Blockchain Is Central To The Coevolution Of AI and Humans
Fears abound: AI is going to replace humans. But, this is not necessarily the truth. After all, AI could introduce new types of jobs, rather than eliminate humans. The question is, How?
The need for humans in AI architecture is already clear today. People all over the world are employed to label data and train AIs to ensure they can be trusted. Humans clean, annotate, and debias data in order to mitigate AI’s potentially discriminatory or negative impact.
Only a blockchain architecture can achieve these goals in a democratized and transparent manner. Only blockchain consensus methods can be deployed to achieve more trustworthy AI-based datasets.
In a Proof-of-Work framework, a block might only be mined when the performance of such a model exceeds a certain threshold. Certain models encompass deep learning models.
The process for this parallels those carried out by other blockchain ecosystems, though in an AI architecture the miner of a block is a human overseer training the deep neural network to specification.
An incentive model is important because it motivates user participation on a blockchain-based data sharing platform, which solves data-sharing problems through a transparency layer for data transactions. This allows participants to supervise their own records.
A blockchain architecture compensates data owners to quickly cover the cost of data sharing, while balancing incentives for all participants on a given platform.
One approach could be allocating rewards via Shapley values on an incentive layer. A Shapley value represents the average of all marginal contributions to all possible coalitions. Shapley values keep computation time down in a complex system. Shapley values are instrumental to ensuring smaller clients are also incentivized during training.
In AI-training, well-developed blockchain-based systems employ a form of incentivized symbiosis, in which both AI and humans enter into a social contract built on frameworks pioneered in the Web3 space. The structure is encoded into a blockchain technology layer with defined rules and incentives.
Rewards, in the form of utility tokens, are based on performance. They incentivize AI agents to meet specific goals, including data accuracy, as well as meet efficiency standards. Humans contributing high-quality data, properly training AI systems or providing feedback can also earn tokens in a well-architected system.
How to best fuse blockchain and AI for the purpose of AI training remains hotly debated. One model theorizes governing with a Proof of Personhood (PoP) consensus mechanism that only allows human agents to oversee validation processes.
PoP consensus can resist malicious attacks on peer-to-peer networks, including attacks using fake identities, such as in the Sybil attack, in which a single attacker uses several fake identities to compromise a system.
PoP is enforced at the protocol level, and not via application-level contracts like in other implementations. Incentives through financial rewards help encourage participation in AI-training systems. Increased participation contributes to a network’s resilience.
The intersection of AI, Web3 and humans represent a new field of research that is instrumental to ensuring human-AI coevolution–a feedback loop between humans and AI. In this coevolution, AI influences individual and collective behavior. Likewise, humans also direct the evolution of AI systems.
Blockchain, with its transparent incentives and consensus mechanisms, is central in this to ensuring a positive social impact of AI architectures through the democratization of AI training by incentivizing contributors to accurately label data, debias algorithms, and validate models.
