Human Preferences Research Articles

Visual Instruction Tuning with Polite Flamingo

Recent research has demonstrated that the multi-task fine-tuning of multi-modal Large Language Models (LLMs) using an assortment of annotated downstream vision-language datasets significantly enhances their performance. Yet, during this process, a side effect, which we termed as the "multi-modal alignment tax", surfaces. This side effect negatively impacts the model's ability to format responses appropriately - for instance, its "politeness" - due to the overly succinct and unformatted nature of raw annotations, resulting in reduced human preference. In this paper, we introduce Polite Flamingo, a multi-modal response rewriter that transforms raw annotations into a more appealing, "polite" format. Polite Flamingo is trained to reconstruct high-quality responses from their automatically distorted counterparts and is subsequently applied to a vast array of vision-language datasets for response rewriting. After rigorous filtering, we generate the PF-1M dataset and further validate its value by fine-tuning a multi-modal LLM with it. Combined with novel methodologies including U-shaped multi-stage tuning and multi-turn augmentation, the resulting model, Clever Flamingo, demonstrates its advantages in both multi-modal understanding and response politeness according to automated and human evaluations. Code and dataset are available at https://github.com/ChenDelong1999/polite-flamingo

Personal Bests and Gender

Abstract We connect two large bodies of scientific inquiry. First, important theories in the social sciences establish that human preferences are reference-dependent. Second, a separate field of research documents substantial differences in preferences and attitudes across genders. Specifically, we examine the universe of official classic chess games (more than 250,000 subjects and 22 million games). This allows us to study differences across genders both in cognitive performance (intensive margin) and in competitive participation (extensive margin), using the fact that personal bests act as reference points. We find that males and females behave very differently around their personal bests in both margins.

Exploring the potential of ChatGPT in medical dialogue summarization: a study on consistency with human preferences

BackgroundTelemedicine has experienced rapid growth in recent years, aiming to enhance medical efficiency and reduce the workload of healthcare professionals. During the COVID-19 pandemic in 2019, it became especially crucial, enabling remote screenings and access to healthcare services while maintaining social distancing. Online consultation platforms have emerged, but the demand has strained the availability of medical professionals, directly leading to research and development in automated medical consultation. Specifically, there is a need for efficient and accurate medical dialogue summarization algorithms to condense lengthy conversations into shorter versions focused on relevant medical facts. The success of large language models like generative pre-trained transformer (GPT)-3 has recently prompted a paradigm shift in natural language processing (NLP) research. In this paper, we will explore its impact on medical dialogue summarization.MethodsWe present the performance and evaluation results of two approaches on a medical dialogue dataset. The first approach is based on fine-tuned pre-trained language models, such as bert-based summarization (BERTSUM) and bidirectional auto-regressive Transformers (BART). The second approach utilizes a large language models (LLMs) GPT-3.5 with inter-context learning (ICL). Evaluation is conducted using automated metrics such as ROUGE and BERTScore.ResultsIn comparison to the BART and ChatGPT models, the summaries generated by the BERTSUM model not only exhibit significantly lower ROUGE and BERTScore values but also fail to pass the testing for any of the metrics in manual evaluation. On the other hand, the BART model achieved the highest ROUGE and BERTScore values among all evaluated models, surpassing ChatGPT. Its ROUGE-1, ROUGE-2, ROUGE-L, and BERTScore values were 14.94%, 53.48%, 32.84%, and 6.73% higher respectively than ChatGPT’s best results. However, in the manual evaluation by medical experts, the summaries generated by the BART model exhibit satisfactory performance only in the “Readability” metric, with less than 30% passing the manual evaluation in other metrics. When compared to the BERTSUM and BART models, the ChatGPT model was evidently more favored by human medical experts.ConclusionOn one hand, the GPT-3.5 model can manipulate the style and outcomes of medical dialogue summaries through various prompts. The generated content is not only better received than results from certain human experts but also more comprehensible, making it a promising avenue for automated medical dialogue summarization. On the other hand, automated evaluation mechanisms like ROUGE and BERTScore fall short in fully assessing the outputs of large language models like GPT-3.5. Therefore, it is necessary to research more appropriate evaluation criteria.

