AI and Blockchain — Intersection of Hype or Reality?
One of the areas that blockchain technology is trying to disrupt is Artificial Intelligence, which by nature is a centralised blackbox — in contrast with blockchain technology, which is decentralised and transparent. In this short article I will discuss a few main points that are at the intersection of AI and blockchain.
Key takeways:
AI applications currently cannot run natively on blockchain infrastructure due to lack of speed and the centralised nature of AI data stores.
Blockchain can be useful for edge AI applications and to establish a clear chain of provenance of data sources, which is also useful for data quality.
Crypto can be used to monetise user generated data and content and for data privacy control.
Consensus mechanisms need to be improved for scaleable AI applications to be possible. Proof-of-authority and proof-of-stake hybrids are the most likely direction.
AI requires massive amounts of data for training but can be easily trained using a decentralised system that needs to communicate quickly together. Andrew Ng, one of the thought leaders in Deep Learning, has emphasized how Deep Learning has been enabled by the availability of massive amounts of easily accessible data together with faster computation. Blockchain seems to have neither central data stores or fast computation with currently available infrastructure. Can you do machine learning without centralised harvesting of raw user data? I believe that this is one of the main challenges facing AI applications running on the blockchain — one that requires AI applications to somehow do training in a decentralised manner and that can utilise decentralised data stores spread in a P2P like network. Current deep learning system architecture is simply not compatible with such a deployment mechanism, so it is likely that the initial wave of blockchain AI applications merely use cryptocurrency for payments or some form of transactional token, while using the ledger functionality of blockchain to authenticate trust in a dataset or result. Edge AI applications, where a lot of the AI pre-processing and part of the training is offloaded to IoT devices in a decentralised manner, may be a better fit for blockchain applications — since the decentralisation part is common to both technologies.
AI can use blockchain technology to check the source provenance of a dataset which is used for training, possibly helping identify any sources of bias and correcting them. Using blockchain based permissions, the data owners can also add and revoke permission to their dataset, although current AI systems will face difficulty in quickly adding or removing data from their models, especially once they have been trained and incorporated into other models. Transfer learning, for example, does not seem to fit very well with the stated goals of many blockchain projects — this is an area that will need more research in the future to fit in well. Data privacy is also a recurrent theme in the intersection between blockchain and AI, and AI makes this a more important issue since incorrect implementations can lead to massive data breaches. Using a blockchain based data permissions model can help mitigate these risks and also allow for data quality issues to be addressed. A transparent system that tracks the provenance of data and its sources, together with a transparent rating of its quality and truthfulness would help ensure that fake news, false data and low quality information do not get inadvertently incorporated into AI models. Decentralised data permissions would also open the way for users to gain monetary value from their own generated data, enabled by micro-payments via suitable crypto tokens. This unlocking and monetisation of user data and user content follows the original spirit of Vannevar Bush’s Memex design which inspired the Internet, the WWW, Wikipedia and a lot of other modern information sharing systems.
Presently available blockchain infrastructure is too slow for large scale AI applications that would run directly on it — which negates a lot of the benefits of smart contracts and AI applications. Faster and more robust smart contract infrastructure would help unlock the benefits of having an AI system create a secure smart contract that is tailor made to a specific situation or transaction. I believe that this an area of value that may have to wait until the next generation of blockchain systems that can handle huge volumes of transactions become commonplace.
Current consensus mechanisms for DLTs — mostly based on proof-of-work — are too costly for widespread AI applications to utilise on a large scale. Work on alternative consensus mechanisms such as proof-of-stake or proof-of-authority (or some hybrid) is needed to ensure that there is a safe manner of agreeing on the values in a distributed ledger without requiring computationally wasteful and energy expensive processes that currently underlie a lot of the blockchain technology. AI itself may make newer consensus mechanisms work more efficiently — in my opinion this is a roadblock that needs to be solved before scaleable AI applications can be deployed purely on blockchain systems. A hybrid of proof-of-stake and possibly proof-of-authority seems to be the best mix for future scaleable AI applications that are compatible with blockchain technology for practical everyday usage.
In the interim, AI can add value to blockchain applications, by helping users make sense of the huge volumes of data that blockchain ledgers contain, and also by making processes more efficient and thus helping reduce the overall cost of transactions. AI and blockchain opportunities in smart contracts is one of the most attractive areas, opening up new ways of having AI integrate itself into the global economy in a controlled and well restricted manner that is beneficial for all parties.
Related Content
Some additional reading on the web based on my past interviews and conference panels.
AI and DLT technology podcast and video interview of Angelo Dalli on Carla Maree Vella’s (Carla Maree Vella) My Blockchain Island site.
Interview with EAK Digital at the Delta Summit on blockchain and AI between Angelo Dalli and Erhan Korhaliller.
Get more insight from the Malta Blockchain Summit panel on AI and Blockchain moderated by Angelo Dalli with the participation of Vince Vella, Sally Eaves, Joseph Dureau and Mark Mueller-Eberstein — watch the recorded livestream at http://bit.ly/2qJRmcO
Credits: Robot image via www.vpnsrus.com
