Recent work shows that we can use partial verification instead of money to implement truthful mechanisms. In this paper we develop tools to answer the following question. Given an allocation rule that can be made truthful with payments, what is the minimal verification needed to make it truthful without them? Our techniques leverage the geometric relationship between the type space and the set of possible allocations.

Many real-world systems such as taxi systems, traffic networks and smart grids involve self-interested actors that perform individual tasks in a shared environment. However, in such systems, the self-interested behaviour of agents produces welfare inefficient and globally suboptimal outcomes that are detrimental to all - common examples are congestion in traffic networks, demand spikes for resources in electricity grids and over-extraction of environmental resources such as fisheries. We propose an incentive-design method which modifies agents' rewards in non-cooperative multi-agent systems that results in independent, self-interested agents choosing actions that produce optimal system outcomes in strategic settings. Our framework combines multi-agent reinforcement learning to simulate (real-world) agent behaviour and black-box optimisation to determine the optimal modifications to the agents' rewards or incentives given some fixed budget that results in optimal system performance. By modifying the reward functions and generating agents' equilibrium responses in a sequence of offline Markov games, our method enables optimal incentive structures to be determined offline through iterative updates of the reward functions of a simulated game. Our theoretical results show that our method converges to reward modifications that induce system optimality. We demonstrate the applications of our framework by tackling a challenging problem in economics that involves thousands of selfish agents and tackle a traffic congestion problem.

Purpose - The purpose of this paper is to report on empirical work conducted to open up algorithmic interpretability and transparency. In recent years, significant concerns have arisen regarding the increasing pervasiveness of algorithms and the impact of automated decision-making in our lives. Particularly problematic is the lack of transparency surrounding the development of these algorithmic systems and their use. It is often suggested that to make algorithms more fair, they should be made more transparent, but exactly how this can be achieved remains unclear. Design/methodology/approach - An empirical study was conducted to begin unpacking issues around algorithmic interpretability and transparency. The study involved discussion-based experiments centred around a limited resource allocation scenario which required participants to select their most and least preferred algorithms in a particular context. In addition to collecting quantitative data about preferences, qualitative data captured participants' expressed reasoning behind their selections. Findings - Even when provided with the same information about the scenario, participants made different algorithm preference selections and rationalised their selections differently. The study results revealed diversity in participant responses but consistency in the emphasis they placed on normative concerns and the importance of context when accounting for their selections. The issues raised by participants as important to their selections resonate closely with values that have come to the fore in current debates over algorithm prevalence. Originality/value - This work developed a novel empirical approach that demonstrates the value in pursuing algorithmic interpretability and transparency while also highlighting the complexities surrounding their accomplishment.

Web-based collaborative systems, where most computation is performed by human collectives, have distinctly different requirements from traditional workflow orchestration systems, as humans have to be mobilised to perform computations and the system has to adapt to their collective behaviour at runtime. In this paper, we present a social orchestration system called SmartOrch, which has been designed specifically for collective adaptive systems in which human participation is at the core of the overall distributed computation. SmartOrch provides a flexible and customisable workflow composition framework that has multi-level optimisation capabilities. These features allow us to manage the uncertainty that collective adaptive systems need to deal with in a principled way. We demonstrate the benefits of SmartOrch with simulation experiments in a ridesharing domain. Our experiments show that SmartOrch is able to respond flexibly to variation in collective human behaviour, and to adapt to observed behaviour at different levels. This is accomplished by learning how to propose and route human-based tasks, how to allocate computational resources when managing these tasks, and how to adapt the overall interaction model of the platform based on past performance. By proposing novel, solid engineering principles for these kinds of systems, SmartOrch addresses shortcomings of previous work that mostly focused on application-specific, non-adaptive solutions.

