How Can Blockchain & AI Come Together?
Blockchain and AI are the two technologies set to change the world, but what can they do together?
by Juan Aranovich, editorial research and Adam Kreitzman, intern
Intro
ChatGPT has taken the internet by storm over the past month, as social media platforms are flooded with examples of just how close we are to having AI disrupt a significant amount of industries.
Blockchain and AI are widely considered to be the two technologies that will shift society in the coming years, but there hasn’t been much talk surrounding the intersection between the two and what it could mean for the future.
While there exists a wealth of potential for both technologies to positively impact each other, the crossover is currently hamstrung by limitations around scalability, infrastructure, and the challenges of developing novel technology.
How are the challenges facing AI and blockchain similar?
The similarities in the challenges that developers face when it comes to building the two technologies is uncanny. Both have question marks surrounding the scalability of their applications, the trust that humans have in the technology, and a sizable moat around them that can make gaining additional users challenging.
ChatGPT set a record of being the fastest platform to accumulate one million users after it reached the number five days after launching. However just days after, it became difficult to access as it would reject users due to their servers being full, despite the fact that language models are some of the easier ones to scale considering the simplicity of text data compared to other forms such as 2-dimensional or 3-dimensional imaging. As data becomes more complex and multidimensional, so do the models and their applications.
On the blockchain side, issues with the scalability of smart contracts and blockchains in general are highly documented, with much of the focus being on upgrading infrastructure to allow for broader use. There are attempts to scale Bitcoin with the Lightning Network, and to scale Ethereum with layer 2 technologies, as well as several other approaches throughout the blockchain ecosystem such as Cosmos app chains and Avalanche subnets.
Compounding the two technologies together only amplifies the challenges of each, as it requires the deployment of AI models within the limitations of blockchain.
Ethical considerations of AI
Perhaps the biggest takeaway from the DeFi space, and crypto in the broader sense, is the ability to bring equitable, decentralized ownership of assets to the world, while enabling everyone to participate in all facets of a digital economy.
For AI to play a large role in this, it is imperative that it maintains neutrality, but studies we have seen so far have all shown the exact opposite.
In fact, much of the research revealed a bias against particular regions and ethnicities, and even particular languages, as studies suggest it struggles with Arabic. Understanding why this is the case requires looking at how AI models are built, which is that they emerge out of the data that they trained on. In the case of language models, the most advanced AI is being developed mainly with either the English or Chinese language.
But this is just one example, and there are others where the ramifications could be extremely costly. For instance, if AI models were to be trained on patients to try and diagnose disease based on genomic data for instance, it could be prone to misdiagnosis if applied to populations it was not trained on. Recognizing that AI is not an abstract entity with knowledge but rather a computer that is trained to recognize specific patterns within a set of data is essential for understanding the consequences that could arise from its use.
Two possible ways of applying AI with blockchain would be using it for authentication purposes and market optimization. This could include anything from fraud detection to the optimization of structuring liquidity.
But these are sensitive subjects to integrate AI within, as failing to properly distinguish fraud for certain groups or having liquidity structured disadvantageously for certain people both have negative ramifications for imposing fairness in a digital economy.
Bias already exists within all systems and societies to some degree, so AI is not necessarily introducing a new problem, but its use can either mitigate or perpetuate existing biases.
Roadblocks
The application of AI relies primarily on three things: sufficient data for modeling, a clear use-case with predictive features, and the infrastructure necessary to deploy it for use. As it stands today, all three factors pose challenges in the blockchain ecosystem.
One of the pillars of blockchain technology is the transparency and open-source data of all transactions that happen on top of it. It yields lots of possibilities for how it could be utilized in the future. For instance, the data of every single Ethereum block containing transactions, all of which are time-stamped. Additionally, oracle price feeds give us historical price data for all cryptocurrencies. There is not a lack of available data, but just because data exists does not necessarily mean that there is a clear AI-driven use for it. On top of that, aggregating available data can be quite a challenge, not to mention expensive, as accessing the requisite APIs for high-quality blockchain data costs thousands of dollars a month.
Another roadblock is the lack of obvious use-cases for AI in crypto. Additionally, a lot of research is required to figure out whether an AI-based approach can ultimately be successful. Launching a successful AI application requires identifying product-market fit and actually developing a solution that serves it. AI still has yet to make a significant inroads on its own, so finding a solution that couples the two together is an added challenge.
Infrastructural development is the aspect that can be considered the most mature at present. There are several projects that are attempting to provide infrastructure for deploying AI on-chain or using blockchain to facilitate the use of AI.
Developers
Another significant challenge in the intersection of blockchain and AI is finding developers who have a deep understanding of both technologies. Many developers specialize in one area or the other, but the intersection of these two technologies requires a unique combination of skills that are difficult to find. This can make it challenging for companies looking to build decentralized AI applications to find the talent they need to get their projects off the ground.
