Partnered With:

Latest research from a16z Crypto: Four new business models emerging from the convergence of AI and blockchain.

Read this article in 35 Minutes

The current situation and opportunities of the integration trend between blockchain and AI.

Original video: Web3 with a16z, AI & Crypto

Original author: Dan Boneh (Professor at Stanford University and Senior Research Advisor at a16z crypto), specializing in cryptography, computer security, and machine learning; Ali Yahya (General Partner at a16z crypto), formerly worked at Google Brain and was also one of the core contributors to Google's machine learning library TensorFlow.

Compiled and translated by: Qianwen, Chain Catcher

Stephen King once wrote a science fiction novel called "The Diamond Age", in which there is an artificial intelligence device that serves as a mentor to people throughout their lives. When you are born, you are paired with an AI that knows you very well - your preferences, your decisions, and guides you towards the right direction. This sounds great, but you would never want this technology to fall into the hands of middleman giants. Because this would give the company a lot of control and lead to a series of privacy and sovereignty issues.

We hope that this technology can truly belong to us, and thus a vision emerges that you can use blockchain to achieve this. You can embed artificial intelligence in smart contracts. With the power of zero-knowledge proofs, data privacy can be maintained. In the next few decades, this technology will become increasingly intelligent. You can choose to do anything you want or change it in any way you wish.

So what is the relationship between blockchain and artificial intelligence? What kind of world will artificial intelligence lead us to? What is the current situation and challenges of artificial intelligence? And what role will blockchain play in this process?

AI and Blockchain: A Battle of Equals

The development of artificial intelligence, including the phenomenon described in "The Diamond Age", has actually been around for a long time, but has recently experienced a leap in development.

First of all, artificial intelligence is largely a top-down, centrally controlled technology. Cryptography, on the other hand, is a bottom-up, decentralized collaborative technology. In many ways, cryptocurrency is a study of how to build decentralized systems that can enable large-scale human cooperation without a truly centralized authority. From this perspective, it is a natural fit for these two technologies to come together.

Artificial intelligence is a sustainable innovation that can enhance the business models of existing technology companies and help them make top-down decisions. The best example of this is Google, which can decide what content to present to users among billions of users and billions of page views. Cryptocurrency, on the other hand, is essentially a disruptive innovation, and its business model is fundamentally at odds with that of large technology companies. Therefore, this is a movement led by rebels on the edge, not by those in power.

Therefore, artificial intelligence may be closely related to privacy protection, and the two promote and interact with each other. As a technology, artificial intelligence has established various incentive mechanisms, leading to less and less privacy for users, as companies want to obtain all of our data. And artificial intelligence models trained with more and more data will become more effective. On the other hand, artificial intelligence is not perfect, and models may have biases, which may lead to unfair results. Therefore, there are also many papers on algorithm fairness at this stage.

I believe that we will embark on a path towards artificial intelligence, where everyone's data will be aggregated into these massive models for optimization. Meanwhile, cryptocurrency is moving in the opposite direction by increasing personal privacy and empowering users to control their data sovereignty. It can be said that encryption technology is a technology that competes with artificial intelligence because it helps us distinguish between human-created content and AI-created content from a wealth of information. In a world where AI-created content is rampant, encryption technology will become an important tool for maintaining and preserving human content.

Cryptocurrency is like the Wild West because it has no authority and anyone can participate. You have to assume that some of the participants are malicious. Therefore, there is a greater need for tools to help you filter out honest participants from dishonest ones, and machine learning and artificial intelligence are intelligent tools that can be very helpful in this regard.

For example, there are projects that use machine learning to identify suspicious transactions submitted to wallets. This way, these transactions will be flagged and submitted to the blockchain. This can effectively prevent users from accidentally submitting all their funds to attackers or doing something they will regret later. Machine learning can also be used as a tool to help you predict which transactions may involve MEV.

Just as the LLM model can be used to detect fake data or malicious activity, conversely, these models can also be used to generate fake data. The most typical example is deepfakes. You can create a video of someone saying something they never actually said. However, blockchain can actually help mitigate this problem.

For example, there is a timestamp on the blockchain that shows the date when you said something. If someone forges a video, you can use the timestamp to deny it. All of this data, the truly authentic data, is recorded on the blockchain and can be used to prove that the deeply forged video is indeed fake. Therefore, I believe that blockchain may help combat forgery.

We can also rely on trusted hardware to achieve this. Cameras and devices such as our phones will sign the captured images and videos as a standard. This is known as C2PA, which specifies how cameras sign data. In fact, a Sony camera can now take photos and videos and generate a C2PA signature on the video. This is a complex topic that we will not delve into here.

