"Software is eating the world" has become one of the iconic phrases of the last decade in the software industry. It was quoted in 2011 by software legend and venture capitalist extraordinaire Marc Andreessen, and it summed up the idea that companies that operated primarily in the physical world are transitioning into the digital economy - a trend that will essentially transform every business into a software company.
Jesus Rodriguez is CEO of IntoTheBlock, a blockchain and cryptocurrency market analytics company. This article is a preview of a talk he will give this week on the Big Ideas stage at Consensus 2022 in Austin, Texas.
In recent years, the development of machine learning (ML) and artificial intelligence (AI) has permeated all areas of the software industry, leading many experts to claim that "machine learning eats software." Cryptocurrencies and digital assets are based on code and programmability, and are therefore likely to be influenced by ML-AI trends. The intersection of ML-AI with digital assets will likely usher in a new era where intelligence becomes a native component of crypto assets.
The idea of intelligent crypto-assets is conceptually trivial but fraught with practical challenges. What are some of the fundamental ML trends that can rapidly impact the next generation of crypto assets? What about the key scenarios that can benefit from smart cryptocurrencies, or some of the key technical challenges that need to be overcome for cryptocurrencies to become smart. This paper explores some of these ideas and develops a thesis on the potential for crypto and ML to intersect.
Only Cryptocurrencies Can Be Inherently IntelligentImportant
point to keep in mind when considering AI-ML in the context of crypto assets is that cryptocurrencies are the only asset class in history that has the potential to become inherently intelligent. AI ML capabilities in traditional asset classes like commodities or equities are implemented in vehicles like robo-advisors or quant strategies that live outside the asset itself. Even though there is an obvious role for these vehicles in the crypto space, crypto assets can natively embed these AI ML capabilities into assets. This benefit is, of course, a side effect of cryptocurrencies' programmable and digital capabilities. Crypto assets are based on code, and that code could take the form of AI-ML models.
Machine learning will eat up cryptocurrencies, but how?
AI-ML will likely play an important role in the next decade of the crypto market. While the first phases of cryptocurrency focused on digitization and automation, the next iteration seems destined to focus on intelligence. There are many applications of AI ML in the crypto economy today, but we cannot claim that crypto assets are inherently intelligent. In the near future, we should expect crypto assets and protocols to begin incorporating AI-ML as native capabilities that allow them to learn and adapt their behavior based on their environment or markets.
The inevitability of digital assets becoming smart stems in part from the amazing evolution of AI-ML technologies in recent years. In the context of cryptocurrencies, we should not think of AI-ML as a generic thing, but rather as a group of interconnected methods. From this perspective, there are a small number of AI-ML schools that seem particularly well suited for crypto applications. Let's look at some of the most popular techniques in terms of their potential for crypto technologies.
by many to be the
most important development of the last decade of AI-ML
, transformers are behind the revolution in natural language understanding (NLU) and are gaining acceptance in other areas such as computer vision. Models such as OpenAI's GPT-3 or NVIDIA's Megatron are capable of generating synthetic text indistinguishable from real text, performing highly complex question-answer interactions, or even making inferences about text forms. Models such as OpenAI's DALL-E 2 or Google's Imagen are capable of generating artistic images from text forms, bridging the gap between intelligence and different domains.
Given the impact transformers have had in NLU and computer vision, it's not hard to imagine the impact they will have in domains like NFTs that rely on visual representations and textual interactions.
(Facebook) AI Research recently referred to self-supervised learning (SSL) as the "dark matter of AI" as an analogy to the fundamental role this new type of technology can play in the next generation of AI models. Conceptually, SSL seeks to enable intelligent capabilities similar to the way babies learn through observation and interaction. SSL seeks to overcome some of the limitations of traditional supervised learning methods that must be trained with large amounts of labeled data. Models such as Meta's DINO are able to classify objects in images without prior training.
The applications of learning without huge amounts of labeled data seem perfect for cryptocurrencies. Decentralized finance (DeFi) could be an immediate beneficiary of these methods.
Graph NeuralNetworksBlockchain recordsrepresent
the largest source of data in the crypto space. From a structural standpoint, blockchain datasets are inherently hierarchical, modeling relationships between addresses, transactions, or blocks. Graph neural networks (GNNs) are the AI ML discipline that specializes in learning over hierarchical datasets. Companies like Google's DeepMind use GNNs to predict traffic in Google Maps or even understand the structure of glass.
GNNs seem like a perfect AI ML technique for crypto assets. If blockchains are ever to become intelligent, GNNs will likely play a key role in developing knowledge from their own datasets.
Learning (DRL) became something of a pop culture after DeepMind's AlphaGo defeated multiple-time Go world champion Lee Sedol. AlphaGo mastered Go by playing an unimaginable number of games against itself and correcting its own mistakes. This form of trial-and-error learning through interaction is the essence of DRL.
Since AlphaGo, DRL has been at the center of remarkable AI ML successes. DeepMind's AlphaFold shocked the scientific community by being able to predict protein structure from an amino acid sequence, a discovery that may usher in a new era in medicine. Another standout DRL model from DeepMind was MuZero, which is able to master games like Go, chess, and Atari without even knowing the rules.
The DRL principle of learning by trial and error seems relevant to many crypto domains such as DeFi or NFTs, where conditions are constantly changing. Finally, most crypto protocols are based on strong game-theoretic rules, and DRL have proven to excel at games.The Road to Intelligence in CryptoCyberpunk legend
and science fiction author William Gibson once said, "The future is already here-it's just not evenly distributed." This quote could serve as a philosophical guideline for us as we think about the path to smart crypto. The emergence of cryptocurrencies coincided with the golden era of AI-ML research and technology development. Today, AI-ML technologies are on the rise, and it's only a matter of time before they become first-class citizens in the crypto space as well. The use cases seem to be everywhere. Let's look at some of the most obvious ones.