Web-based collaborative systems, where most computation is performed by human collectives, have distinctly different requirements from traditional workflow orchestration systems, as humans have to be mobilised to perform computations and the system has to adapt to their collective behaviour at runtime. In this paper, we present a social orchestration system called SmartOrch, which has been designed specifically for collective adaptive systems in which human participation is at the core of the overall distributed computation. SmartOrch provides a flexible and customisable workflow composition framework that has multi-level optimisation capabilities. These features allow us to manage the uncertainty that collective adaptive systems need to deal with in a principled way.

This paper explores the challenges around fair information access when the limits of human attention require algorithmic assistance for 'finding the diamond in the coal mountain'. While often demanded by users, the seemingly intuitive concept of fairness has proven to be very difficult to operationalise for implementation in algorithms. Here we present two pilot studies aimed at getting a better understanding of the conceptualisation of algorithmic fairness by users. The first was a multi-stakeholder focus-group discussion, the second a user experiment/questionnaire. Based on our data we arrive at a picture of fairness that is highly dependent on context and informedness of users, and possibly inherently misleading due to the implied projecting of human intentions onto an algorithmic process.

The Generalized Second Price (GSP) auction has appealing properties when ads are simple (text based and identical in size), but does not generalize to richer ad settings, whereas truthful mechanisms such as VCG do. However, a straight switch from GSP to VCG incurs significant revenue loss for the search engine. We introduce a transitional mechanism which encourages advertisers to update their bids to their valuations, while mitigating revenue loss. In this setting, it is easier to propose first a payment function rather than an allocation function, so we give a general framework which guarantees incentive compatibility by requiring that the payment functions satisfy two specific properties. Finally, we analyze the revenue impacts of our mechanism on a sample of Bing data.

The Generalized Second Price (GSP) auction has appealing properties when ads are simple (text based and identical in size), but does not generalize to richer ad settings, whereas truthful mechanisms such as VCG do. However, a straight switch from GSP to VCG incurs significant revenue loss for the search engine. We introduce a transitional mechanism which encourages advertisers to update their bids to their valuations, while mitigating revenue loss. In this setting, it is easier to propose first a payment function rather than an allocation function, so we give a general framework which guarantees incentive compatibility by requiring that the payment functions satisfy two specific properties. Finally, we analyze the revenue impacts of our mechanism on a sample of Bing data.

Vehicle-to-grid (V2G) is a promising approach whereby electric vehicles (EVs) are used to store excess electricity supply (e.g., from renewable sources), which is sold back to the grid in times of scarcity. In this paper we consider the setting of a smart car park, where EVs come and go, and can be used for V2G while parked. We develop novel allocation and payment mechanisms which truthfully elicit the EV owners' preferences and constraints, including arrival, departure, required charge, as well as the costs of discharging due to loss of efficiency of the battery. The car park will schedule the charging and discharging of each EV, ensuring the constraints of the EVs are met, and taking into consideration predictions about future electricity prices. Optimally solving the global problem is intractable, and we present three novel heuristic online scheduling algorithms. We show that, under certain conditions, two of these satisfy monotonicity and are therefore truthful. We furthermore evaluate the algorithms using simulations, and we show that some of our algorithms benefit significantly from V2G, achieving positive benefit for the car park even when agents do not pay for using it.

. Sharing economy applications need to coordinate humans, each of whom may have different preferences over the provided service. Traditional approaches model this as a resource allocation problem and solve it by identifying matches between users and resources. These require knowledge of user preferences and, crucially, assume that they act deterministically or, equivalently, that each of them is expected to accept the proposed match. This assumption is unrealistic for applications like ridesharing and house sharing (like airbnb), where user coordination requires handling of the diversity and uncertainty in human behaviour. We address this shortcoming by proposing a diversity-aware recommender system that leaves the decision-power to users but still assists them in coordinating their activities. We achieve this through taxation, which indirectly modifies users’ preferences over options by imposing a penalty on them. This is applied on options that, if selected, are expected to lead to less favourable outcomes, from the perspective of the collective. The framework we used to identify the options to recommend is composed by three optimisation steps, each of which has a mixed integer linear program at its core. Using a combination of these three programs, we are also able to compute solutions that permit a good trade-off between satisfying the global goals of the collective and the individual users’ interests. We demonstrate the effectiveness of our approach with two experiments in a simulated ridesharing scenario, showing: (a) significantly better coordination results with the approach we propose, than with a set of recommendations in which taxation is not applied and each solution maximises the goal of the collective, (b) that we can propose a recommendation set to users instead of imposing them a single allocation at no loss to the collective, and (c) that our system allows for an adaptive trade-off between conflicting criteria.