Usually, newspapers do not publish camera-taken photos as they are. They will crop and authorize the photos before publishing. Once you start editing the photos, it means that the recipients, final readers, and users on the browser will not see the original photos, and therefore cannot perform C2PA signature verification.

The question is, how can users confirm that the image they are seeing is indeed signed correctly by the C2PA camera? This is where ZK technology comes in handy, as you can prove that the edited image is actually the result of downsampling and grayscale scaling of the correctly signed image. This way, we can replace C2PA signatures with simple zk proofs and correspond them one by one with these images. Now, readers can still confirm that what they are seeing is a real image. Therefore, zk technology can be used to combat this information.

How to Break Through in the Blockchain Industry?

Artificial intelligence is essentially a centralized technology. It benefits greatly from economies of scale, as running on a single data center makes things more efficient. In addition, data, machine learning models, and machine learning talent are usually controlled by a few technology companies. So how can we break through? Cryptocurrencies can help us achieve the decentralization of artificial intelligence by using technologies such as ZKML, which can be applied to data centers, databases, and machine learning models themselves. For example, in terms of computation, using zero-knowledge proofs, users can prove that the actual process of reasoning or training models is correct.

In this way, you can outsource this process to a large community. In this distributed process, anyone with a GPU can contribute computing power to the network and train models in this way, without relying on a large data center that centralizes all GPUs.

From an economic perspective, it is uncertain whether this makes sense. But at least through the right incentives, the long tail effect can be achieved. You can utilize all possible GPU capabilities. All these people can contribute computing power for model training or inference, which can replace large tech companies that control everything. To achieve this, various important technical issues must be addressed. In fact, there is a company called NVIDIA that is building a decentralized GPU computing market mainly for training machine learning models. In this market, anyone can contribute their own GPU computing power. On the other hand, anyone can use any computing power existing in the network to train their large machine learning models. This will become an alternative choice to centralized large tech companies such as OpenAI, Google, and Meta.

Imagine a scenario where Alice has a model that she wants to protect. She wants to send the model to Bob in an encrypted form. Bob receives the encrypted model and needs to run his own data on it. How can he do this? This is where fully homomorphic encryption comes into play. If the user has the encrypted model and plaintext data, they can run the encrypted model on the plaintext data, receive the encrypted result, and send it back to Alice. She can then decrypt it and see the plaintext result.

This is actually an existing technology. The question is, can we extend it to larger models? This is a significant challenge that requires more effort from companies, as the current technology is effective for medium-sized models.

Current Situation, Challenges, and Incentive Mechanisms

I believe that decentralization in computing is necessary. The first issue is verification, which can be solved using ZK, but currently these technologies can only handle smaller models. The challenge we face is that the performance of these cryptographic primitives is far from meeting the needs of training or inference for super-large models. Therefore, a lot of work is being done to improve the performance of the proof process, so that increasingly larger workloads can be efficiently proven. Meanwhile, some companies are also using other technologies that are not just cryptographic, but use game theory properties to let more independent people work together. This is an optimistic approach that does not rely on cryptography, but it still aligns with the larger goal of decentralized artificial intelligence or helping to create an artificial intelligence ecosystem. This is the goal proposed by companies such as OpenAI.

The second major issue is the problem of distributed systems. For example, how do you coordinate a large community to contribute to a network's computing power, making it feel like an integrated and unified computing infrastructure? There are many challenges involved, such as how to reasonably decompose the workload of machine learning and allocate different workloads to different nodes in the network, as well as how to efficiently complete all of this work.

Currently, the technology can be applied to medium-sized models, but cannot be applied to models as large as GPT-3 or GPT-4. However, there are other methods available. For example, we can train multiple people and compare the results, creating a game-theoretic incentive mechanism to discourage cheating. If someone cheats, others may complain that their calculated training results are incorrect, and the cheater will not receive a reward. We can also distribute data sources in the community to train large machine learning models. Similarly, we can collect all data and train models ourselves, rather than relying on a centralized organization. This can be achieved by creating a market, similar to the computing market we described earlier.

We can also take an incentivized approach, encouraging people to contribute new data to a large dataset that can be used to train models. The challenges and validation challenges involved are similar. You must somehow verify that the data contributed by people is indeed good data. This data is neither duplicate data, nor randomly generated junk data, nor artificially generated unrealistic data.

In addition, it is also important to ensure that the data does not in any way undermine the model, otherwise the performance of the model will actually deteriorate over time. Perhaps we must rely on a combination of technical and social solutions, in which case you can also establish credibility by using site indicators that are available to certain community members. This way, when they contribute data, it will be more reliable than before.

