Blockchain, multi-agents and dynamic learning in supply chains

Most of us in the supply chain professional community would not probably escape a day without noticing a story of how the blockchains are the talk of the 'supply chain town', with their ambitious claim to become the most promising step towards attaining the unvanquishable double T's of an agile supply chain network vis a vis traceability and transparency.There are two other wizkids of AI that promise to craft a more dynamic element to the plotline-: intelligent multi-agents and dynamic learning . Here, I try to sketch a thread on possible interlude among these 3 game-changers in a relatively less explored context.

Recently I was going through a research paper which was addressing the problem of supply chain disruption management with multi-agent based framework.Given that the treatment of problem was indeed , I found that the approach was silo-type with agents coordinating across enterprise legacy systems .The entire risk identification approach was centered on KPI benchmarking and case based learning approach to response initiation. However, every subject matter expert on risk management would confirm that the topology of risk propagation (especially supply chain risks) as well as their interfaces of influence are anything but simple-linear and predictable.

I have two critiques to the silo-type enterprise solutions approach to supply chain networks

1.There is problem of consistency and interoperability during communication of inter-organisational multi-source heterogenous data. A proposed solution to this challenge is an ontology based middle layer can act as a globally distributed and decentralized network .On the application layer, decentralized apps can access the blockchain layer for retrieving data at the front-end.

2.The learning process should be dynamic and continuous-improvement based, meaning the simulation of policies on any given problem scenario should not be totally dependent on actor-to-actor KPI, rather the response to interplay between actors and environment as a dynamic whole.For instance the response to a demand spike could be wrongly triggered due to information mismanagement (example: manipulated inventory data,information asymmetry) by a given actor(enterprise node in the chain) and since there is no consensus mechanism on an ecosystem level standard of data, this can affect a chain of erroneous responses by the agents.

Dynamic and context adaptive learning: Given the above-mentioned two scenarios I propose a block-chain enabled (digitized) SC which is mediated through  Multi-agent reinforcement learning (MARL) mechanism as the foundational decision support system framework (DSSF).The global distributed network of multi-agents and need for continuous improvement based learning promotes A3C-RL (advantage actor-critic agent) as a possible candidate for the purpose of simulation and optimization.Also researchers claim that A3C has a faster convergence rate and stability among the various RL algorithms.The ontology framework provide a unified schema as for mapping from the blockchain data layer to destination data layer via the API of individual supply chain agents.Hence the integrated metadata of every executed block can be retrieved and transformed to specific needs of the agents with better access to lineage of the data. A formidable task would be to structure 'information governance' so that data access-control roles are properly delineated prior to adoption among the SC parties in accordance with standard security protocols

Blockchain enabled SC coordination: In adherence to the reference architecture advocated by IBM BCNA (Blockchain network architecture), the data layer of the block chain stores all the on-chain and off-chain data.The integration tier ( of BCNA) is responsible for mapping the organizational identity to Blockchain identity is required to access/invoke the chain code provided by the Blockchain solution. The integration tier needs to map Blockchain identities to appropriate Blockchain solution roles for authorization. I propose that blockchain layer and and its entities act as a coordination mechanism in addition to providing peer-peer transaction transparency. A communication between on-chain data amongst different networks could be enabled with API integration and ontology based interoperability.This area needs more use-case based empirical validation.

Multi-agent (MAg) ecosystem: The promising ingenuity of MAg's as coordinators of decision-making ( agent based modelling) in distributed and decentralized complex adaptive systems has been studied and reiterated in literature for the last two decades.The agents share and access data via the certification authority that oversees the quality and access rights of participants.Let's take the case of a global supply chain network(GSC).Each node of the GSC is represented as an agent with autonomous behavior and a local utility function.The learning process consists of adapting the local behavior of each SC node (agent) with the aim of optimizing an intended global behavior ( for example total cost of operation of SC network) of the SC. In addition to self-awareness in maximising local rewards (value) ,the agents possess "collective intelligence or diplomatic awareness" that cause them to collaborate for global optimality.

There needs to be further organizational leadership investment and diligence in developing a reproducible implementation of the 'high-level' view of infrastructure described above. Given the tremendous evolution of business landscape with the advent of great marriage between the fields of SCM and AI, we could at least claim that there is hope for changes to happen slowly but steadily.