Human Prosocial Preferences Are Related to Slow-Wave Activity in Sleep.

Prosocial behavior is crucial for the smooth functioning of the society. Yet, individuals differ vastly in the propensity to behave prosocially. Here, we try to explain these individual differences under normal sleep conditions without any experimental modulation of sleep. Using a portable high-density EEG, we measured the sleep data in 54 healthy adults (28 females) during a normal night's sleep at the participants' homes. To capture prosocial preferences, participants played an incentivized public goods game in which they faced real monetary consequences. The whole-brain analyses showed that a higher relative slow-wave activity (SWA, an indicator of sleep depth) in a cluster of electrodes over the right temporoparietal junction (TPJ) was associated with increased prosocial preferences. Source localization and current source density analyses further support these findings. Recent sleep deprivation studies imply that sleeping enough makes us more prosocial; the present findings suggest that it is not only sleep duration, but particularly sufficient sleep depth in the TPJ that is positively related to prosociality. Because the TPJ plays a central role in social cognitive functions, we speculate that sleep depth in the TPJ, as reflected by relative SWA, might serve as a dispositional indicator of social cognition ability, which is reflected in prosocial preferences. These findings contribute to the emerging framework explaining the link between sleep and prosocial behavior by shedding light on the underlying mechanisms.

Batch Active Learning of Reward Functions from Human Preferences

Data generation and labeling are often expensive in robot learning. Preference-based learning is a concept that enables reliable labeling by querying users with preference questions. Active querying methods are commonly employed in preference-based learning to generate more informative data at the expense of parallelization and computation time. In this paper, we develop a set of novel algorithms, batch active preference-based learning methods, that enable efficient learning of reward functions using as few data samples as possible while still having short query generation times and also retaining parallelizability. We introduce a method based on determinantal point processes (DPP) for active batch generation and several heuristic-based alternatives. Finally, we present our experimental results for a variety of robotics tasks in simulation. Our results suggest that our batch active learning algorithm requires only a few queries that are computed in a short amount of time. We showcase one of our algorithms in a study to learn human users’ preferences.

Voice-based age, gender, and language recognition based on ResNet deep model and transfer learning in spectro-temporal domain

In personal identity recognition systems, detecting a person's age, gender, and language using voice signal characteristics is a crucial issue, especially because of the importance of security considerations. Age, gender, and language classification problems are important in signal processing because they are used to analyze and understand human behavior, interactions, and preferences. This can be especially useful in the fields of human-computer interaction, psychology, and social science research. In this paper, a new system for detecting a speaker's age, gender, and language based on deep learning models is presented. Deep learning models have shown great efficacy in various fields of signal processing. For this paper, a range of deep models were tested, including convolutional neural networks (CNNs), recurrent neural network (RNN), and a fine-tuning ResNet34 architecture. Additionally, techniques such as transfer learning were applied to improve the efficiency of the proposed system. The input voice signals are preprocessed by applying the spectro-temporal transform to obtain additional features that can be fed to the ResNet34 model, which is designed specifically for the task of voice signal processing. The dataset used in this paper was sourced from the Mozilla common voice initiative, which is dedicated to advancing speech recognition and language identification technologies. The performance of the proposed algorithm was evaluated in the presence of Gaussian noise to determine its robustness. The experimental results demonstrated that the proposed algorithm significantly outperformed basic algorithms and other deep neural networks in terms of age and gender recognition from voice signals.

