AI & Blockchain: The Intersection of Top Tech Trends
AI & Blockchain: The Intersection of Top Tech Trends
Artificial intelligence and blockchain are the two most talked-about tech trends of the past two years by a wide margin. There’s good reason why these technologies have received so much attention. AI promises to automate many tasks and is often better at modeling complex situations than humans. Blockchain, on the other hand, offers greater data security and privacy while also reducing the overhead and centralized power of major institutions.
Combining blockchain and AI together may sound counterintuitive or like a pitch for the next big scam ICO. However, there are real reasons why these two technologies might work well together. They each address the flaws of the other, balancing the worst tendencies of each. As we’ll see in this article, AI of the future may very well rely on datasets stored on the blockchain and distributed computing based on blockchain innovations. In turn, blockchains could use AI to monetize user-controlled data, create a marketplace for AI models, and even create autonomous organizations.
In this article, we’ll explore the ways these two technologies are currently interacting. It’s important to keep in mind, however, that both AI and blockchain are in their infancy. There will be many developments over the coming decades that revolutionize the ways we think and talk about these technologies. Combining the two will not be without its challenges. To be sure, the weaknesses of blockchain as a data storage and retrieval system make it harder to use than a traditional database for AI applications. Additionally, hard coding a smart contract or decentralized organization is much easier than writing an AI that can do those things itself safely. Nevertheless, if we can get to that point, the potential opportunities are exciting.
Before we get into the potential applications of blockchain and AI together, let’s discuss why it’s so difficult to mix the two. In many ways, blockchain and AI are like oil and water. They work in such different ways and are philosophically opposed to each other in many of their aims. The contrast between the philosophies is what makes the potential combinations so enticing. It’s also the biggest hurdle to development of these new applications.
1.1 Centralized vs. Decentralized
The first major philosophical difference between AI and blockchain is the locus of power in the system. AI relies on huge, complete-as-possible data stores and massive computing power to train algorithms. As a result, the companies with the most data and resources get the most out of AI. In turn, they are enriched by AI’s advancements and have more resources to invest in building better AI. As such, AI is a centralizing technology. It incentivizes consolidation of data and computing power. One common concern is that AI will make the rich richer and it inherently creates income inequality for the poor who can’t take advantage of AI.
Blockchain could address this philosophical challenge of AI. Blockchain ledgers decentralize control of data and computing resources while still making the data and resources available to the overall network. Of course, this decentralization comes at the cost of network latency, and serious efforts would need to be made to speed blockchain ledgers up if they’re to be used alongside AI. However, there is promise in the idea that consumers can still own their data or computing power and rent it to major corporations as needed.
Smaller companies, governments, and even NGOs could also take advantage of user data if that data were stored on a blockchain. Access to decentralized computing power would also make it possible for anyone to run AI models. The upshot is blockchain could democratize access to AI, allowing anyone to develop and use AI models on real datasets from worldwide users.
1.2 Transparent vs. Black Box
Another philosophical difference between AI and blockchain is how they treat transparency. Blockchain’s founding principle is transparency. The ledger of public blockchains is open for anyone to see. While data has been anonymized, it’s still available on the ledger and the blockchain is transparent about how transactions get added. By nature, the open peer network creates trust between parties using publicly understood cryptography.
In contrast, machine learning algorithms and neural networks are notoriously hard to understand. While certain types of statistical algorithms require an advanced understanding of linear algebra, other deep learning techniques are completely a black box, even to the researchers who develop them.
This transparency problem in AI isn’t a problem that blockchain can solve. It’s simply the nature of deep learning and our current level of understanding about how algorithm training works. Perhaps future research will lead to greater understanding of these algorithms.
What blockchain can solve, however, is public access to the data these models are trained on, which can shed important light on the potential weaknesses or biases of a model. After all, an AI model is only as good as the data it is trained on. Public, independent audits of training data is key to equitable AI, and blockchain data storage could facilitate that.
2. Potential Applications of Blockchain & AI Combined
Philosophically, blockchain and AI pull in opposite directions. AI focuses on fast, complex insights that rely on massive data and computing resources. Blockchain can provide those data and computing resources, but it will be slower, more transparent, and decentralized about the process. Speed is currently a bottleneck, but blockchain scaling solutions currently under development could serve data more quickly.
