ChatGPT and other (perceived) breakthroughs in AI and ML (machine learning) applications are the new talk of town. Right when the investor community supposedly turned their back on web3 to ape into the new and shiny AI, a new imperative emerged: Blockchain x AI (aka Hype²).
Source: https://www.forbes.com/sites/charliefink/2023/02/16/this-week-in-xr-ai-supplants-metaverse-web3-bigscreen-unveils-sleek-vr-headset/?sh=3ffe73666f8c
Now, despite all the hype and buzzword bingo, there actually is an interesting intersection of both technologies. I personally looked for solutions combining AI models and DLT 4 years ago, but unfortunately both technologies were not mature enough, so I stepped back into a “wait and monitor for 2-3 years” status.
And indeed, beyond the memes and overhyped projects (both old and new), there are some interesting synergies, overlaps and augmentations of both technologies combined that are worth exploring for future inventions. I am not going to cover specific projects – there are other posts for that – but I will provide a quick breakdown of what we will look for in the coming months and years and where we see the biggest potential. Disclaimer: I think it will easily take another 3-5 years before any combined solutions can be deployed at scale, but that’s the usual timeframe for early-stage investments anyway.
Interesting further reads and my inspiration back then (Kudos to Matt Turck): https://www.linkedin.com/pulse/ai-blockchain-introduction-matt-turck/ ; https://mattturck.com/
Let’s get a basic understanding for AI:
AI is the ability of machines/computer programs to perform tasks that would typically require human intelligence. It uses algorithms, which are sets of rules that the computer follows to perform a specific task. There are two broad categories: weak AI (specific tasks like playing chess) and strong AI (human-like intelligence).
Here is where it gets interesting: AI requires large amounts of high-quality data to learn from, as well as the ability to process and analyze that data quickly. For example, computer vision requires large datasets of labeled images or videos to learn how to recognize and classify objects. The quality of the data is just as important as the quantity – it must be accurate, representative, and free from biases. Data preparation (structuring), cleaning, and preprocessing are very resource intensive.
Algorithms are trained with machine learning, which involves multiple steps and works like training a dog (or your kid) with incentives and penalties in an iterative process – gladly, it doesn’t involve cleaning poop.
Prep work: Data collection and cleaning – Data labeling (e.g. image labeled with the object it contains) – Training data selection (the more data used, the better) – Algorithm selection (based on the task).
Training process: Training (algorithm adjusts its internal parameters to better fit) – Model evaluation (trained model is evaluated with unused labeled data) – Model refinement (Retraining using additional data or different parameters).
Data quality is critically important for the accuracy and reliability of AI algorithms. Poor-quality data can lead to biased, inaccurate, or unreliable AI models and thus wastes resources and time in preparation steps.
Where does Blockchain come in?
For a broader definition of Blockchain, DLT or Crypto, please refer to google or ChatGPT. Let’s cut to the chase: Data is used in blockchains to record transactions and other types of information (like code or identities) in a secure, transparent, and tamper-proof manner – maintained by a network of nodes. Cryptographic hashes ensure the integrity of the data in the blocks (like a checksum), i.e. once a block is added to the blockchain, it cannot be modified or deleted.
The secure and transparent recording of data is one of the key features that makes blockchains so promising for a wide range of applications, including AI. Blockchains can help to ensure data quality by providing a system for recording and validating data. Multi-party validation and decentralization make it harder to compromise data once on chain. However – shit in, shit out still applies (aka the Oracle problem).
How can blockchains and AI complement each other?
Blockchains can provide a secure and transparent way to store and share data (privately) through encryption and access control mechanisms. For example, healthcare AI algorithms may require access to sensitive patient data, and blockchains can provide a secure and transparent way to share this data.
That way, blockchains can serve as a backend solution for multi-party AI training by allowing data sharing with full provenance and increased reliability of datasets through validation and notarization of data (think of a “quality consensus”). Immutability of data provides an additional security layer for data labeling and AI inputs. A blockchain-based system to record the history of datasets allows organizations to ensure that the data used for training is accurate, reliable, and trustworthy
Decentralized marketplaces for AI algorithms and data can be created without requiring a platform owner (DeFi for data – let’s call it DeDa©), so that organizations can securely and transparently trade and exchange datasets without intermediaries. In addition, smart contracts might be able to automate at least parts of the training process to reduce the time and cost of training AI algorithms. If the training itself is run on a decentralized computing network (reminds me of Skynet…that didn’t end so well for humanity though), resource sharing could bring costs / time down even further.
Tokens can be used to provide incentives and reward economies, for example by representing the value of data and allowing providers to monetize the data they collect and share, thereby incentivizing high-quality datasets. Smart contracts could also reward AI trainers for achieving certain performance metrics or alternatively reward the contribution of computing resources or expertise in collaborative training scenarios.
On the other end, AI can be used to improve the efficiency and security of blockchains by leveraging AI algorithms to detect and prevent fraud, detect anomalies in the network (hello, DeFi exploits!), and optimize the performance of the blockchain.
Going a step further, I could even envision baking predictive analytics into the consensus mechanism to increase throughput with dynamic load balancing based on continuous analysis of transaction data on a blockchain. In the same manner, AI can enable the identification of patterns and trends within a blockchain network.
Verdict
There you go, behind the fluff and fairy dust are actual use cases that have the potential to advance both technologies by working in tandem. And these examples are surely not exhaustive. Time, tests and multiple failures will tell which parts are fiction and which have solid footing. Some of these visions are even scary to a point that The Matrix and Skynet become reality, machines will enslave us, and we will all be begging for the red pill. Well, in a way we are already enslaved since the dawn of smartphones and social media, so how bad can it really be?
Be that as it may, if you are working on a fantastic solution that combines Blockchain and AI and you think it has the potential to save or end mankind, please reach out. If you got inspired to build impressive tech, kindly leave a honorable mention. You can also reach out if you just want to exchange opinions.