Otherwise, it takes a very long time to truly cover the data distribution. One of the biggest challenges in machine learning is that the model can only cover the distribution range that the training dataset can reach. If some inputs are far beyond the distribution range of the training data, then your model may behave completely unpredictably. In order to make the model perform well in edge cases, black swan data points, or data inputs that may be encountered in the real world, we need a dataset that is as comprehensive as possible.

Therefore, if you have an open, decentralized market that provides data for datasets, you can allow anyone in the world with unique data to provide that data to the network, which is a better way. Because if you try to do this as a central company, you have no way of knowing who owns this data. Therefore, if you can create an incentive mechanism that encourages these people to come forward and provide this data voluntarily, then I believe you can actually achieve significantly better coverage of long-tail data.

So we must have some mechanism to ensure that the data you provide is authentic. One method is to rely on trusted hardware, embedding some trusted hardware in the sensor itself, and we only trust data that is correctly signed by the hardware. Otherwise, we must have other mechanisms to distinguish the authenticity of the data. There are currently two important trends in machine learning. First, the performance measurement methods of machine learning models are constantly improving, but they are still in the early stages, and it is actually difficult to judge the performance of another model. Another trend is that we are becoming increasingly adept at explaining how the model works.

Therefore, based on these two points, sometimes I may be able to understand the impact of the dataset on the performance of machine learning models. If we can understand whether third-party contributed datasets contribute to the performance of machine learning models, then we can reward this contribution and create incentives for the existence of this market.

Imagine if you could create an open market where people can contribute trained models to solve specific types of problems, or if you created a smart contract with some kind of test embedded in it, and someone could use zkml to provide a model and prove that it can solve the test, that would be a final solution. You now have the tools to create a market where people are incentivized to contribute machine learning models that can solve certain problems.

AI and Encryption: How to Form a Business Model?

I believe the vision behind the intersection of cryptocurrency and artificial intelligence is that you can create a set of protocols to distribute the value obtained from this new technology of artificial intelligence to more people, so that everyone can contribute and share the benefits brought by this new technology.

Therefore, the people who can profit will be those who contribute computing power, contribute data, or contribute new machine learning models to the network, so that better machine learning models can be trained to solve more important problems.

Network demanders can also benefit. They use this network as the infrastructure for training their own machine learning models. Perhaps their models can contribute some interesting things, such as the next generation of chat tools. In these scenarios, because these companies will have their own business models, they can drive value acquisition themselves.

The people who establish this network will also benefit. For example, creating a token for the network that will be distributed to the community. All of these people will collectively own this decentralized network, which can be used for computing data and models, and can also gain some value from all economic activities conducted through this network.

You can imagine that every transaction, every payment method that calculates fees for data or model fees through this network, may be charged a certain fee, which will enter the treasury controlled by the entire network. Token holders collectively own this network. This is essentially the business model of the network itself.

Artificial Intelligence Promotes Code Security

Many listeners may have heard of co-pilots, which is a tool used to generate code. You can try using these collaborative generation tools to write Solidity contracts or cryptographic code. However, it is important to note that this is actually very dangerous because often times, when you try to run it, these systems will actually generate code that can run but is not secure.

Actually, we recently wrote a paper on this issue, which pointed out that if you try to get a collaborative pilot to write a simple encryption function, the encryption function it provides is correct. However, it uses an incorrect operation mode, so you will end up with an insecure encryption mode.

You may wonder why this happens. One reason is that these models are basically trained based on existing code, which is trained in github repositories. Many github repositories are actually vulnerable to various attacks. Therefore, the code learned by these models can work normally, but it is not secure. This is like producing garbage with low quality, which produces garbage. Therefore, I hope that people will be careful and cautious when using these generative models to generate code, carefully check whether the code has really done what it should do, and done it safely.

You can use artificial intelligence models, combined with other tools, to generate code and ensure that the entire process is error-free. For example, one idea is to use the LLM model to generate specifications for a formal verification system, requiring LLM to generate a specification as a formal verification tool. Then, the same LLM instance is required to generate a program that complies with the specification, and then a formal verification tool is used to see if the program really complies with the specification. If there are vulnerabilities, the tool will also capture them. These errors can be fed back to LLM as feedback, and ideally, LLM can modify its work and generate another correct version of the code.

Finally, if you repeat the operation, you will eventually get a piece of code that, ideally, can fully satisfy this return value and, after formal verification, can also satisfy this return value. Moreover, since humans can read this trace, you can see that this is the program I want to write through this trace. In fact, many people are already trying to evaluate the ability of LLM to find software vulnerabilities, such as Unity smart contracts, C and C plus.

So, will we reach a point where code generated by LLM is less likely to contain bugs than code generated by humans? For example, when we talk about autonomous driving, are we concerned that it is less likely to crash than a human driver? I believe this trend will only become stronger, and the degree to which artificial intelligence technology is integrated into existing toolchains will also increase.

