The convergence of Blockchain and AI can enhance machine learning and enable AI to create and trade financial products
Artificial Intelligence solutions will soon run on top of blockchains, increasing machine learning capability and even creating new financial products. Blockchain-AI convergence is inevitable because both deals with data and value. Blockchain enables secure storage and sharing of data or anything of value. AI can analyze and generate insights from data to generate value.
We will consider two (out of many) areas where blockchain and AI can be combined. First, we’ll explore how machine learning models can be collaboratively improved on blockchains. We will see this in the near future since Microsoft already working on this. Then, we will look at how AI agents can autonomously create and trade new financial instruments over blockchains. Finally, we’ll look at how organizations can position themselves for the blockchain-AI convergence.
First, a reality check. Most of the chatter about startups combining blockchain and artificial intelligence is largely hype. These companies are very young, with few clients and not much commercialization.
Many of these companies raised money through initial coin offerings (ICO), meaning that their solutions weren’t as thoroughly vetted as they would have been had they raised venture capital money. Some of these small companies might well become successful, and the use cases they are working on are definitely important. We may hear more about them in the coming years but won’t focus on them here.
Blockchain and AI Basics
Blockchain is a digital ledger that can record not just economic transactions but virtually anything of value.
Public blockchains such as Ethereum are open to the public. Private (permissioned) blockchains are invitation-only and are often used in corporate environments. Private blockchains are faster than public blockchains because participants are known and trusted and transactions can be verified faster.
A key feature of blockchain is that it enables unrelated parties to transact and share data with each other on a common ledger. Transactions are validated using cryptography and consensus mechanisms such as Proof of Work. This is significant because participants don’t have to trust each other or rely on third-party validators to transact with each other.
Blockchain has powerful implications for financial transactions (e.g. Bitcoin) and even sharing sensitive data across organizations. People can be confident that the data on a blockchain is trustworthy even if they can’t see the underlying data.
Blockchain uses cryptography to ensure that data, transactions, and identities can be:
- Incorruptibly, securely and irreversibly recorded
- Verified as trustworthy while remaining private — participants can verify the veracity of data without needing to look at the data, and only see what they are authorized to see
- Easily shared so that everyone in a blockchain network has an identical copy of the entire ledger, including updates as they occur
For those interested in a more comprehensive blockchain guide, check out this post.
Artificial intelligence involves using computers to do things that require human intelligence. AI models can be used to analyze, classify and make predictions from data. Unlike traditional software, AI models can also improve (learn) over time as they are fed new data.
Machine learning, a subset of artificial intelligence, is used to glean insights from data. In general, larger datasets help create better machine learning models. Equally important is the quality of the data — datasets need to be updated with recent and relevant data so that models can remain effective.
Data is central to AI effectiveness, and blockchain enables collaborative and secure data sharing. Blockchain can ensure the trustworthiness of data and can enable more data to be securely shared before AI extracts insights from it.
Microsoft: Improving Machine Learning Models on the Blockchain
Researchers at Microsoft are working on ways to collaboratively improve machine learning models hosted on public blockchains. This collaboration is incentivized since blockchain makes it possible to reward people who help improve models.
While huge advances are being made in machine learning, the benefits are not widely accessible. People with limited resources can’t always access cutting edge machine learning systems that are highly centralized and use proprietary datasets that are expensive to re-create. Furthermore, even the best models become outdated if they aren’t regularly retrained with new data.
Microsoft is trying to make AI decentralized and collaborative using blockchain. In this future, people can easily and cost-effectively run advanced machine learning models on everyday devices and apps (e.g. laptops, browsers, mobile) and collectively contribute data and improve models.
Allowing advanced AI models and large datasets to be widely shared, updated and trained could increase the rate of AI adoption and effectiveness
Microsoft is developing a Decentralized & Collaborative AI on the Blockchain framework to enable the AI community to collaboratively train models and build datasets on public blockchains. Importantly, people can use the machine learning models for free. Some of many applications include developing virtual assistants or recommender systems (e.g. what Netflix uses to recommend shows). Proof of concepts was created using Ethereum.
Blockchain usage makes sense because it offers participants trust and security. You can be 100% sure of what code you are interacting with. Instead of requiring specialized cloud services, Microsoft’s framework puts public models into smart contracts that codify model specifications. Models can be updated on the blockchain or used off-chain on the user’s local device at no transaction cost. The immutable nature of blockchain and smart contracts means that the model will always perform to specification. Once the model is updated and validated, every user will see it as the ‘one true version.’
Blockchain also provides an incentive system that encourages participants to contribute to data that improves models. The ability to verify and track changes allows us to accurately compute and payout rewards (in tokens) for contributions that improve AI models.
Microsoft’s researchers claimed that it costs $0.25 to update a Perceptron model on Ethereum. In the future, they hope that even this fee will not be required. Users are rewarded depending on how much their contribution helped the model improve. Good contributions are rewarded while bad (malicious) contributions are punished by taking away user deposits.
