True Utility Function Research Articles

Aversion to external feedback suffices to ensure agent alignment

Ensuring artificial intelligence behaves in such a way that is aligned with human values is commonly referred to as the alignment challenge. Prior work has shown that rational agents, behaving in such a way that maximizes a utility function, will inevitably behave in such a way that is not aligned with human values, especially as their level of intelligence goes up. Prior work has also shown that there is no “one true utility function”; solutions must include a more holistic approach to alignment. This paper describes apprehensive agents: agents that are architected in such a way that their effective utility function is an aggregation of a partial utility function (built by designers, to be maximized) and an expectation of negative feedback on given states (reasoned about, to be minimized). Agents are also capable of performing a temporal reasoning process that approximates designers’ intentions in function of environment evolution (a necessary feature for severe mis-alignment to occur). We show that an apprehensive agent, behaving rationally, leverages this internal approximation of designers’ intentions to predict negative feedback, and, as a consequence, behaves in such a way that maximizes alignment, without actually receiving any external feedback. We evaluate this strategy on simulated environments that expose mis-alignment opportunities: we show that apprehensive agents are indeed better aligned than their base counterparts and, in contrast with extant techniques, chances of alignment actually improve as agent intelligence grows.

Two-Stage Auction Mechanism for Long-Term Participation in Crowdsourcing

Crowdsourcing has become an important tool to collect data for various artificial intelligence applications, and auction can be an effective way to allocate work and determine reward in a crowdsourcing platform. In this article, we focus on the crowdsourcing of small tasks such as image labeling and voice recording, where we face a number of challenges. First, workers have different limits on the amount of work they would be willing to do, and they may also misreport these limits in their bid for the work. Second, if the auction is repeated over time, unsuccessful workers may dropout of the system, reducing competition and diversity. To tackle these issues, we first extend the results of the celebrated Myerson's optimal auction mechanism for a single-parameter bid to the case where the bid consists of the unit cost of work, the maximum amount of work one is willing to do, and the actual work completed. We show that a simple payment mechanism is sufficient to ensure a dominant strategy from the workers, and that this dominant strategy is robust to the true utility function of the workers. Second, we propose a novel, flexible work allocation mechanism, which allows the requester to balance between cost efficiency and equality. While cost minimization is obviously important, encouraging equality in the allocation of work increases the diversity of the workforce as well as promotes a long-term participation on the crowdsourcing platform. Our main results are proved analytically and validated through simulations.

Utility Preference Robust Optimization with Moment-Type Information Structure

Utility Preference Robust Optimization with Moment-Type Information Structure In some decision-making problems, information on the true utility function of the decision maker may be incomplete, which may bring potential modeling risk. In “Utility Preference Robust Optimization with Moment-Type Information Structure,” Guo, Xu, and Zhang propose a maximin utility preference robust optimization model where information about the DM’s preference is constructed by moment-type conditions. The authors propose a piecewise linear approximation approach to tackle the maximin problem, reformulate the approximate problem as a single mixed integer linear program, and derive error bounds for the approximate ambiguity set, the optimal value, and the optimal solutions. To examine the performance of the model and the computational schemes, they carry out extensive numerical tests and demonstrate the effectiveness of the model and efficiency of the computational methods.

In Search of Sustainable Development: A Critical Analysis of the Modern Moscow Experience

The uncontrolled growth of large cities leads to environmental degradation. A decrease in the quality of life and a negative reaction of the population to the changes take place. The externality theory is chosen by the authors as a methodological approach to explore urban conflicts, since it allows identifying and assessing the real economic costs of decisions made by city authorities. Comparison of the provisions of the concept of sustainable urbanism and the actual state of Moscow allowed the authors to draw a conclusion about the anti-sustainable development trend of this city. Sustainable urbanism sees a high quality urban environment characterized by ecological and social health as the main feature of a modern city. Moscow does not comply with these principles, which is proved by numerous conflicts with residents. The tools for urban planning optimization are proposed. The fair concession method and the formation of a true utility function are proposed as objective criteria in the process of urban planning. That allows us to prevent and / or resolve some types of urban conflicts.

