AI and blockchain: What kind of sparks will the collision of these two popular technologies create?

Editor's note: Blockchain and AI are the two hottest technology directions today. In the eyes of ordinary people, these two technologies seem to have little overlap, because blockchain and AI belong to the two extremes of the technology spectrum: one is to cultivate centralized intelligence on a closed data platform, and the other is to promote decentralized applications in an open data environment. However, data strategist, technology investor and AI consultant Francesco Corea believes that the integration of AI and blockchain may have a revolutionary impact on the entire technology paradigm. Let's see how he analyzes it.

It is undeniable that AI and blockchain are two important technologies that promote the pace of innovation and introduce dramatic changes to every industry. Each technology has its own technical complexity and business impact, but if these two powerful forces are combined, they may redesign the entire technological paradigm (and humanity) from scratch.

This article aims to take a peek at the potential of the convergence of AI and blockchain and discuss the standard definition, challenges and benefits of this alliance, as well as some interesting players in the field.

I. Be prepared

I've been talking and writing about AI for a while now, so I won't waste time defining what AI is and what it is not.

However, I haven’t touched on the topic of blockchain and cryptocurrency so far, so I will use this first part to introduce what blockchain is and how it works.

• Blockchain is a secure, distributed, and stable database shared by all parties in a distributed network. This database can record transaction data (either on-chain or off-chain) and is easy to audit.
• A blockchain is a secure, distributed, immutable database shared by all parties in a distributed network.

In short, blockchain is “a technology that allows people who don’t know each other to trust a shared record of events.”

Data is stored in rigid structures called blocks, which are linked to each other in a chain through hash values ​​(each block contains a timestamp and a link to the previous block through its hash value). Blocks have a header (which contains metadata) and a body containing the actual transaction data. Since each block is linked to the previous one, it is extremely difficult to tamper with any information without the consensus of the network as the number of participants and blocks continues to grow.

The network can verify transactions through different mechanisms, but there are two main mechanisms, one is "proof-of-work" and the other is "proof-of-stake". Proof-of-work (Satoshi, 2008) requires participants (called "miners") to solve complex mathematical problems to add blocks, which in turn require a lot of electricity and hardware capabilities to decode. Proof-of-stake (Vasin, 2014), on the contrary, tries to solve this energy efficiency problem and give more mining power to participants with more currency (proof-of-stake has many derivatives, and its famous "ledger fork problem" has also raised some doubts).

Other mechanisms include Byzantine Fault Tolerance (Castro and Liskov, 2002), Quorum Slicing (Mazieres, 2016), and various derivatives of Proof of Stake, but we will not discuss that for now.

The last feature that needs to be explained is the classification of blockchains according to different network access rights, such as whether anyone can browse freely (permissionless vs permissioned), or participate in the formation of consensus (public vs private). In the former case, anyone can access the ledger and read and write to it, while in the latter case, pre-determined participants have the right to join the network (of course, in the case of public permissionless, as a reward structure provided to miners).

It should be clear by now that the essential power of this technology is not just that it is disruptive, but that it is a foundational technology that aims to “change the intermediation landscape.” Distributed ledger technology does reduce the costs of verification and networking, which in turn affects market structure and ultimately enables the formation of new markets. In recent work, Iansiti and Lakhani (2017) also provide a fascinating comparison between blockchain and TCP/IP, showing how blockchain has slowly gone through the same four stages that previous foundational technologies such as TCP/IP have gone through: single use, localized use, substitution, and transformation. As they explain, the “newness” of such a technology makes it difficult to understand the solution domain, while its “complexity” requires a larger institutional shift to foster a more conducive adoption climate.

However, one thing is also true, that is, blockchain is changing traditional business models and distributing value in a direction that is opposite to the previous technology stack: if investing in applications made more sense than investing in protocol technology 15 years ago, in the blockchain world, value will be concentrated in the shared protocol layer, and the profit level at the application layer will be very meager (see Joel Monegro's "fat protocol" theory).

This is a stack consisting of "fat" protocols and "thin" applications

——Joel Monegro

Finally, to end this introductory chapter, I would also like to mention that the possibilities of blockchain are not limited to transactions, but also to the creation of (smart) contracts that are triggered by specific events and thresholds and can be easily traced and audited.

Additional Information: Initial Coin Offering (ICO)

A big part of the hype surrounding this new phenomenon is the Initial Coin Offering (ICO). Even though many people are throwing money at it because the name evokes the most common (and most valuable) Initial Public Offering (IPO), an ICO is nothing more than a sale of tokens, which are the smallest functional unit of a particular network.

Hopefully ICO experts will forgive my crude definition, but an ICO is a hybrid concept that combines elements of equity distribution, pre-sale/crowdfunding campaigns, and currencies with limited powers and application domains.

