example. Data shows that its safety is 7 to 10 times higher than human driving, and it has millions of miles of real-world data to support it.
Chris Dixon: Perhaps this is the model for the widespread application of AI - first achieve a breakthrough in a specific field (such as autonomous driving) and prove that it performs better than humans, and then promote it to other industries.
What is the ideal future of the Internet?
David George: What do you think the ideal Internet should look like?
Chris Dixon: We are at a crossroads. The original vision of the Internet was a decentralized network that the community could jointly own and manage, and the economic benefits of the network should flow more to users rather than a few large companies. But now, the flow of funds on the Internet has changed, and more and more benefits are concentrated in the hands of a few technology giants.
David George: Yes, advertising revenue on social platforms has reached tens of billions of dollars, but creators can only get a small portion of it.
Chris Dixon: Currently, the top five Internet companies in the world by market value may have occupied more than 50% of the market share of the entire industry. The Internet has become a closed ecosystem dominated by a few companies.
David George: So now technology companies have mastered users and are starting to find ways to get users to spend more time on their platforms.
Chris Dixon: Yes, they have climbed to the top of the Internet, and then kicked away the ladder to prevent new competitors from entering. This is why we are so concerned about the construction of blockchain and decentralized networks. If the future Internet is completely controlled by a few companies, the space for innovation will be greatly compressed. Relying on centralized platforms to build a business is like building on quicksand, which may collapse at any time. Real innovation should be built on an open ecosystem, not controlled by a few companies.
David George: So, our focus is on how to enable small technology companies to survive and grow in this ecosystem. I am still very optimistic about the future. Through your efforts and the promotion of the entire industry, decentralized technology and open source AI are being accepted by more and more people. Todays discussion is great, thank you for your participation.
Chris Dixon: Thank you for your invitation.
*All content on the Coinspire platform is for reference only and does not constitute an offer or recommendation of any investment strategy. Any personal decision made based on the content of this article is the responsibility of the investor, and Coinspire is not responsible for any gains or losses arising therefrom. Investment is risky, so make decisions carefully!