Fuzzy-Set-Based Multi-Attribute Decision-Making, Its Computing Implementation, and Applications

This paper reflects the results of research analyzing models of multi-attribute decision-making based on fuzzy preference relations. Questions of constructing the corresponding multi-attribute models to deal with quantitative information concomitantly with qualitative information based on experts’ knowledge are considered. Human preferences may be represented within the fuzzy preference relations and by applying diverse other preference formats. Considering this, so-called transformation functions reduce any preference format to fuzzy preference relations. This paper’s results can be applied independently or as part of a general approach to solving a wide class of problems with fuzzy coefficients, as well as within the framework of a general scheme of multi-criteria decision-making under conditions of uncertainty. The considered techniques for fuzzy preference modeling are directed at assessing, comparing, choosing, prioritizing, and/or ordering alternatives. These techniques have served to develop a computing system for multi-attribute decision-making. It has been implemented in the C# programming language, utilizing the “.NET” framework. The computing system allows one to represent decision-makers’ preferences in one of five preference formats. These formats and quantitative estimates are reduced to nonreciprocal fuzzy preference relations, providing homogeneous preference information for decision procedures. This paper’s results have a general character and were applied to analyze power engineering problems.

Contextual panel conditioning and reward models in large language models

Direct preference optimization (DPO) aims to match human preferences while reducing the complexity of reinforcement learning. Traditional methods such as reinforcement learning with human feedback (RLHF) first match reward models with cues and preferences, and then use reinforcement learning (RL) to find policies that maximize rewards. In contrast, DPO simplifies the process by directly optimizing the policy to satisfy preferences without explicit reward functions or RL processes. DPO is a more direct and potentially more efficient way to fine-tune a language model to remain consistent with human feedback. Additionally, OpenAI mentioned that they trained the model by imitating human ratings to help improve RLHF. The next step is to fit the model to a data set containing rich "conditions". For example, the training model generates a panel containing memories, conditions, goals, plans, and future tasks, and uses this panel for training. These conditions transform the "creative writing task" into the task of "distributing materials", reducing entropy in creative writing. Conditional reinforcement learning fine-tuning (C-RLFT) enables large language models to understand and generate human-like text, adapt to new information, and personalize responses while maintaining relevance and coherence. Future improvements include improving conditional panels using RLHF or RLAIF, iteration between datasets and models, aligning models with real-world needs, and building new base models based on 0-order optimization. These directions aim to make large language models more efficient, consistent with human preferences, and able to run in a variety of environments, including edge computing devices. Hello, here is some text without a meaning. This text should show what a printed text will look like at this place. If you read this text, you will get no information. Really? Is there no information? Is there a difference between this text and some nonsense like "Huardest gefburn"? Kjift – not at all! A blind text like this gives you information about the selected font, how the letters are written and an impression of the look. This text should contain all letters of the alphabet and it should be written in the original language. There is no need for special content, but the length of words should match the language.

On not being alone in lonely places: preferences, goods, and aesthetic-ethical conflict in nature sports

ABSTRACT Ethical questions normally arise in sport because its participants are human moral agents and because its practice community entails the observance of rules and responsibilities that humans generally owe one another in a social practice of voluntary competition. Since nature sports are not defined by this kind of inter-agential activity, it would appear that there are no comparable ethical constraints on their pursuit. This paper considers conflicts of preference versus right between humans, how these are resolved, and whether these rights are relevant in assessment of nature sport activity vis-à-vis nonhuman creatures. Relying on a goods perspective instead of a rights framework, via Korsgaard, I argue against an assumption that human preference is sufficient to override consideration of nonhuman animals’ functional goods.

One and the same? How similar are basic human values and economic preferences.

