Blockchain & AI: sum much greater than its parts
how the technologies will amplify each other's capabilities
AI on the blockchain?
The recent explosion in generative AI technology, most notably OpenAI’s ChatGPT and DALL-E models, has captured the world’s attention and imagination.
Global interest in the term “AI“ (Google trends)
Unsurprisingly, the crypto industry has run with this narrative and created the meme “AI on the blockchain.” After all, who wouldn’t want to combine disruptive data networking (blockchain) with disruptive data processing (AI)? The logic seems to make sense, but there’s a lot of confusion about how that would actually work. Many just brush it off as the buzzword soup of the day — to be forgotten tomorrow when a new shiny narrative emerges.
While I believe the AI/crypto hype is warranted, I do think its often misguided. No, it doesn’t make sense to put AI directly on a blockchain; the economics & tech limitations will never make it viable (I’m not even sure what the point of that would be). But, they will still work together in meaningful, trust-minimized ways, amplifying each other’s capabilities.
How did we get here?
Researchers have been developing modern AI since the 1930s, but recent breakthroughs in data, compute, and algorithm technologies have led to an explosion in progress, specifically in generative AI (technology that can produce various types of content including text, imagery, audio and synthetic data).
The most significant breakthrough came in 2017 when a team at Google Brain created the transformer, a deep learning model initially used for language translation. However, researchers soon discovered its groundbreaking potential to power AI generated text and images using large language models (LLMs), which companies like OpenAI have leveraged in their products that are sweeping the globe today.
Source: TechTarget (link)
For more on the history of modern AI, I found this article by Ars Technica to be very helpful (link).
Generative AI today: ChatGPT
The most popular generative AI product today is ChatGPT, a supercharged chatbot built on-top of OpenAI’s GPT-4 large language model (LLM). More impressive than its ability to have human-like conversations on any topic, is its seemingly limitless potential as a general-purpose automation tool. This topic has been covered ad nauseam lately on twitter, but some basic examples include:
AI as a commodity
At a recent technology conference, investor Chamath Palihapitiya made an interesting analogy between AI and refrigeration:
The people and the person that invented refrigeration made some money. But most of the money was made by Coca Cola who used refrigeration to build an empire. And I view these large language models (LLMs) as refrigeration. Will there be some money made in it? I think so. But the Coca Cola has yet to be built. And those are the companies that are really gonna monetize it.
Clip: (link)
The value isn’t so much the AI itself — LLM technology will likely commoditize over time — but the utility that it can power. In other words, AI is simply the fuel; the real value is the utility built on top.
Building on top of ChatGPT
Generative AI functionality will be limited unless it can interact with the rest of your digital life. This is where ChatGPT plugins come in: APIs that connect to popular third-party applications and user-specific data. This interoperability allows ChatGPT to perform a wide range of actions:
Retrieve real-time information (sports scores, stock prices, latest news, etc.)
Retrieve knowledge-based information (company documents, personal notes, etc.)
Perform actions on behalf of the user (booking a flight, ordering food, etc.)
Plus much more as new APIs get released
Microsoft, a major investor in OpenAI, will soon leverage the technology to enhance Microsoft 365 through the launch of Copilot (announcement). With minimal input, users can automate sophisticated work functions across PowerPoint, Word, Excel, Outlook, and Teams:
All applications in the suite become interconnected and controlled by Business Chat, a Microsoft Office version of ChatGPT. For example, you can give it natural language prompts like “tell my team how we updated the product strategy” and it will generate a status update based on the morning’s meetings, emails, and chat threads.
Seeing is believing though — I highly recommend watching Microsoft’s recent demo to really appreciate the technology (video).
Generative AI systems, powered by LLMs, are developing at an accelerating rate. However, they are only as valuable as their utility, which requires seamless interoperability with the rest of the digital world.
Intersection of blockchain & AI
Now that we’ve established the value of generative AI, how does crypto enter the picture? There are many moral/ideological reasons to decentralize the technology (democratize intelligence, model transparency, etc.), but I’ll strictly focus on the economic implications:
Compounding AI
One of most exciting implications of open AI systems is “compounding AI.” Today, the top deep learning models are run and maintained by large research institutions or technology companies. They are siloed by design, to keep the proprietary “secret sauce” hidden from competitors. However, this limits the pace of innovation. Decentralized machine learning networks, like Bittensor (TAO), enable multimodal architectures where users are incentivized to contribute highly demanded intelligence (models, compute, data, etc.). These models even learn from each other, creating a global system of different LLM models that get stronger and more functional as the network grows. Theoretically, these networks have much higher intelligence potential than closed systems, given their collaborative learning environments.
For more information on Bittensor: (link)
Composability scales utility
Like software applications & compute legos (link), AI systems will also benefit from open access and interoperability. Think about Copilot: users of Microsoft 365 will find the application & data composability extremely valuable. But what if you also use Gmail, iOS, and Telegram? The more fragmented/siloed your data is, the less AI can do for you. For AI systems to provide the highest level of utility and functionality, they will have to securely interoperate across all user applications & access all your databases. This is only possible with decentralized architecture.
Again, this doesn’t necessarily mean “putting AI on the blockchain.” For example, Bittensor essentially serves as a decentralized oracle connecting off-chain machine learning to a trust-minimized intelligence marketplace. If successful, an entire ecosystem of interoperable, permissionless applications will be built on top of its AI engine. In comparison, to build on top of OpenAI, you must seek permission from management, negotiate a business contract, and develop/deploy a specific API to connect the two systems. This is a much slower and exclusive process that limits innovation & development.
Data access
AI models are only as powerful as the data used to train them. Centralized systems today are unable to collect datasets from separate parties who don’t want or can’t share their data, greatly limiting functionality. Web3 projects like Ocean Protocol facilitate decentralize data markets, enabling users to monetize their data while completely preserving content privacy. This allows LLMs to be trained with sensitive data (like medical records) without having control of the actual data.
For more information on Ocean Protocol: (link)
Authentication
Generative AI will soon have the ability to create artificial videos that look and sound exactly the real thing, unfortunately democratizing deepfake technology. It will become increasingly difficult to decipher authenticity in the digital world, but blockchains provide tools to automate identity verification and prevent impersonators/scammers (“don’t trust, verify”).
AI will also be highly valuable for the crypto industry
GitHub, a Microsoft subsidiary, recently launched GitHub Copilot X, a ChatGPT-like product specifically built for software developers. Some highlights include:
Not only does it have the potential to materially increase developer efficiency & productivity, the automation features make it possible to create blockchain software with little-to-no code. This is significant since most smart contract coding languages are unique to crypto, so these automation tools will help ease the technical transition from web2 (30M developers) to web3 (only 100k developers currently).
AI will also enhance blockchain application security, by materially improving code auditing capabilities and automating the detection and resolution of real-time protocol hacks.
Convergence of blockchain & AI
Its not just a meme — blockchain and AI will increasingly depend on each other to drive innovation and stay competitive in their respective markets. Trust-minimized compute environments (powered by blockchain) optimize AI intelligence, interoperability, data security, and user-authenticity. While AI automation tools will 10x blockchain developer efficiency and application safety. As the two technologies continue to compound on each other, the next few years are going to be wild.
Cover photo cred: Coin Rivet