While Microsoft’s framework isn’t operating at scale yet, their vision could soon become the norm. Allowing advanced AI models and large datasets to be widely shared, updated and trained could increase the rate of AI adoption and effectiveness.
Digital Investment Assets Traded by AI Agents over Blockchain
Blockchain is already used to store and trade financial instruments such as cryptocurrencies and security tokens (cryptographic tokens backed by underlying assets). However, this is a nascent market that is only a few years old. Security tokens themselves are even more nascent— by one measure, the total security token market cap for January 2020 was a mere $52.7 million.
AI will create and trade digital investment assets over high-speed private blockchains
Clearly there is not enough activity (and data) to apply AI to financial products traded over blockchain yet. However, as data volumes going through blockchains increases, AI can glean insights from data, help create financial products, and even trade these products autonomously.
- Stage I: Blockchain proof of concepts
- Stage II: Tokenization of assets on blockchain
- Stage III: Digital investment assets traded on blockchains, powered by machine learning
- Stage IV: AI as economic agents that trade digital investment assets
We are in the second stage where assets can be tokenized and traded on blockchains. Tokens can represent underlying securities, physical assets, rights to cash flow, or utilities. Tokenizing and trading assets on blockchains reduces transaction costs and settlement time while improving auditability. AI and machine learning become applicable for pattern detection and predictive algorithms. However, we don’t have enough on-chain activity to apply AI yet.
The third stage will see the introduction of native digital assets. Tokens can go from representing an underlying asset to becoming the underlying asset. While this concept is hard to digest now, it will be helped by the future explosion of complex blockchain data. This kind of financial engineering will create new revenue sources for financial firms. Applying AI and machine learning will create a competitive advantage.
To be sure, these native digital assets will be highly exotic products. They will exist on blockchains and have their own economic behavior and unique cash flows. They will be created by either human or AI-driven financial engineering. Their risk, predictive and pricing models will be AI-driven because they might be too complex for humans.
These exotic and complex financial products might bring back memories of the Asset-Backed Securities, CDOs, and Credit Default Swaps that led to the 2008 financial crisis. Yes, the downside risk is there and these products must be regulated. Still, native digital assets are likely the next evolution of financial engineering and we will see them eventually.
The fourth and final stage will see AI become economic agents. AI algorithms will actively trade digital investment assets over a blockchain-powered tech stack. Evolutionary (genetic) algorithms could generate, test and trade multiple strategies, kill off under-performing strategies, and continually tweak the winning strategies to maximize trading profit. All with minimal human supervision.
In this new world, AI will create and trade digital investment assets over high-speed private blockchains. Institutional investors will buy these assets because they trust the ability of the issuing firms. This means that incumbent firms will have a huge advantage.
This future might seem hard to fathom, and the details are necessarily vague because nobody has done this yet. However, the underlying blockchain technology and AI methods already exist. We simply need increased blockchain activity, improved AI capabilities, and corporate adoption.
Remember, if you had told people in 2009 that everyone would be talking about magical internet money called Bitcoin within 10 years, you would have been laughed out of the room.
Positioning your Organization for Blockchain and AI Convergence
Specific use cases for combining blockchain and AI will depend on company needs but the underlying theme will be data. Blockchain will ensure that data is secure, private and trustworthy. AI models will use this data to become more effective.
Companies can prepare themselves to develop combined AI and blockchain solutions by improving their digital and data capabilities
Executives must first determine the specific business needs and determine whether blockchain and AI can address these needs. If they already have AI initiatives in place, they can explore how blockchain could improve them. Alternatively, companies sitting on valuable data could monetize it by joining a blockchain ecosystem and sharing data with people building AI models.
For instance, an autonomous car company could store data collected by its cars on a blockchain. When self-driving cars go mainstream, they will collect huge amounts of driving data from on-board cameras and sensors. This data is used to improve the neural networks powering self-driving functions.
Storing this data securely and maintaining driver privacy is a business need. Storing data on blockchains can anonymize driver information, ensuring driver privacy. The car company can still use the data to improve its self-driving neural nets.
From a monetization perspective, the car company could share aggregated and anonymized driving data with insurance companies. Insurers can use the data to price self-driving car insurance more intelligently since self-driving cars have a different risk profile than regular cars. In the end, driver privacy is protected, the car company improves its self-driving capabilities, and the driver may get insurance at a better price.
The reason we haven’t yet seen many examples of joint adoption of blockchain and AI is that implementation at scale is challenging. Many businesses are still in the early stages of implementing blockchain and AI in isolation. Companies are still figuring out how to structure their organizations and modify business processes for blockchain and AI.
Companies can prepare themselves to develop combined AI and blockchain solutions by improving their digital and data capabilities. Digital transformation is a precursor to AI and blockchain adoption. Managing data and business processes using digital systems provides AI initiatives with firm-wide data, enabling AI implementation at scale.
Executives must also understand how to upgrade current data infrastructure to enable future AI and blockchain adoption. They must understand what kind of data needs to be collected and where the current gaps are. Building these core capabilities is like laying the foundation for a house — it greatly improves the chances of building successful blockchain and AI solutions.
All Rights Reserved for Raj Shroff