Intelligent NFTsThere are already
some applications where generative AI-ML methods are used to create NFTs. However, the impact of AI-ML should extend to all areas of NFTs. Let's imagine NFTs that incorporate speech and language capabilities to establish a dialogue with the user, answer questions about its meaning, or interact with a particular environment. Imagine starting a conversation with your favorite digital assistant using a visual NFT that can change its appearance depending on the nature of the dialogue. Similarly, imagine using AI ML transformation models trained on millions of paintings to create unique NFTs that capture unique aspects of the masters' style.
Intelligent DeFi ProtocolsDeFi protocolsare
all about automation, but they're not exactly intelligent. Integrating AI ML capabilities into DeFi protocols seems inevitable. We can imagine a new generation of automated market maker (AMM) protocols that can adjust balances in pools using real-time predictive models based on existing market conditions. Similarly, we can envision credit protocols that adjust the size of loans based on an intelligent profile of the requesting addresses.
Intelligent L1-L2 Blockchains Artificial intelligenceinfluences
all aspects of software infrastructure, such as networking, computing, or storage systems, and blockchains are unlikely to be an exception. It is not far-fetched to think about intelligent consensus protocols that improve performance based on predictive models. Similarly, we can think of blockchains developing smart economies to control computational costs in terms of gas or other equivalents.
Smart crypto apps and dappsUser experience
seems to be one of the most obvious areas for adoption of AI ML capabilities. It's only a matter of time before wallets or exchanges start incorporating native intelligence features that help improve investment and trading decisions, which today depend entirely on human subjectivity.
programmable stablecoins seems to be very present these days after the collapse of Terra UST. What if instead of thinking of this form of stablecoins as programmable, we think of them as programmable but also intelligent? Instead of programmable stablecoins that adjust the peg based on statically defined economic gymnastics, how about using AI-ML algorithms that organically learn from market conditions. A combination of AI-ML and human oversight seems like an interesting approach to explore in this space.AI-ML influences cryptocurrencies, but cryptocurrencies can also contribute to AI-ML Therelationship
between cryptocurrencies and AI-ML is more bidirectional than most people think. While the scenarios where AI-ML can impact the next generation of crypto assets and infrastructures are fairly clear, there are some non-obvious areas where crypto can impact AI-ML technologies.
Decentralized AI (dAI) is an emerging technology movement that seeks to leverage decentralization of computation, as well as tokenization mechanisms, to mitigate some of the increasing centralization problems of AI ML technologies. A subset of the general dAI approach is mechanisms that leverage cryptoassets to create an economy where companies and individuals are incentivized to share data and AI-ML models.
Data is the stream of AI ML, but today it is highly controlled by a small number of incumbents, and there are virtually no incentives for companies to collaborate and share data to break this monopolistic cycle. The introduction of smart tokenomics and incentive mechanisms could organically help companies collaborate and share the benefits of creating and training AI ML models for specific tasks on a regular basis.
Bias and fairness are another hot topic in AI-ML today that could be greatly impacted by the use of native crypto technologies. The datasets used in training AI-ML models are riddled with biases, discrimination, and toxic data points that can affect the knowledge of AI models.
Although there have been many advances in quantifying and monitoring the fairness of AI ML models, there are no robust accountability and benchmarking mechanisms that the entire industry trusts. Imagine using a blockchain layer to track the biases and fairness scores of specific AI ML models and reward models that improve their fairness scores. This is an entry-level scenario for the use of blockchain technologies in AI ML infrastructures.
There is no doubt that AI-ML should be a fundamental element of the next generation of digital asset technologies, but there is also a lot of concrete value that cryptocurrencies and blockchains can deliver in the world of AI-ML. Fundamentally, cryptocurrencies could serve as an economic and accounting layer that supports the building of more equitable and democratic AI-ML solutions.From digitization to automation to intelligence, AI-MLis influencing
every single area of the software world, and cryptocurrencies are unlikely to be an exception. The core principles of digital asset technologies have focused on democratizing financial services through digitization and automation. Intelligence is one of the next frontiers for cryptocurrencies, and we are likely to see the impact across the board. From smart NFTs, DeFi protocols to new forms of crypto assets, the incorporation of AI-ML will likely usher in a new era of innovation in the crypto economy. The technologies and use cases are already in place. It's time to start building them.
Also in the series 'Big Ideas':
The Coming InDAOstrial Revolution by Julie Fredrickson.
Distributed, autonomous organizations give people the ability to build bigger, crazier things on radical timelines, just as the advent of the corporation paved the way for the Industrial Revolution.
Trustless Evidence: Web 3 helps document war crimes in Ukraine by Jonathan Dotan
In an era of misinformation, blockchain technology can renew our faith in provable truth, not least during the current conflict in Ukraine, says Jonathan Dotan, founding director of The Starling Lab.
How Web 3 is changing philanthropy by Rhys Lindmark
Rhys Lindmark, a "Big Ideas" speaker at CoinDesk's Consensus Festival, on how the crypto generation could rewrite the rules for charitable giving.
Let's use new forms of money to engage with our communities by Matthew Prewitt
More local money could reduce the incentive to leave communities in need of resources, says Matt Prewitt, president of RadicalxChange Foundation.
Forecasts, prediction markets and the age of better information by Clay Graubard and Andrew Eaddy.
Quantified forecasts are an invaluable yet underutilized tool, and prediction markets appear to be a critical tool for their adoption.