In Dr. Andrew Ure’s Philosophy of Manufacturing (1835) he defines the Factory as: ‘the idea of vast automation…being subordinated to a self-regulating moving force, with the main difficulty being in the ‘distribution of the different members into one cooperative body’². Ethereum, one of the most actively used blockchains has pioneered the use of decentralised applications (Dapps) and decentralised autonomous organisations (DAOs) which run on peer-to-peer networks and replaces the need for a centralised computing. The term DAO first appeared to describe a system and infrastructure that is conceived as a self-governing system on peer-to-peer network. Evidently, the concept of self-regulation is important in distributed systems and in evolutionary computation, using metaheuristic procedures to optimise the structure and evolution of the system. These could be argued as form of niche construction on both a digital and physical level — the coordination of developer effort at a physical level, and the emergence of a self-regulating ‘cybernetic’ system architecture.

The existing Ethereum blockchain is run on tools like Parity³ and Geth⁴which help optimise scaling problems across the system, and continue the flow of distribution. The emergence of ‘parachaining’⁵ and new ‘sharding’⁶techniques has bolstered the speed of decentralised Web 3.0 protocols by allowing smoother integrations and transactions between blockchains. These have provided new sandboxes for evaluating the flux of new computational tools that are accessible to remodel existing structures of data and cooperation.

Jack Dorsey has expressed his ambition of running Twitter on a blockchain; speaking at the Oslo Freedom Forum 2020 he said “Blockchain and Bitcoin point to a future, point to a world, where content exists forever.”⁷ Blockchain is still an emergent technology but the adoption of the technology into enterprise use cases has shown how versatile the technology is and how dynamic self-organising systems can be. But it also highlights the increasing need for a ‘trusted’ system architecture for AI and the difficulties that emerge in distrusted environments.

The emergence of new clusters of highly-skilled blockchain engineers and developers are evolving blockchain and distributed protocols at increasing speed without the organisation of a central system. Decentralised platforms like Ethereum and Polkadot exist to organise the network and bring forth new tools for developers, but this type of environment provides the appropriate structure to experiment in AI mutuality and broaden the applications of AI.

Arguably, one could suggest that the proliferation of distributed ledger tools and affordable blockchain services could spark a new type of swarm intelligence and adaptation, shaping new types of multi-agent systems in new configurations overnight and deploying them across the whole network. There are a significant proportion of connections across blockchain and AI which are largely undiscussed and unmentioned. I’ve broken these down into two sections: (A) the emerging technical architecture to support increasing mining, blockchain development and deep learning, and (B) the technical compatibility of these fields:

(A) Technical Hardware & Infrastructure for Neural Nets and Mining

Technical Infrastructure + Compatibility: Blockchain technology has developed a new era of global hardware infrastructure and high frequency computational tools. Cryptocurrency mining has layered a new technical infrastructure around the world, focused in cluster formation of GPUs, application-specific integrated circuits (ASIC), Tensor Processing Units (TPU), CPUs that could be used to power complex deep learning algorithms. The Sophon unit⁸ owned by Chinese technology company Bitmain includes bespoke silicon chips for AI as well as for crypto mining and blockchain. The development of ASICs clouds, suitable for both Bitcoin and Litecoin mining can also be applied to neural networks.

Hypothetically, cryptocurrency mining pools and miner ecosystems could be converted into new shells for running large AI models across the network. Imagine the whole of the bitcoin mining pool with more electricity annually than the whole of Argentina running a single AI model or Neural Net? Mining pools have become a new type of computational swarm intelligence, directed by one goal of mining currencies collectively as a pool. By pooling computational resources together, swarms of users agglomerate processing power to collect daily rewards for their contributed hashing power. For example, the mining pool Ethermine has 174,530 active miners across the network which all providing aggregated hashpower to work together to solve Ethereum blocks currently active. This is just one example of a mining pool, but what happens if these resources move across to contribute to building AI systems? The profitability of cryptocurrency mining has been fluctuating wildly over the last months*.

If the cryptocurrency revolution collapses, it will leave the door open to spare GPUs, ASICS, and computational pools to be redirected to deep learning systems being trained at speed across vast networks and pools. We have unconsciously created a mining market that can evolve into AI.

