How two disruptive technologies empower each other
Blockchain and AI
Artificial intelligence and blockchain technology — aside from the Internet of Things (IoT), no other technologies are currently receiving as much attention, from both futurists and big businesses alike.
But while much has already been written about the potential and the ramifications that these technologies can have on their own, what about their combination? Can these two technologies actually join forces? And if so, how can they serve to help each other realize their full potential?
This article is based on extensive research into current academic and industry-wide debate over exactly these aforementioned questions. Targeted to a beginner’s audience, let us now discuss why blockchain and AI can work together really well, and what their synergistic, combined potential could mean for technology, business, and society as a whole.
Before we combine these two technologies, however, let us briefly discuss and achieve a common understanding of these two technologies.
The blockchain is akin to a linked list, or table, that is stored on all nodes of a network, with each participating node storing the full, identical, real-time-updated version of the database. The table/database contains transaction data, with a timestamp and the respective wallet addresses, between which transactions have taken place.
Transactional data is not stored raw, but multiple transactions are packed together in a block, which is hashed/encrypted using a hash function. Each block contains the hash of the recent transaction data, as well as a reference to previous blocks, creating a chain of blocks, explaining the origin of the name “blockchain”.
Transactions are verified through a consensus mechanism between all participating nodes, where at least 51% of them have to approve/validate a transaction in order for it to be accepted. As transactions are signed by a user’s private key and identified by public key, and also due to the encryption of blocks, records on the blockchain are immutable and tamper-proof.
The main features of the blockchain are encryption, immutability, decentralization (though not necessarily), deterministic logic (valid/invalid), and attack-resilient (Salah et al. 2019). Currently, the biggest issues with blockchain are security, scalability and interchain/cross-chain compatibility, as well as speed.
While the term “AI” dates back to the infamous Summer Conference at Dartmouth College in 1956, the real discussion and development of AI has gone on to gain massive traction since the early 2000s. It is generally agreed-upon that the advent of big data and massive computing power are especially the real enablers of progress in the field of artificial intelligence.
This article follows the Oxford Dictionary’s definition that AI describes “the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision making and translation between languages.”
Through machine learning and training with data sets, algorithms are developed, which continue to get better and better at recognizing certain patterns, thereby making a successful action/categorization more likely, as the algorithm is further developed.
Five key characteristics of current AI models are that they are centralized, changing, probabilistic, volatile, data, knowledge and decision-centric.
It is important to distinguish between what Goertzel (2007) consider to be narrow AI vs. Artificial General Intelligence (AGI). Currently, practically all AI models are narrow AI –designed and trained to perform one single human skill at the best possible level, e.g. the recognition of cancer tumors from the analysis of x-ray scans.
However, the real goal of AI should be to develop AGI — a machine system that is capable of acquiring and applying knowledge, being able to reason and think in a variety of domains (Goertzel 2007, p. 6), and able to solve problems like a human can. It furthermore includes the capability of using general and specialized knowledge, the ability to learn from humans, the environment, and other intelligent systems and performs experiential learning — growing ever-better at solving novel types of problems, the more it is exposed to them (Goertzel 2007, p.7).
Biggest AI Issues at the moment are…
Overall, the biggest issues that AI on the way to AGI currently face are
· Security: a massive concern in empowering AI is that, at some point, it will take autonomous decisions and actions which will hurt individuals or society.
· Transparency/Explainability: another big concern in empowering AI and “setting it free” is that it will make damaging decisions, and that these decisions cannot be understood or comprehended afterward in order to be possibly corrected.
· Centralization: most AIs are owned by, and stored, on the servers of single companies, and usage is restricted to paying subscribers, with the AI not being available to be used, fed and trained by a larger amount of people.
· Costs: the usage of centralized servers and computing power is expensive, yet it is even more costly for companies who are not generating relevant data themselves to get access to relevant data for training algorithms/AI.
· Lack of data (quality and quantity): the quality of AI algorithms cannot be any better than the quality of data with which they are being fed and trained. Since the quality of training is also a result of the quantity of useful data, a lack of data quality and quantity limits the quality of AI algorithms that are developed. And, as most data is centralized in companies’ data silos, this is a massive issue for AI development.
· Monetization: developed AI algorithms have a hard time monetizing their existence, as they will have to build a service model around it.
