I. Setting the stage
The IoT and consumer hardware industry have seen multiple failures and a few exits over the last 12–18 months (while the B2B side has been doing a bit better overall) and some critics have been recently made to the industry to slow down.
In spite though of the current pushback, the sector is still increasing and attracting capital and talents. Clearly, there are multiple reasons on why this is the case, but I firmly believe that one of those is the convergence of IoT and Artificial Intelligence with the Blockchain as the infrastructural backbone, which is unlocking the next step not only on the tech-side but also on the business one.
The industry has indeed evolved from merely creating products, to create networks of products (namely, Internet of Things), to eventually create Intelligent networks of products (I-IOT). The transition between the first and the second class was straightforward: I will create more (different) products and try to link them together. This generated many new possibilities, but it was clear from day one that it came with a series of issues hard to tackle, such as security/privacy issues, validation/authentication problems, and connectivity bottlenecks.
And this is where AI and Blockchain come in. The second transition indeed was made possible through a combination of improvements in computing powers, device miniaturization, ubiquitous wireless connectivity and efficient algorithms (Porter and Heppelmann, 2014). The new class of smart products will be (and already are, to some extent) able to monitor, control, optimize, and automatize processes and products with an accuracy previously not imaginable.
Of course, as often happens, the bonus on integrating those fundamental technologies was that they ended up modifying IoT as much as IoT was impacting them in turn.
This convergence is however not accidental, but rather an inevitable necessity almost designed by default:AI needs data, IoT needs intelligence and insights, and both of them need security and transparent marketplaces.
The magnitude of this convergence is so high that will affect several sectors swinging from energy and manufacturing to home environment, robotics and drones, supply chain, logistics, and healthcare. Every field which is historically data-rich but information-poor will be touched (or should I say brutally hit?) by those technologies.
I will explore how in the next few sections.
II. How Blockchain is changing IoT
The blockchain as a technology is basically providing the IoT stack with a secure data infrastructure to capture and validate data. As simple as that. Or at least it is a simple statement that contains three different nuances:
- Securing data better: the first one is indeed the concept of storing data securely. We know that blockchain protocols are not designed to heavily store data (they are indeed ledgers, not databases), but they can provide “control points” to monitor data access (Outlier Ventures, 2018);
- Creating the right incentive structure: a blockchain can create the right incentive structure to share IoT data, which is something we are currently missing. Cross-sectional data have been proved to have the most disruptive impact when applied across different industries, but the problem of how and why sharing data in the first place remains. Blockchain (and tokenization) can be used to solve this economic dilemma, and once data are shared can be more easily validated, authenticated and secured;
- Creating a network of computers: distributing the workload and implement parallel computing tasks is something it is usually attributed to new AI or High-Performance Computing (HPC) applications, but a blockchain would be essential in this development authenticating and validating the single nodes of those networks (companies that are working in this space are Golem, iExec, Onai, Hadron, Hypernet, DeepBrain Chain, etc.).
III. How Blockchain can change AI
AsI have already previously mentioned, blockchain can affect AI in multiple ways:
- Help AI explaining itself (and making us believe it): the AI black-box suffers from an explainability problem. Having a clear audit trail can not only improve the trustworthiness of the data as well as of the models but also provide a clear route to trace back the machine decision process (i.e., where data are coming from, who wrote the original algorithm, what data was used for training, etc.) and can establish the foundations for “algorithms standards” (e.g., which main algorithms, packages, and framework have been developed using a specific training set). This is also essential in machine-to-machine interactions and transactions (Outlier Ventures, 2017), and provides a secure way to share data and coordinate decisions, as well as a robust mechanism to reach a quorum (extremely relevant for swarm robotics and multiple agents scenarios). Rob May expressed a similar concept in one of his last newsletters;
- Increase AI effectiveness: a secure data sharing means more data (and more training data), and then better models, better actions, better results…and better new data. A network effect is all that matter at the end of the day (AIBlockchain is an example of a multi-application intelligence that uses different sets of data);
- Lower the market barriers to entry: let’s go step by step. Blockchain technologies can secure your data. So why shouldn’t you store all your data privately and maybe sell it? Well, you probably will. So first of all, blockchain will foster the creation of cleaner and more organized personal data. Second, it will allow the emergence of new marketplaces: a data marketplace (low-hanging fruit, currently pursued by companies such as Ocean Protocol, OpenMined, Neuromation, BurstIQ, AtMatrix, Effect.ai, Datum, Streamr, Deuro, Datawallet, etc.); a models marketplace (e.g., Dbrain, etc.); and finally even an AI marketplace (examples are SingularityNET, Fetch.ai, doc.ai, Computable Labs, Agorai, and similar). Hence, easy data-sharing and new marketplaces, jointly with blockchain data verification, will provide a more fluid integration that lowers the barrier to entry for smaller players and shrinks the competitive advantage of tech giants. In the effort of lowering the barriers to entry, we are then actually solving two problems, i.e., providing a wider data access and a more efficient data monetization mechanism. It is also possible that a blockchain-enabled AI will eventually create new organizational structures for intelligent agents to cooperate or compete;
- Reduce catastrophic risks scenario: an AI coded in a DAO with specific smart contracts will be able to only perform those actions, and nothing more (it will have a limited action space).
IV. How AI can change IoT
AI is feeding itself with the new stream of data coming from the physical world and the billions (if not trillions) of sensors and “things” that are capturing and monitoring everything we do.
At the same time though, as soon as an AI starts making sense of IoT data flows, it will:
- Increase data efficiency: an AI will inform back those sensors on what data should be captured and stored, and above all where those sensors should be placed to be both more efficient and more effective;
- Save costs: it is fair to think that an algorithm performance should be tested continuously, and once reached the optimal level with data marginal return approaching zero (i.e., a point in which adding more data does not improve the prediction outcome), an AI will not store or capture more data, resulting in energy, servers, computation, cloud, and infrastructural savings. In addition to that, unplanned downtime prediction is a second cost saving possibility an AI will open for an IoT ecosystem;
- Increase security: an AI could clearly be able not only to fight potential external threats for an IoT network but even predict them (e.g., AnChain);
- Compute on-the-fly: edge/fog computing is quickly becoming a hot topic since it allows on-device computation, which in turns reduces the response time for an action, limits the exposures to privacy and compliance issues, and solves the huge connectivity bottleneck problem. A few startups are already working in this direction (Foghorn, Mythic, Neureal, SONM, Nebula AI, etc.) as well as big incumbents as Google (who recently released, in addition to federated learning, an entire stack made by an Edge TPU and a Cloud IoT Edge platform), but likely things will change here due to the rapid development of specialized training and inference chips and the forthcoming introduction of the 5G. Cloud is still necessary for computationally intensive operations and to store data centrally to guarantee (maybe?) an extra layer of security (meaning, network disaster recovery), but custom chips and edge computing algorithms can do most of the operations the final customer needs directly on the device.