Bayesian Automatic Relevance Determination for Utility Function Specification in Discrete Choice Models

Specifying utility functions is a key step towards applying the discrete choice framework for understanding the behaviour processes that govern user choices. However, identifying the utility function specifications that best model and explain the observed choices can be a very challenging and time-consuming task. This paper seeks to help modellers by leveraging the Bayesian framework and the concept of automatic relevance determination (ARD), in order to automatically determine an optimal utility function specification from an exponentially large set of possible specifications in a purely data-driven manner. Based on recent advances in approximate Bayesian inference, a doubly stochastic variational inference is developed, which allows the proposed MNL-ARD model to scale to very large and high-dimensional datasets. Using semi-artificial choice data, the proposed approach is shown to be able to accurately recover the true utility function specifications that govern the observed choices. Moreover, when applied to real choice data, MNL-ARD is able discover high quality specifications that can outperform previous ones from the literature according to multiple criteria, thereby demonstrating its practical applicability.

Nonparametric estimation of utility function in first-price sealed-bid auctions

I propose a simple method to nonparametrically estimate the utility function in first-price, sealed-bid auctions when bidders’ participation is exogenous. Using a pair of bid densities, each for a different number of bidders, I establish a contraction mapping that converges to the true utility function. The method employs nonparametric estimates in place of the true bid densities to construct the contraction mapping and computes the utility function by iterating the contraction operator. A Monte Carlo study shows that the method approximates the true utility function and valuation density.

Multidimensional Auctions and Scoring Rules

An auction is a mechanism to buy (sell) a product or service, where potential sellers (buyers) have to submit bids, which fully determine the winner of the auction and the price he or she has to pay (or receives). The main objective of this study is to determine what the maximum possible total welfare resulting from a multidimensional procurement auction can be in certain settings, to derive equilibria for first-score, second-score and second-preferred-offer auctions and to see under which conditions the maximum possible total welfare is obtained in these equilibria. Some of the results which will be obtained are that there exist ex-post equilibria as well as a dominant strategy equilibrium in second-score auctions, while there exist only ex-post equilibria in first-score and second-preferred-offer auctions and that the maximum possible total welfare is obtained in any ex-post equilibrium of a first-score auction in which the scoring rule equals the buyer's true utility function.

Learning a decision maker's utility function from (possibly) inconsistent behavior

When modeling a decision problem using the influence diagram framework, the quantitative part rests on two principal components: probabilities for representing the decision maker's uncertainty about the domain and utilities for representing preferences. Over the last decade, several methods have been developed for learning the probabilities from a database. However, methods for learning the utilities have only received limited attention in the computer science community. A promising approach for learning a decision maker's utility function is to take outset in the decision maker's observed behavioral patterns, and then find a utility function which (together with a domain model) can explain this behavior. That is, it is assumed that decision maker's preferences are reflected in the behavior. Standard learning algorithms also assume that the decision maker is behavioral consistent, i.e., given a model of the decision problem, there exists a utility function which can account for all the observed behavior. Unfortunately, this assumption is rarely valid in real-world decision problems, and in these situations existing learning methods may only identify a trivial utility function. In this paper we relax this consistency assumption, and propose two algorithms for learning a decision maker's utility function from possibly inconsistent behavior; inconsistent behavior is interpreted as random deviations from an underlying (true) utility function. The main difference between the two algorithms is that the first facilitates a form of batch learning whereas the second focuses on adaptation and is particularly well-suited for scenarios where the DM's preferences change over time. Empirical results demonstrate the tractability of the algorithms, and they also show that the algorithms converge toward the true utility function for even very small sets of observations.

Robust Bayesian decision theory applied to optimal dosage.

We give a model for constructing an utility function u(theta,d) in a dose prescription problem. theta and d denote respectively the patient state of health and the dose. The construction of u is based on the conditional probabilities of several variables. These probabilities are described by logistic models. Obviously, u is only an approximation of the true utility function and that is why we investigate the sensitivity of the final decision with respect to the utility function. We construct a class of utility functions from u and approximate the set of all Bayes actions associated to that class. Then, we measure the sensitivity as the greatest difference between the expected utilities of two Bayes actions. Finally, we apply these results to weighing up a chemotherapy treatment of lung cancer. This application emphasizes the importance of measuring robustness through the utility of decisions rather than the decisions themselves.