Both the basic human values approach and the economic preference approach have been developed and applied to represent fundamental drivers of human behavior in various domains by measuring people's underlying preferences and motivations. Both of them have been used, however, in isolation from each other, the former primarily in social psychology studies and the latter mainly in economic studies. But how similar are they? Finding that they differ may suggest that combining them to explain human behavior might be beneficial. To the best of our knowledge, only a few studies have attempted to explore and empirically examine the theoretical and empirical link between variables in both approaches. The current study tries to fill this gap by examining relations between basic human values and major economic preferences. We examine the associations between the values of self-transcendence, self-enhancement, openness to change and conservation, and the economic preferences of risk aversion (or seeking), altruism, trust, and positive and negative reciprocity. We propose mechanisms as to how they may be associated with one another. For example, we expect an association between conservation and risk aversion as both are motivated by attributing importance to stability and the status quo, or between self-transcendence and altruism, as both are motivated by concern for others. For the empirical analysis we employed convenience samples collected in Poland and Germany. Results in both samples support our expectations: several values and economic preferences are linked in theoretically predictable ways, but only to a weak or moderate extent. We conclude that they are not mutually exclusive but may rather be complementary, and therefore likely both relevant for investigations into explaining behavior.

Climate and biodiversity change constrain the flow of cultural ecosystem services to people: A case study modeling birding across Africa under future climate scenarios

Global change is currently impacting ecosystems and their contributions to people (i.e. ecosystem services). These impacts have consequences for societies and human well-being, especially in Africa. Historically, efforts have focused on assessing global change from a social or biophysical perspective, treating them as separate entities. Yet, our understanding of impacts to social-ecological systems remains limited, particularly in the Global South, due to a lack of data, tools, and approaches accounting for social and ecological aspects of ecosystem services. This is especially relevant for cultural ecosystem services as they are less tangible. We use a simple indicator and important provider of a multitude of cultural ecosystem services, birding, to understand how climate, biodiversity, and land use change will impact cultural ecosystem services across Africa. We explore how emerging tools and data can overcome limitations in mapping and modeling cultural ecosystem services, particularly in analyzing human preferences and behavior at large spatiotemporal scales and in data-poor regions. Leveraging crowdsourced data from eBird and using machine learning techniques we map and model recreational birding to assess the underlying social-ecological relationships and the impact of future climate and environmental change. We show that bird species richness, protected areas, accessibility, and max temperature contribute most to birding suitability across the continent. Further, we show spatial shifts in the suitability of birding under three future climate scenarios (SSP126, 370, and 585). Models suggest climate and biodiversity change will increasingly constrain the flow of birding related cultural ecosystem services across Africa. This has implications for human-nature interactions, development of countries, management of protected areas, and overall human well-being in the future. More generally, we highlight opportunities for crowdsourced datasets and machine learning to integrate non-material ecosystem services in models and thus, enhance the understanding of future impacts to ecosystem services and human well-being.

Exotic pet owners’ preferences for different ectothermic taxa are based on species traits and purchase prices in the United States

The exotic pet trade has resulted in substantial invasion and disease risks, owing to the release of pets into new environments. Scientists have conjectured that pet owners acquire and release species with undesirable traits because they are imperfectly informed about the traits of these animals. However, few studies have used social science methods to elicit pet owners’ preferences for exotic pets. In 2019 we administered a best-worst choice survey to 1,055 exotic pet owners in the United States (who own pet reptiles, amphibians, fish, or invertebrates) to examine how human preferences and incomplete information may contribute to the risks of the exotic pet trade. Respondents preferred colorful and patterned species. On average, respondents preferred medium-sized amphibians and reptiles, small fish, and large invertebrates, although they demonstrated heterogeneity in preferences with respect to the adult size of pets. Respondents also preferred amphibians and reptiles with medium life expectancies and fish and invertebrates with long life expectancies, although they again demonstrated heterogeneity in preferences with respect to pets’ life span. Respondents preferred docile animals, and were more likely to purchase lower-cost pets. We found some evidence that respondents’ decision to purchase exotic pets depended on whether these animals were native, rare, had unusual morphological features, and breed easily. Respondents’ decision to purchase specific taxa as exotic pets also depended on their age, education, and housing. Most respondents stated that they searched for information on pets’ diet, behavior, adult size, life span, costs of care such as equipment or veterinary costs, and whether the animal was captive bred before purchasing these animals. Excepting pets’ diets, fewer than half of respondents had been offered information on pets’ traits by sellers. On average, respondents rated the information they had been offered as average. Respondents typically obtained additional information about pets from online searches. Our results suggest that certification systems that provide critical information on exotic pets’ behaviors, adult size, longevity, fecundity, and husbandry needs should be implemented to prevent pet owners acquiring animals that they may subsequently abandon.