When combined, these conflicting but powerful technologies could enable several interesting applications. Generally, these applications focus on democratizing the benefits of and access to AI.
2.1 Distributed Computing for AI
One of the leading potential combinations of blockchain and AI is using the idea of mining networks as a way to source computing power for training algorithms. Algorithm training in AI and machine learning requires intensive CPU utilization in order to compute thousands or millions of training sessions that teach the algorithm how to make better decisions. Waiting on algorithm training is one of the major slowdowns facing machine learning development. If data scientists and AI researchers had access to more computing resources, they could develop algorithms faster.
Several startups are working on the idea of connecting AI researchers with the thousands of GPUs out there that are currently mining cryptocurrencies. At times, AI researchers may be willing to pay more than miners would make from mining cryptocurrency. It could make sense for the same GPU networks that currently power proof-of-work to power the next generation of algorithm training.
If successful and affordable, these training networks could make algorithm training available to anyone. Currently, if you want to develop serious AI algorithms you need access to centralized server farms. Instead, using decentralized GPU mining networks, anyone anywhere in the world could have access to a supercomputer to train their algorithms.
2.2 Keeping Data Private
Blockchains anonymize data. While there are still ways to deduce what data belongs to which person, this anonymization makes a blockchain ledger a great place for research. Instead of centralized institutions like Google or Amazon gathering data on millions of users, anyone with access to a blockchain ledger could use anonymized data available there to create analyses, make predictions, and even train algorithms.
Using wide-reaching, decentralized, and anonymous data could also make the algorithms trained on that data less biased. Instead of Google’s algorithms being trained on Google’s mostly white and mostly western user data, global anonymous blockchain data could allow Google to create algorithms that are more representative of the global population as a whole.
2.3 A Marketplace for Algorithms
As soon as you democratize access to computing and data, AI begins to become a community-driven development opportunity. Already, we’re seeing open source packages for common machine learning models like TensorFlow or PyTorch. Blockchain based data and computing could explode this trend and create a marketplace for algorithms. Companies could pay data scientists for their algorithm’s insights or buy access to the algorithm outright for ongoing use.
Instead of AI development being internal and proprietary, it could be an open marketplace where anyone can participate and contribute value. As we’ve seen with the open source movement and the rise of as-a-service software and infrastructure, these types of marketplaces encourage innovation much more than private development.
2.4 A Marketplace for Data
Another side effect of blockchain records is users gain more control over their data. Currently, Google keeps your search history as well as all your interactions with Google products like YouTube, Google Maps, and Android devices. You gave that data away for free, and now Google will use it to create more value and make more money.
A blockchain-based infrastructure could give you back control of your data. Perhaps you could share information about yourself with Google in exchange for personalized services. At the end of the day, though, Google wouldn’t own the data you shared, and you could revoke access at any time. This opens the door for you to sell access to your personal data. Corporations might pay consumers for their information in order to drive insights. Such a marketplace could provide everyone online with a passive income stream just from existing and creating data worth selling.
2.5 Decentralized Autonomous Organizations
The most distant but also the most revolutionary application of blockchain and AI combined is the idea of an autonomous organization. Right now, decentralized autonomous organizations (DAOs) already exist, but their rules are hard coded as smart contracts. The organization can perform actions, but it doesn’t make independent decisions. It only follows the rules the smart contract developer wrote.
If we add AI algorithms into the development of DAOs, however, independent decision making becomes a possibility. These organizations could make business or philanthropic decisions based on market data. They could acquire resources and decide on the best ways to allocate those resources given the context. We move toward companies and organizations that can run themselves without any outside intervention.
Since these companies run on a decentralized network, they’d be nearly impossible to shut down. So, there’s significant danger in creating them. However, if built well, they could be the key to good governance, economic equality, and reduced workload for human knowledge workers.
The combination of AI and blockchain holds many exciting potential applications. It’s worth remembering, though, that these technologies are still very young. Most of these hybrid applications will take years to develop or may not even develop at all. They’re certainly powerful ideas, but researchers and developers should take great care when working on such projects. The ethical implications of both blockchain and AI are massive, and the stakes are high for the economy, governance, and society as a whole.