We propose a hybrid auction mechanism for sponsored search, where bidders can be truthful or not, and are accordingly treated differently. Our class of hybrid mechanisms give incentives for non-truthful bidders to bid truthfully, while behaving as a non-truthful auction if no bidders are truthful. Our motivation is that the Generalized Second Price (GSP) auction (the current mechanism of choice) has appealing properties when ads are simple (text based and identical in size). But GSP does not generalize to richer ad settings, whereas truthful mechanisms, such as VCG do. Hence there are incentives for search platforms to migrate to truthful mechanisms, but a straight switch from GSP to VCG either requires all bidders instantly bid truthfully or incurs significant revenue loss. We introduce a transitional mechanism which encourages advertisers to update their bids to their valuations, while mitigating revenue loss. The mechanism is equivalent to GSP when nobody has updated their bid, is equivalent to VCG when everybody has updated, and it has the same allocation and payments of the original GSP if bids were in the minimum symmetric Nash equilibrium. In settings where both GSP ads and truthful (TF) ads exist, it is easier to propose a payment function than an allocation function. We give a general framework for these settings to characterize payment functions which guarantee incentive compatibility of truthful ads, by requiring that the payment functions satisfy two properties. Finally, we compare the revenue of our transitional mechanism with revenues of GSP and VCG mechanisms when run on a sample of Bing data.

Current storage offerings provide a small number of options in the form of fixed prices with volume discounting. This leaves storage operators to guess how much data customers will add over time. Instead, we propose that the operator elicits basic information about future usage. Such information can be used to operate the system more efficiently. In particular, we show how prices can be calculated that encourage customers to accurately report the range of their future usage while ensuring that the operator covers his costs.

We demonstrate how crowdsourcing can be used to automatically build a personalized tourist attraction recommender system, which tailors recommendations to specific individuals, so different people who use the system each get their own list of recommendations, appropriate to their own traits. Recommender systems crucially depend on the availability of reliable and large scale data that allows predicting how a new individual is likely to rate items from the catalog of possible items to recommend. We show how to automate the process of generating this data using crowdsourcing, so that such a system can be built even when such a dataset is not initially available. We first find possible tourist attractions to recommend by scraping such information from Wikipedia. Next, we use crowdsourced workers to filter the data, then provide their opinions regarding these items. Finally, we use machine learning methods to predict how new individuals are likely to rate each attraction, and recommend the items with the highest predicted ratings.

We examine trade-offs among stakeholders in ad auctions. Our metrics are the revenue for the utility of the auctioneer, the number of clicks for the utility of the users and the welfare for the utility of the advertisers. We show how to optimize linear combinations of the stakeholder utilities, showing that these can be tackled through a GSP auction with a per-click reserve price. We then examine constrained optimization of stakeholder utilities. We use simulations and analysis of real-world sponsored search auction data to demonstrate the feasible trade-offs, examining the effect of changing the allowed number of ads on the utilities of the stakeholders. We investigate both short term effects, when the players do not have the time to modify their behavior, and long term equilibrium conditions. Finally, we examine a combinatorially richer constrained optimization problem, where there are several possible allowed configurations (templates) of ad formats. This model captures richer ad formats, which allow using the available screen real estate in various ways. We show that two natural generalizations of the GSP auction rules to this domain are poorly behaved, resulting in not having a symmetric Nash equilibrium or having one with poor welfare. We also provide positive results for restricted cases.