The introduction of a new unregulated means of raising funds is definitely an interesting innovation, but it also raises several concerns in a community that is not ready yet. I would love to hear your feedback, but here are the key points to consider when evaluating ICOs:

• Tokens have additional utility in value exchange, and a company selling tokens with the sole goal of raising money would send a bad signal to the market. Tokens are used to build a user base and incentivize stakeholders to participate in the ecosystem at the earliest stage. A good white paper alone is not enough;
• Beware of unrestricted token sales;
• Be wary of token sales with no time limit;
• Be wary of token sales that do not clearly state the (current and future) number and value of tokens (this may sound a bit ridiculous, but you would be surprised at the opacity of ICOs).

II. How AI will change blockchain

While blockchain is extremely powerful, it has its own limitations. Some of these are technology-related, while others stem from the old-school culture inherent in the financial services sector, but all of them will be impacted by AI in some way:

1. Power consumption: Mining is an extremely difficult task that requires a lot of electricity (and money) to complete. AI has been proven to be an effective means of optimizing power consumption, so I think similar results can be achieved in blockchain. This may lead to a decrease in investment in mining hardware;
2. Scalability: The blockchain is growing steadily at a rate of 1MB every 10 minutes, and has now accumulated 85GB. Satoshi Nakamoto (2008) first proposed "blockchain pruning" (i.e. deleting unnecessary data about fully consumed transactions) as a possible solution, but AI can introduce new decentralized learning systems such as federated learning, or introduce new data sharding technologies to make the system more efficient.
3. Security: Even though blockchain is almost impossible to attack, the deeper layers and applications of blockchain are not so secure (such as DAO, Mt Gox, Bitfinex, etc.). The incredible progress made in machine learning in the past two years makes AI an excellent ally of blockchain to ensure secure application deployment, especially given the fixed nature of the system architecture;
4. Privacy: Privacy concerns about owning personal data raise regulatory and strategic concerns about competitive advantage. Homomorphic encryption (operating directly on encrypted data), Project Enigma, or Project Zerocash are definitely possible solutions, but I think this issue is closely tied to the previous scalability and security issues, and I think they are equally important;
5. Efficiency: Deloitte (2016) estimates that the total operating cost of blockchain verification and shared transactions is about $600 million per year. An intelligent system may eventually be able to calculate in real time the probability of a particular node being the first to perform a particular task, thereby allowing other miners to choose to abandon their efforts on that particular transaction, thereby reducing the total cost. In addition, even if there are certain structural constraints, better efficiency and lower energy consumption may also reduce network latency, thereby making transactions faster;
6. Hardware: Miners (not necessarily companies but also individuals) invest incredible amounts of money into specialized hardware components. Since power consumption has always been a key issue, many solutions have been proposed and more will be introduced in the future. As long as the system becomes more efficient, some of the hardware may be converted (sometimes partially converted) to use neural networks (mining giant Bitmain is doing this);
7. Lack of talent: This is a leap of faith, but in the same way that we are trying to automate data science itself (unsuccessfully to my knowledge so far), I don’t see why we can’t create virtual agents that can create new ledgers (or even influence and maintain them);
8. Datagate: In the future, when all our data is on the blockchain and companies can buy it directly from us, we will need help granting access, tracking data usage, and generally figuring out what happens to our personal information at computer speeds. That’s the job of (smart) machines.

III. How blockchain changes AI

In the previous section, we quickly touched on the impact that AI might eventually have on blockchain. Now let’s flip the tables and look at how blockchain might impact the development of machine learning systems. In more detail, blockchain can:

1. Help AI explain itself (and make us trust it): AI black boxes suffer from the problem of explainability. Having a clear audit trail not only improves the credibility of the data, but also the credibility of the model, and also provides a clear way to trace the machine decision-making process.
2. Improved effectiveness of AI: Secure data sharing means more data (and more training data), which leads to better models, better actions, better outcomes… and better new data. At the end of the day, network effects are the most important thing.
3. Lowering barriers to entry: Let’s take it one step at a time. Blockchain technology protects your data. So why can’t you store all your data privately, or maybe sell it? You probably would. So first, blockchain will facilitate the creation of cleaner, more organized personal data. Second, blockchain will facilitate the emergence of new markets: data markets (this is easier to implement); model markets (this is much more interesting); and even AI markets eventually (see what Ben Goertzel is trying to solve with SingularityNET). So easy data sharing and new markets, combined with blockchain data verification, will provide smoother integration, thereby lowering the barriers to entry for small businesses and shrinking the competitive advantage of tech giants. In the effort to lower barriers to entry, we are actually solving two problems, namely providing broader access to data and more effective data monetization mechanisms;
4. Increase trust in humans: Once some of our tasks are managed by automated virtual agents, clear audit trails will help robots trust each other (and help us trust them). This will eventually increase machine-to-machine interactions (Outlier Ventures, 2017) and transactions, with itemized data and coordinated decisions, plus a secure means of reaching quorum (highly relevant for swarm robotics and multi-agent scenarios). Rob May also expressed a similar concept in his recent email newsletter.
5. Reduced catastrophic risk scenarios: An AI written in a DAO with a specific smart contract can only perform those actions and nothing more (so its action space is also limited).