Another obstacle is that the main driver for machine learning is Python, which may not be the best language for blockchain development. This creates a disconnect between the two technologies and makes it difficult for developers to integrate them seamlessly. To overcome this challenge, some companies are developing new programming languages that are specifically designed to work with both blockchain and AI. For example, the Seahorse Language, which is being developed on the Solana blockchain, aims to provide developers the ability to write Solana programs in Python in order to bridge the gap between web2 and web3.
What is being built?
While there are certainly challenges to developing on-chain AI, that doesn’t mean that it isn’t happening. Several teams are currently working on AI-powered blockchain solutions and infrastructure.
Some projects are AI-based protocols that attempt to improve the user experience of using blockchain, while others are blockchain-based protocols that attempt to improve the user experience of AI. The fact that both infrastructural improvements are possible with the technology of each is a good testament to the potential impact AI and blockchain will have on each other in the future.
As of now there are three major projects that contribute to infrastructural development, which are Fetch, Ocean, and SingularityNet.
1. Fetch
One of the biggest projects is Fetch.ai, which is a proof-of-stake blockchain built with the Cosmos SDK that allows people to seamlessly deploy digital twins and set up Autonomous Economic Agents (AEAs). It is also available for use on both the Ethereum blockchain and the Binance Smart Chain. By setting up AEAs and digital twins, Fetch believes people will benefit from Impermanent Loss protection, will be potentially protected from instances of rugpulls, and even be able to take advantage of arbitrage opportunities. Their platform is geared towards helping users with asset management and curbing the risks present when the user is unable to instantaneously access funds.
Fetch has seen an uptick in active users recently, with its highest usage coming in December of 2022. While these figures are consistent with the increased usage in dApps that has happened over the last month, it is surprising to see 2021 level activity, which has not been the case throughout most of crypto. Fetch represents the most advanced AI tool that has integrated within blockchain in terms of its finished product and usage, which may very well be why it is the highest protocol by market cap. Being able to automate aspects of risk management when markets are 24/7 is a key area of development that can protect users when they are inevitably away from the market.
2. Ocean
Another major protocol that has seen traction is Ocean Protocol, which allows people to make their private data accessible for use on the platform. Accessing useful information can be challenging for anyone attempting to harness machine learning, and Ocean tries to solve this problem by allowing data sharing with its compute-to-data method which allows for remote computation to happen on these datasets. Those who create/share high-quality datasets are rewarded with Ocean tokens from users who access the data, manufacturing a new market powered by blockchain that enables decentralized data sharing and allowing users to profit from datasets they own or create. It also maintains the privacy and security of the data because it never leaves ownership of the creator or is exported from the server. This is useful because oftentimes there are tons of compliance issues with sharing data, and this alleviates the problem by allowing for remote access to data while never actually letting the platform users gain ownership of the data.
Since the beginning of 2023, there has been a slight increase in Ocean Protocol usage, but the weekly usage has seen a major decrease ever since it originally launched back in 2020, when there were thousands of unique users. It is not out of line with dApp usage across the entire crypto landscape, however, as many have fallen off substantially since early 2022.
3. SingularityNet
Rounding out the three AI-driven protocols with the most traction is SingularityNet, a platform that aims to facilitate the connection of AI services to those who need it. It acts as a decentralized exchange where anyone can list their own AI service for others to purchase. Additionally, it rewards tokens to people who lend processing power to the network in order to help the AI networks run. Effectively, SingularityNet aspires to be a hub of AI activity that brings services to users while helping complex networks execute tasks.
These charts all tell the same story, albeit to different degrees, of AI protocols seeing increased usage since the end of 2022/beginning of 2023. It is important to note that the release of ChatGPT, as well as news that Microsoft is investing 10 billion dollars in the technology, spurred conversation into blockchain projects connected to AI. Seeing whether the trend keeps pace or plateaus will be key to understanding whether the jump in usage was purely speculative or not.
Other potential use-cases
One key technology that has shown promise is zero-knowledge proofs, which have the potential not only to speed up blockchain transactions, but could also have ramifications for increasing security of data with its encryption. Being able to improve data security would have major ramifications for the potential of zero-knowledge encryption to impact the data and AI space moving forward.
Another way blockchain can facilitate more data security and improve machine learning output is through a process called decentralized swarm learning. Similar to how Ocean Protocol maintains data privacy and security, swarm learning is a practice of machine learning model communication that happens across a blockchain network in order to decentralize the sharing process. The idea is that in instances where data sharing is impossible due to compliance issues, models which train separately can combine and adjust their weights in order to incorporate input from all datasets into the final model. Studies have shown that this method can be just as effective as training models on all datasets locally. What is interesting about this approach is that people can unite from different places to form models that are effective in the real world while being rewarded for the significance their model adds. This framework will allow a truly decentralized and merit-based way for people to collaboratively develop AI applications.
Conclusion
The synchronous development of AI and blockchain technologies is starting to take form, and while it is in its infancy, AI infrastructure and access to on-chain data are both improving and may potentially lead to breakthrough applications in the future. While there are certainly challenges to the integration of the two technologies, both have the focus of developers who are trying to scale them and search for real-world cases to apply them to. At some point, there is bound to be an overlap of the two as scalability and real-life uses come into fruition for both.