There is none deceived but he that trusts --Benjamin Franklin

Introduction

Given the current state of knowledge, it is relatively easy to have artificial intelligent agents to find patterns and to formulate predictions following some objective criteria. But to get become useful in comparison with human intelligence, it is crucial that those agents are able to ask: Why? Posing the question is an exercise on causal reasoning, a realization of the awareness of cause and effect.  It is also a matter beyond logic formalisms--  the interface matters. That is because once the intelligent agent is able to ask why?, it will be also in position to ask counterfactual questions, such as how an intervention may chance the output, or even the causal relationship itself \cite{Kyburg_1991}. This is particularly important in the case of blockchain-based AIs because directionality matters (causation works in one way) and because the substrate of the AI is a fully or partially immutable ledger. 
One may attempt to reduce the proposed solution to the problem to the application of well known methods, such as bayesian networks; since blockchain systems are deterministic and it is desirable that the constructs operating on top of them (e.g. smart contracts) behave with some degree of predictability, establishing an appropriate intuition of the priors by access to on-chain data and some sort of data integrity-verified artifacts (e.g oracles) may at least partially achieve this purpose. However, such a simplistic approach would deprive the AI of context: the world is not presented to us as a data feed, but rather as a dynamical experience in which the embeddedness into a social context \cite{W_lfer_2012} dictates the response, specially in the forming stages of intelligence. Particularly, the repeated use of metaphors such "circles of trust" in industry and personal relations hints at the tendency that humans have to at least implicitly compute similarity metrics (to define the boundaries of that circle, or space) and to elaborate mechanisms to detect violations to certain social laws and descriptive models of economic behavior \cite{Liu_2014}. In a way, to engage in the world we require that other agents are "trust verified".
Trust is fundamental to the human experience, yet it is little understood. But AI, web analytics, and blockchain technology have come to change that. For the first time in history, we have real-time data to map how attention flows, and understand how people actually assess risk and commit resources. And with AI we can augment our own intellect \cite{Engelbart_1995}, to understand complex socio-technical systems. In some cases this is a full departure from the prevailing economic theories, that were developed using experiments conducted on small groups, incomplete and delayed macroeconomic data, or theoretical models which are completely dissociated from reality. There is no reason why economics and the social sciences have to be called "soft" anymore; there is no such thing as hard and soft sciences, a scientist should always operate in the realm of facts and quantitative evidence, otherwise, he is only a commentator.
Computationally, biologically, and socially, humans "need" to trust \cite{Fareri2015}. And even when dealing with "trust-less" systems such as distributed ledger technology, everyone (including AIs) need to trust "on the design". This work is concerned with the role of trust asymmetry \cite{Venegas} in the causal reasoning process of blockchain-based AIs. We will approach it from the perspective of the machine, of the method and apparatus needed for the intelligent agent to make consequential decisions in the world: is there an asymmetry, then why? And, where and when trust asymmetry breaks?

Related work

 

Integrating social information with traditional network layers

A blockchain-based AI becomes aware of the environment via off-chain data. To conceptualize this using the OSI model as a reference, we use a cross-layer stack (layer 7 and layer 6).  In practice, the financial activity logged in the blockchain is the expression (e.g. a conversion event) of the consumption, flow of attention, and commitment of resources in adjacent networks such as the web and mesh IoT. For instance,  when a web browser creates a request (e.g. GET / HTTP) a Java based application could log the hits, detect the device type (mobile or desktop), and other features included in the user-agent header. Also by looking down in the stack to the routing level, ISP data is used to obtain geographical origin, redirect path, and destination. This sort of "alternative data" becomes specially useful when studying permissioned and semi-centralized networks, since not all data is publicly available and suitable proxies are required.    

Ripple

Formally, Ripple is not a blockchain, but a common ledger based on a proprietary technology to cater to the privacy needs of the banking industry --therefore, some transactional and network activity details are unavailable to the public.  We tracked daily usage for the 100 most popular services among prospect Ripple users over the course of 18 months (548 days in total). We use daily prices as target variable since a general audience-prospect user will be inclined to look at the daily prices, while professional investors usually focus on daily returns. The services included those directly related to cross-border payments operations (e.g. gateways) and other peripheral to the economy, including price information services, wallets, and the like. We investigate the long-term market structure, specifically the demand signals from that segment of new comers. We started with one hundred services, and after many rounds of elimination making different formulas compete with each other for accuracy (using symbolic regression), the simplest and most meaningful relationship expresses price as a function of two constant values and the demand for the services of a particular exchange In other words, the simpler predictive model to provide any insight traces back the rise of Ripple among this segment to the popularity of one of the prominent exchanges (of the centralized type) that listed the coin. One such predictive model can be expressed as  Price = a*exchange1 — b  
A more complex model has the form Price = a + b*sma(wallet_users, 21)*sma(wallet_users, 37) + c*wallet_public² — d*exchange — e*gateway*sma(wallet_users, 21)*sma(wallet_users, 37)
According to this, from May 2017 to December the use of a particular wallet created support levels of 21 and 37 days (using simple moving averages) and it had the bigger impact of all variables discovered (increasing the use of this wallet has a positive impact on price 100% of the times). This means that the usage of this service serves like a “canary in the mine”, i.e a prolonged decreased in usage (being all conditions equal, such as not having a similar alternative replacing the use of the wallet) will indicate weakening demand fundamentals.