Although the interaction between AI and blockchain technology can bring many benefits, there is still a big problem that bothers me.

AI was born in an open source environment where data is the real moat. But as this data becomes democratized (and software becomes open source), how can we ensure AI can succeed and continue to grow? What will be the new moat? My only guess at this stage is... talent.

IV. Decentralized Smart Company

There are many startups working on blockchain and cryptocurrency. However, I am only interested in those working on the intersection (or integration) of AI and blockchain technology, which are obviously not very common. Such companies are mainly concentrated in San Francisco and London, but there are also examples in New York, Australia, China and European countries.

The number of startups in this category is really small, so it’s hard to further categorize them. I usually like to try to understand the underlying model of a group of companies and the impact/type of application they have on the industry, but given the small number of data points, it’s difficult to do such analysis here, so I’ll simply categorize them as follows:

Decentralized intelligence: TraneAI (training AI in a decentralized manner); Neureal (peer-to-peer AI supercomputing); SingularityNET (AI marketplace); Neuromation (comprehensive dataset generation and algorithm training platform); AI Blockchain (multi-application intelligence); BurstIQ (healthcare data marketplace); AtMatrix (decentralized robotics); OpenMinedproject (data marketplace for training machine learning locally);

• Conversational platforms: Green Running (home energy virtual assistant); Talla (chatbot); doc.ai (quantitative bio and healthcare insights);
• Prediction platforms: Augur (collective intelligence); Sharpe Capital (crowdsourced sentiment prediction);
• Intellectual property: Loci.io (IP discovery and mining);
• Data provenance: KapeIQ (fraud detection for healthcare entities); Data Quarka (fact checking); Priops (data compliance); Signzy (KYC)
• Transactions: Euklid (Bitcoin investment); EthVentures (investment in digital tokens). For other (theoretical) financial applications, see Lipton (2017);
• Insurance: Mutual.life (P2P insurance), Inari (general insurance);
• Others: Social Coin (citizen reward system); HealthyTail (pet analysis); Crowdz (e-commerce); DeepSee (media platform); ChainMind (cybersecurity).

Here are some reviews:

It is interesting to see that many AI-blockchain companies have advisory boards that are larger than the team size. This may be an early sign that the convergence is not yet complete, indicating that we don’t know more than we do know;

I’m personally very excited to see the first category of startups grow (decentralized intelligence), but I also see huge growth in conversational and predictive platforms, as well as intellectual property. I categorized the other examples as “miscellaneous” because I don’t think they represent specific categories at this stage, but rather individual attempts to pair AI with blockchain;

Evaluating these companies is extremely difficult. The websites are often cryptic, making it unclear what they do and how they do it (which is a bit contrary to the transparency you buy into blockchain for), and the technology requires a high-tech education to fully evaluate it. It’s a difficult task to cut through the fog of hype, and it’s easy to be fooled. But let me give you a specific example: Ever heard of Magos AI? While researching the company for this article, I read several articles about this AI-driven blockchain prediction platform company (from Wired, Prnewswire, etc.) that had just completed an ICO of over $500,000 and made big promises about its deliverables.

But if you think they should share the ICO materials/information and want to check their website, their website is weirdly not accessible. Of course, sometimes this happens. But I was still not satisfied because I read about it in Wired and I wanted to know more. I managed to find out who its co-founder is, but I can't find his profile on Linkedin. Another weird thing. But some people don't like social activities, especially if you consider that there was no sign of the company three months ago. Let's take a look at the other team members. There is no information either, and I can't find any traceable evidence of their past experience (except that the CTO is an analyst, but I can't find any evidence of this). I tried to dig deeper into their technology: I wanted to find their white paper, blue paper, yellow paper, or whatever book. But I could only find relevant comments, but no text.

Two final points: I don’t consider myself a blockchain expert at all, but I’ve read a lot about it. And I also believe I’m pretty knowledgeable about AI and what’s going on in the industry. These guys claim to have built 5 different neural networks that can achieve the same accuracy in different areas that are more complex than Libratus (or DeepStack), but I’ve never heard of such a network - which is very strange. Well, maybe I can write to them and ask to meet and get to know them. But you know what? Their address is AXA’s Zurich office.

After 5 minutes of research, I finally Googled two keywords: "Magos scam". It seems that these guys ran away with the money. They must have gone somewhere to build that neural network. So please keep an eye on it.

My view is that exponential technology is very good and can advance human development, but as the benefits it brings increase, the potential for "negative integration" will also increase exponentially. Be vigilant.

V. Conclusion

Blockchain and AI are two extremes of the technology spectrum: one is to cultivate centralized intelligence on a closed data platform, and the other is to promote decentralized applications in an open data environment. However, if we can find a smart way to make these two work together, the total positive externalities can be magnified instantly.

The fusion of these two technologies certainly has technical and ethical implications, such as how should we edit (or even forget) data on the blockchain? Is an editable blockchain a solution? Will the fusion of AI-blockchain push us down the path of becoming data hoarders?

Honestly, I think the only thing we can do is keep experimenting.