Blockchain for AI
Based on my research, experts agree that blockchain holds the potential to aid AI and solve some of its problems. In essence, the ways that blockchain enables and empowers AI can be summarized as follows:
Blockchain is the decentralized, secure infrastructure for the data that will make AI reliably smarter, autonomous and comprehensible to outsiders within predefined limits.
In more detail, the ways in which blockchain technology can strengthen and promote the growth of AI adoption and refinement are as follows.
· Refinements of artificial intelligence through permissioned access to high-quality data: for AIs to get better, they need access to high-quality data. As some of the most valuable data in the world are sensitive and personal, permissioned access through a blockchain-based access management layer could finally allow for permissioned and traceable access to those data treasures, e.g. for medical records and data, AIs could train and have their skills refined in the detection of tumours, cancer, etc. (Salah et al. 2019, p. 10143).
· Limit AI activities through smart contracts: the reason we don’t have an AGI right now is the fear of letting AIs roam freely and things getting out of control. An idea proposed by various researchers is that by setting up AIs using blockchain-based smart contracts, the AI will be limited to actions enabled within the framework of the programmed smart contract. In other words, each AI decision and action would represent a transaction, only one permitted by the code for seen could actually be initiated. An AI skill/application would equal a smart contract.
· Creat “explainable AI”. The blockchain could allow tracking & understanding of AGI deductions: if we are to let AI roam freely, we want to be able to understand the directions and inferences it makes to arrive at its decisions. Both so we can understand errors if aborting the AI should be necessary, but also so we can profit from AIs’ new learnings and potentially even transfer them to another field of study. If an AI is based on the blockchain, all its deductions and inferences will be logged as a tamper-proof transactional record (Dinh/Thai 2018, p. 51). And, unlike human beings who can be masters of self-deceit, an AI won’t be able to cover the tracks of its own thinking.
· Blockchain layer to secure IoT for analysis: The Internet of Things enabled by sensors and smart devices allows for the collection of billions of data points in real-time, providing valuable training data for AI models. Yet, IoT devices in unsecured networks are vulnerable and easily exploited and used in DDoS attacks. Blockchain technology could secure IoT devices implemented as an access management layer. The control of devices would only be able with a majority consensus being achieved, hence hackers would have to hack more than 51% of nodes in the network (Dinh/Thai 2018, p. 51). Through such an access management layer, data sharing with AIs could be realized, enabling all parties to access and trust measured data in real-time.
· Democratization of AI: by using an open-source platform like ChainIntel, AI models can be run in decentralized apps being executed on local devices using distributed AI execution (Salah et al. 2019, p. 10139). Thereby, AI functions like facial recognition or semantic analysis can be conducted in dApps, computational network and storage space can be provided by Ethereum or IPFS blockchain (Salah et al. 2019, p. 10139). This decentralized, distributed approach is a promising, powerful alternative to the current titan-dominated monopolization of AI through Google, Microsoft, Facebook and the like.
· Allow financial compensation for AI services rendered/permissioned data sharing: connecting data and AIs through a blockchain layer and a specifically designed cryptocurrency/-token could propel AI growth. For private or business applications, it could allow companies to buy “AI-as-a-service” where AI service providers are paid for execution of analysis of private data and/or the execution of novel machine learning algorithms (Salah et al. 2019, p. 10143). This would also allow a certain democratization as SMEs could also access AI on a pay-per-use-basis. Exactly this is the vision of KaasY project, which pursues the goal of establishing an AI-as-a-Service marketplace. Yet, for data owners, too, such a token would be powerful, as it could allow them to be compensated for granting access to their data to machine learning algorithms against payment on a permissioned basis (Dinh/Thai 2018, p. 51).
AI for Blockchain
· Monitoring incoming transactions and activity: AI can help in bringing more security to the blockchain by monitoring incoming transactions and node activity, in case they need to detect if there are any abnormalities or irregularities, such as transaction spoofing or an attempt to hijack any part of the network (Dinh/Thai 2018). If given this ability to detect such irregularities in time, AI should then be able to trigger mechanisms and warnings early on, before a potential hack may have a chance to succeed. While this use case has to be explained in further detail, the hope is that dangerous blockchain hacks can be prevented early on.