China is already deploying their currency processors to deep learning⁹. Regardless of the use of blockchain in the processes, the hardware infrastructure through mining pools, processing centres and warehouses could facilitate a new affordable kind of AI development at scale. Even if these mines are unsuccessful like the Hyperblock Bitcoin Mining Servers and Datacenter¹⁰ their hardware is available cheaply, like the 13,000 servers that went on sale in 2020, there is an unrealised potential for this architecture to be repositioned into AI and NNs.

Social Structure + Swarm Intelligence: The mining and blockchain community, albeit different are composed of highly skilled engineers and developers, who have all rapidly adapted and taught themselves new computational processes and infrastructure. As mentioned above, the technical infrastructure is in place for increasing AI development, and so are the reserves of competent, socially connected (forums, communities) and highly skilled engineers and developers. It seems likely that the synergies or germination between the field of AI will rapidly evolve with the advent of new tools and protocols gifted by more community focused blockchain eco-systems. The open-sourced distribution of these tools will continue to evolve the space considerably.

(B) Technical Compatibility with Present AI

Centralisation: Blockchain networks are not immutable as they have always been purported to be, but a 51% attack is usually more costly than any potential benefits it can provide. Such decentralisation is thus very pragmatic. While it responds to legal, political, or economic threats, it is also based on more or less precise cost-benefit estimates, and tends to not spend more resources on decentralisation. Is absolute transactional transparency and data chains an effective way of compiling information? If companies like OpenAI upload their training data and models to Github, what’s the difference? How can we think about larger participatory models for deep learning, and establish more reliable and non-biased data sets?

Computational Intensity + Scalability: Proof of Stake algorithms are evolving the blockchain model from computationally intensive Proof of Work algorithms using sharding techniques to efficiently distribute the blockchain whilst maintaining the core privacy and security features. Distributed computational networks suffer from the ability to scale due to the increased latency and complexity of the systems. With large cloud providers like AWS, IBM Cloud, Google Cloud all running blockchain and Ethereum blockchains, global scaling into existing cloud platform is resolved and at affordable prices. Hypothetically the reach of blockchain and distributed systems will continue to evolve with the growth of cloud platforms.

Problems: One of the biggest challenges a decentralised system faces is its continuous maintenance and community governance. Every additional measure of decentralisation to minimise external control also diminishes the control powers of the creators of the system. Blockchain and distributed systems raise a fundamental point when thinking about Super-intelligent AI over narrow AI, where it is difficult to distinguish whether this incredible robustness and distribution is a feature or hinderance. It’s most likely that a super-intelligent system using a decentralised architecture would be harder to deal with in the classical ‘pull the plug out’ scenario. Nevertheless, it is important to envisage the paradigms and germination between these two technologies and evaluate models of how breakthroughs in core fields will impact research across a plethora of industries.

Conclusion

Blockchains and peer-to-peer systems have shown great promise in shifting traditional models of data privacy, transparency and self-organising systems. This emerging technology will play some part in the the advancement and proliferation of AI, in what form it is hard to say, but it is apparent that the hardware of the existing infrastructure for mining and blockchains will play an important role in the development of AI in China and the world. New forms of distributed communities have seen the power of decentralised systems and arranged themselves without a centralising authority. In tool based systems like Ethereum, Polkadot, Hyperledger and Solana blockchains, it is really the potential of human interaction and developers to experiment in building new systems for AI — this will certainly be an important consequence of blockchain. Dr. Andrew Ure’s definition of the factory also resembles that of a hypothetical AGI system, which automates and self-regulates its parameters through one cooperative entity or system.

The rise of community tools like Discord and Reddit have created new hidden structures of creativity and collaboration within these spaces. It is important not to under-estimate how key these tools are in accelerating research, ideas and development in the field of AI and Blockchain.

However, there are significant warnings that the emergence of new forms of swarm intelligences and ownerless formations will make a sandbox for AI research and development if we are not cautious. Blockchain systems are still in their infancy, Ethereum is only 6 years old, but they are showing enormous potential to reshape our economies and how we distribute data across the world. But at what cost? For now, we have built the skeleton that will continue to germinate, seed and mutate, but at least we will begin to see the shape of process that has previously been hidden.

Written in April 2021 and thus unaware of present market conditions.*

‍