The moral machine experiment on large language models.

As large language models (LLMs) have become more deeply integrated into various sectors, understanding how they make moral judgements has become crucial, particularly in the realm of autonomous driving. This study used the moral machine framework to investigate the ethical decision-making tendencies of prominent LLMs, including GPT-3.5, GPT-4, PaLM 2 and Llama 2, to compare their responses with human preferences. While LLMs' and humans' preferences such as prioritizing humans over pets and favouring saving more lives are broadly aligned, PaLM 2 and Llama 2, especially, evidence distinct deviations. Additionally, despite the qualitative similarities between the LLM and human preferences, there are significant quantitative disparities, suggesting that LLMs might lean toward more uncompromising decisions, compared with the milder inclinations of humans. These insights elucidate the ethical frameworks of LLMs and their potential implications for autonomous driving.

Better Together: 14-Month-Old Infants Expect Agents to Cooperate.

Humans engage in cooperative activities from early on and the breadth of human cooperation is unparalleled. Human preference for cooperation might reflect cognitive and motivational mechanisms that drive engagement in cooperative activities. Here we investigate early indices of humans' cooperative abilities and test whether 14-month-old infants expect agents to prefer cooperative over individual goal achievement. Three groups of infants saw videos of agents facing a choice between two actions that led to identical rewards but differed in the individual costs. Our results show that, in line with prior research, infants expect agents to make instrumentally rational choices and prefer the less costly of two individual action alternatives. In contrast, when one of the action alternatives is cooperative, infants expect agents to choose cooperation over individual action, even though the cooperative action demands more effort from each agent to achieve the same outcome. Finally, we do not find evidence that infants expect agents to choose the less costly alternative when both options entail cooperative action. Combined, these results indicate an ontogenetically early expectation of cooperation, and raise interesting implications and questions regarding the nature of infants' representations of cooperative actions and their utility.

Vascular Underpinnings of Cerebral Lateralisation in the Neonate

Traditionally, adult and neonatal cerebral perfusion have been presumed to be symmetrical. Contrary to this, our adult work shows that supra-aortic cerebral supply is systematically biased towards the left, in terms of both vessel geometry and blood flow volumes. Although this asymmetry is meaningfully related to hand preference, the developmental origins of this association remain unknown. Our detailed investigations of the cerebral vasculature confirm analogous asymmetries in term neonates. Specifically, we demonstrate that the structure and flow of neonatal middle cerebral vessels are consistently asymmetric and predominantly left-dominant. Building on our work from the same cohort, we now report further analyses of these new-found asymmetries. Namely, exploring for the first time, the relationship between arterial lateral biases and the neonatal head-turning response—a reliable early behavioural precursor of handedness that shows a systematic rightward bias in the population. Here, we demonstrate a contralateral relationship between vessel morphology and primitive expressions of lateralisation that predate the establishment of definitive handedness in the course of postnatal development. This relationship mimics patterns observed in adults and suggests that lateralising trends in angiogenesis may ultimately influence the emergence of human lateral preferences.

Effect of Adapting to Human Preferences on Trust in Human-Robot Teaming

We present the effect of adapting to human preferences on trust in a human-robot teaming task. The team performs a task in which the robot acts as an action recommender to the human. It is assumed that the behavior of the human and the robot is based on some reward function they try to optimize. We use a new human trust-behavior model that enables the robot to learn and adapt to the human's preferences in real-time during their interaction using Bayesian Inverse Reinforcement Learning. We present three strategies for the robot to interact with a human: a non-learner strategy, in which the robot assumes that the human's reward function is the same as the robot's, a non-adaptive learner strategy that learns the human's reward function for performance estimation, but still optimizes its own reward function, and an adaptive-learner strategy that learns the human's reward function for performance estimation and also optimizes this learned reward function. Results show that adapting to the human's reward function results in the highest trust in the robot.