You can integrate it into your form validation toolchain, and you can also integrate it into other tools, such as the memory management issue checking tool mentioned earlier. You can also integrate it into your unit testing and integration testing toolchain, so that llm is not just operating in a vacuum. It can receive real-time feedback from other tools that connect it to the ground reality.

I believe that by combining the use of super large machine learning models trained on all the data in the world with these other tools, it may make computing programs better than human programmers. Even though they may still make mistakes, they may be superheroes. This will be an important moment in software engineering.

Artificial Intelligence and Social Graph

Another possibility is that we may be able to establish a decentralized social network, which behaves much like Weibo, but the social graph is completely on the chain. It's almost like a public product that anyone can build on. As a user, you can control your identity on the social graph. You can control your data, who you follow, and who can follow you. In addition, there are a large number of companies building portals in the social graph, providing users with experiences similar to Twitter, Instagram, TikTok, or any other experience they want to create.

But all of this is built on the same social graph, which no one owns, and there is no multi-billion dollar tech company in complete control of it.

This is an exciting world, because it means it can be more vibrant and have an ecosystem built by people together. Each user can have more control over what they see and do on the platform.

But at the same time, users also need to filter signals from noise. For example, it is necessary to develop reasonable recommendation algorithms to filter all content and show you the news sources you really want to see. This will open a door for the entire market, a competitive environment composed of service providers. You can use algorithms, use AI-based algorithms to plan content for you. As a user, you can decide whether to use a specific algorithm, perhaps one established by Twitter or another algorithm. But similarly, you also need tools like "machine learning" to help you filter out noise and help you parse all the junk information. In this world, generative models can create all the junk information in the world.

Why is human proof important?

One highly relevant question is, in a world where artificially generated content is rampant, how can you prove that you are indeed human?

Biometric technology is a possible direction, and one project is called World Coin, which uses retina scanning as biometric information to verify if you are a real person, ensuring that you are indeed a living person and not just a photo of an eye. This system has secure hardware that is difficult to tamper with, so the proof that appears on the other end, which is the zero-knowledge proof that masks your actual biometric information, is difficult to forge in this way.

On the internet, no one knows you're a robot. Therefore, I believe this is where the importance of the human proof project lies, as it will become crucial to know whether you are interacting with a robot or a human. Without human proof, it is impossible to determine whether an address belongs to an individual or a group of people, or whether ten thousand addresses truly belong to one person or are just pretending to be ten thousand different people.

This is crucial in governance. If every participant in the governance system can prove that they are actually human, and they can prove it in a unique way because they only have one set of eyeballs, then the governance system will be more fair and less prone to becoming an oligarchy (based on the preference for the maximum amount locked in a smart contract).

Artificial Intelligence and Art

Artificial intelligence models mean that we will live in a world with unlimited media richness. In such a world, communities centered around any specific media or narratives around specific media will become increasingly important.

For example, Sound.xyz is building a decentralized music streaming platform that allows artists and musicians to upload music and directly connect with our community by selling NFTs. Users can comment on tracks on the sound dot xyz website, and other listeners can see the comments. This is similar to previous SoundCloud functionality. Buying NFTs also supports artists and helps them achieve sustainable development and create more music. But the wonderful thing about all of this is that it actually provides artists with a truly interactive platform with the community. Artists are everyone's artists.

Due to the role of cryptocurrency here, you can create a community around a piece of music, and if a piece of music is only created by a machine learning model without any human elements, then this community will not exist.

We will encounter a lot of music that will be completely generated by artificial intelligence, and it will be very important to establish communities and tools that revolve around art, music, and other types of media to tell stories. It will distinguish the media that we truly care about, truly want to invest in, and spend time engaging with from other general media.

There may be some synergies between the two, such as many music pieces being enhanced or generated by artificial intelligence. However, if there are also human elements involved, such as creators using AI tools to create a new piece of music, they have their own unique voice, artist page, community, and followers.

Now, there is a synergistic effect between these two worlds, and everyone has the best music because artificial intelligence has given them superhuman abilities. However, at the same time, they also have human elements and stories, which are coordinated and realized through encryption technology, allowing you to gather all these people on one platform.

In terms of content generation, this is definitely a brand new world. So how do we distinguish between human-generated art and machine-generated art that needs to be supported?

This actually opens a door for collective art, generated through the creative process of the entire community rather than a single artist. There are already some projects doing this, where the community influences the chain through some voting process and generates art pieces based on prompts from machine learning models. Perhaps what you generate is not just one piece of art, but ten thousand. Then you use another machine learning model, which is also trained based on community feedback, to select the best one from these ten thousand pieces.