· Monitor and analyse transaction patterns: Another beneficial way of employing AI on the blockchain is to have it monitor and analyse real-time and post-transaction patterns — within-chain, on-chain, off-chain and cross-chain, in order to detect and highlight potential inefficiencies and irregularities. AI could help to identify potential for optimization in complexity scaling and hashing, as well as the general functioning of the consensus mechanism; transactions between parent and child chains, as well as main chain and side chains especially seem to be meaningful areas to be analysed and investigated by AI. This use case can be a helpful contributor to relieve at least some of the scalability issues currently pertaining to the blockchain.
· Increasing data security on the blockchain: We have discussed that blockchain can bring security to data to be used for AI, but AI can also help data security and the blockchain. Assuming that data is no longer shared with Facebook & Co, personalization would then be lacking from social media. A solution is to have AI run on the users’ devices, analysing their user and browser history. From there, relevant content can then be pulled (vs. current “push-model”) to the users. As this analysis is done locally, personal data does not leave the device, and sanitization by AI could then provide further protection from personal data theft by content providers (Dinh/Thai 2018).
· Expand smart contract functionality: AI could expand the current functionality of blockchain-based smart contract from simple if-then-else and deterministic logic to more complex situations. Dinh/Thai see the opportunity that AI could analyse provided legal documents to make automated decisions which would be tamper-proof, data-driven, and hence more unbiased than those that are currently made by humans (Dinh/Thai 2018).
Existing Projects & Initiatives Combining Blockchain & AI
· SingularityNET: Started by AI veteran and AGI ambassador/evangelist Ben Goertzel, SingularityNET is on a mission to realize the promising concept of decentralized AI. By enabling any developer in the world to create, use and monetize AI services at scale, it represents an AI marketplace which can help to bring competition to the big AI owners of the world such as Facebook, Google and Microsoft. Businesses and users can easily connect to, and make use of, an AI service that they need, accessing it constantly and only paying for consumption.
· KaaSy: The KaaSy projects endeavour to utilize unused computing power by mining devices which were originally designed for different functions to power a decentralized AI marketplace. Similar to SingularityNET, the idea is to enable a monetization of AI algorithms and further the cause of decentralized AI, also effectively solving the scalability issue by building an incentive scheme with the KAAS token. It is important to note, however, that this company hasn’t posted updates in a few months.
There’s still a long way to go…
There are a few clear conclusions that can be drawn from the above-mentioned use cases and potentials. For one: AI and blockchain do, indeed, have a number of intersections and synergies. Moreover, it seems to be more one-sided, in the sense that blockchain can do more for AI than vice-versa.
Overall, the preliminary state of these findings and predictions must be highlighted. While numerous expert evaluations and academic papers have formed the basis for this article, both industries are still in their infancy; hence, this applies even more toward the potential combined applications that the future may reveal.
Both industries are highly dynamic, and we find ourselves in the privileged, interesting position of having to be patient and see where these two technologies will intersect and merge.
Altogether, a myriad of expert opinions and use cases show that blockchain and AI are an amazing couple with great chemistry, and that better get on the dancefloor together soon and find their rhythm. It is clear, though, that one partner can stand to offer more to the other — blockchain is an essential technology, which could enable AI to become AGI, more transparent, secure and explainable, and thereby realize its full potential in a decentralized, democratic playing field, free of monopolized AI power structures.
Indeed, blockchain is not a nice complement to AI; it should be taken as the revolutionary new data structure that powers AI. If the AIs of the world were put on the blockchain, and the issues of scalability and cross-chain compatibility were worked out, the world could then see the safe rise of explainable AI on the horizon much sooner than expected.
AI, on the other hand, is a great companion and partner for blockchain development and optimization. It can help blockchain technology to become better and more efficient, as well as receive more attention in the public spotlight as one of AI’s biggest enablers.
Of course, there are some major limitations, as of present, with the biggest issues being around what blockchain(s) and what types of blockchain, and what consensus algorithms to best use for AI-hosting blockchain. Overall, it is desirable to see more AI companies experiment with putting their AI on a blockchain. Moreover, it would be fantastic to see more blockchain projects launching which focus on building the blocks and links to enable Blockchain-powered AI.
Dinh, T. N., & Thai, M. T. (2018). Ai and blockchain: A disruptive integration. Computer, 51(9), 48–53.
Goertzel, B. (2007). Artificial general intelligence (Vol. 2). C. Pennachin (Ed.). New York: Springer.