Curve Trajectory Model for Human Preferred Path Planning of Automated Vehicles

Automated driving systems are often used for lane keeping tasks. By these systems, a local path is planned ahead of the vehicle. However, these paths are often found unnatural by human drivers. In response to this, this paper proposes a linear driver model, which can calculate node points reflective of human driver preferences and based on these node points a human driver preferred motion path can be designed for autonomous driving. The model input is the road curvature, effectively harnessed through a self-developed Euler-curve-based curve fitting algorithm. A comprehensive case study is undertaken to empirically validate the efficacy of the proposed model, demonstrating its capacity to emulate the average behavioral patterns observed in human curve path selection. Statistical analyses further underscore the model's robustness, affirming the authenticity of the established relationships. This paradigm shift in trajectory planning holds promising implications for the seamless integration of autonomous driving systems with human driving preferences.

Multimodal Learning‐Based Proactive Human Handover Intention Prediction Using Wearable Data Gloves and Augmented Reality

Efficient object handover between humans and robots holds significant importance within collaborative manufacturing environments. Enhancing the efficacy of human–robot handovers involves enabling robots to comprehend and foresee human handover intentions. This article introduces human‐teaching–robot‐learning‐prediction framework, allowing robots to learn from diverse human demonstrations and anticipate human handover intentions. The framework facilitates human programming of robots through demonstrations utilizing augmented reality and a wearable dataglove, aligned with task requirements and human working preferences. Subsequently, robots enhance their cognitive capabilities by assimilating insights from human handover demonstrations, utilizing deep neural network algorithms. Furthermore, robots can proactively seek clarification from humans via an augmented reality system when confronted with ambiguity in human intentions, mirroring how humans seek clarity from their counterparts. This proactive approach empowers robots to anticipate human intentions and assist human partners during handovers. Empirical results underscore the benefits of the proposed approach, demonstrating highly accurate prediction of human intentions in human–robot handover tasks.

The economic standing of animals

This article considers how cost-benefit analyses of farm animal policy would be altered if animal interests were to be directly included in the computations. Currently, animals lack standing, their preferences for improved living conditions are not considered in the calculation of costs or benefits of farm policies. While animal welfare receives consideration to the extent that humans value it, human preferences are not fully revealed due to incomplete information and the public good nature of animal welfare. Uncertainty associated with the proper role of farm animal standing in cost-benefit analysis validates the avoidance of policies that would be undesirable if animals possessed standing in cost-benefit analyses.

Nutritional habits of nurses during the COVID-19 pandemic. Review of the bibliography

Human dietary preferences and choices are the end result of the interaction of biological, psychological, cultural, economic and social parameters. Eating disorders are a major issue nowadays and the number of people affected is constantly increasing. Various factors influence nurses' healthy eating patterns in the workplace, such as overtime work, cyclical work hours, job stress, the physical work environment, food options offered by hospital canteens, and social factors. Reduced monthly income has a negative effect on nurses' eating habits. Financially challenged nurses are unable to purchase healthy food, especially after the rampant rise in food prices caused by the COVID-19 pandemic. To meet patients' needs, nurses often cut back on meals, which affects their eating habits. Nurses constantly express the need for a calm and dignified place to eat and rest. Quarantine during the COVID-19 pandemic has been blamed for changes in eating habits, weight gain and increased consumption of sweets and snacks. These changes in dietary habits have significant implications for the health impact of the population during the pandemic. Policy makers should ensure that nurses have the necessary time to eat lunch. Nursing departments should implement programs to promote the health of nurses, where their weight will be controlled and special dietary patterns will be given, as well as psychological support to reduce work-related stress.