The statistical price index as an approximation of the constant-utility price index: an empirical analysis using Japanese data sets

It is well known that there is a gap between the constant-utility price index based on consumer demand theory and the official Consumer Price Index (CPI) published by the Statistics Bureau. In this paper, as a benchmark to determine the constant-utility price index, the Almost Ideal Demand (AID) System utility function is assumed to be a true utility function. Our findings suggest that the constant-utility, the Laspeyres and the Törnqvist price indices calculated from the SNA data sets show similar movement, while the official CPI differs. In order to improve the accuracy of the official CPI, it should include agricultural and single households in the survey design. J. Japanese Int. Economies 19 (1) (2005) 37–50.

A model of human nature and personal development

This paper assumes that human nature is universal and postulates the existence of a common, “true utility function” defined on attributes or “substantive goods.” However, individuals have different household production functions so that the capacity to realize the substantive goods varies from person to person. Moreover, the limited life experience of an individual constrains his perception of his constraints and oppurtunities. As a result, individual maximization behavior is based on a perceived utility function and a perceived household production function. Personal development is a process of learning about the true utility or “happiness” function and the true household production function. If a virtuous circle of personal development is activated, imaginary constraints are overcome and real constraints are accepted. On the other hand, a vicious circle of immiserising development may occur if frustrations are misinterpreted.

Manipulation via Endowments

A competitive allocation in an exchange economy has several desirable properties; it is (with minimal assumptions) Pareto optimal and it guarantees that each trader doesn't lose utility in the trade. Besides these properties of efficiency and individual rationality, however, there are undesirable properties such as its manipulability. There is the possibility that an agent can behave as though his utility function is other than the true utility function and achieve a better outcome (higher utility for him) than at a true competitive outcome. Hurwicz (1972) has shown that this manipulability isn't unique to the competitive process, but in fact is common to any reallocation scheme which achieves Pareto optimal, individually rational outcomes. Besides manipulation via the agent's utility function, there is another type of manipulation, namely of the agent's endowment. In theory, an agent's endowment can be known to all people through inspection and therefore undetectable manipulation is impossible. In practice, however, it is often impossible to determine an agent's true endowment. Even if it were possible to determine true endowments, given a private ownership structure, there is no way of preventing an agent from destroying all or some portion of his endowment. In economies in which there are many agents, each of whom initially owns (at most) a small share of the total amount of a commodity the advantages of these types of manipulation would be small. It is straightforward to apply the techniques used by Roberts and Postlewaite (1976) to show (with similar assumptions as used there) that as an economy gets large agents have diminishing incentive to manipulate endowments. However, in economies in which there are small numbers of agents, large agents, or coalitions of agents the manipulation of the competitive mechanism is theoretically possible. Situations in which farmers withhold from market or destroy portions of their crops are possible instances of coalitional attempts to manipulate via endowments. In international trade we normally must consider agents who are not insignificant and potentially can upset competitive allocations through manipulation. Because of this possible manipulation of the competitive mechanism, it becomes interesting to investigate the possibility of designing alternative mechanisms which are not manipulable. In this paper we will show that the problem of manipulation by withholding endowments is not unique to the competitive mechanism, but rather is shared by any Pareto optimal, individually rational scheme of reallocating resources. We show that manipulation via destruction of endowments, however, can be avoided and give an example of a resource reallocation scheme which cannot be manipulated in this way. Finally, we consider manipulation by coalitions and show that if coalitions can pre-trade , that is, form a submarket among themselves, they may be able to enter the larger market in an improved position. We show that all Pareto optimal individually rational reallocation schemes are susceptible to this form of manipulation and discuss stability problems which therefore arise. 2. MODEL

On the flexibility of flexible functional forms: An empirical approach

This paper analyzes empirically the ability of the Translog and Generalized Leontief functional forms to approximate Constant Elasticity of Substitution utility functions. Although these flexible forms provide a good local approximation they do not always provide a good approximation over a range of observations. Hence the finding in practice of observations for which the regularity conditions required by economic theory are not satisfied need not imply the absence of an underlying utility-maximizing process, but may simply reflect the inability of the flexible form to approximate the true utility function over the range of the data.