Search

I report here a comprehensive analysis about the political preferences embedded in Large Language Models (LLMs). Namely, I administer 11 political orientation tests, designed to identify the political preferences of the test taker, to 24 state-of-the-art conversational LLMs, both closed and open source. When probed with questions/statements with political connotations, most conversational LLMs tend to generate responses that are diagnosed by most political test instruments as manifesting preferences for left-of-center viewpoints. This does not appear to be the case for five additional base (i.e. foundation) models upon which LLMs optimized for conversation with humans are built. However, the weak performance of the base models at coherently answering the tests' questions makes this subset of results inconclusive. Finally, I demonstrate that LLMs can be steered towards specific locations in the political spectrum through Supervised Fine-Tuning (SFT) with only modest amounts of politically aligned data, suggesting SFT's potential to embed political orientation in LLMs. With LLMs beginning to partially displace traditional information sources like search engines and Wikipedia, the societal implications of political biases embedded in LLMs are substantial. [ABSTRACT FROM AUTHOR]

Recent advancements in Large Language Models (LLMs) suggest imminent commercial applications of such AI systems where they will serve as gateways to interact with technology and the accumulated body of human knowledge. The possibility of political biases embedded in these models raises concerns about their potential misusage. In this work, we report the results of administering 15 different political orientation tests (14 in English, 1 in Spanish) to a state-of-the-art Large Language Model, the popular ChatGPT from OpenAI. The results are consistent across tests; 14 of the 15 instruments diagnose ChatGPT answers to their questions as manifesting a preference for left-leaning viewpoints. When asked explicitly about its political preferences, ChatGPT often claims to hold no political opinions and to just strive to provide factual and neutral information. It is desirable that public facing artificial intelligence systems provide accurate and factual information about empirically verifiable issues, but such systems should strive for political neutrality on largely normative questions for which there is no straightforward way to empirically validate a viewpoint. Thus, ethical AI systems should present users with balanced arguments on the issue at hand and avoid claiming neutrality while displaying clear signs of political bias in their content. [ABSTRACT FROM AUTHOR]

This work analyzes the prevalence of words denoting prejudice in 27 million news and opinion articles written between 1970 and 2019 and published in 47 of the most popular news media outlets in the United States. Our results show that the frequency of words that denote specific prejudice types related to ethnicity, gender, sexual, and religious orientation has markedly increased within the 2010–2019 decade across most news media outlets. This phenomenon starts prior to, but appears to accelerate after, 2015. The frequency of prejudice-denoting words in news articles is not synchronous across all outlets, with the yearly prevalence of such words in some influential news media outlets being predictive of those words' usage frequency in other outlets the following year. Increasing prevalence of prejudice-denoting words in news media discourse is often substantially correlated with U.S. public opinion survey data on growing perceptions of minorities' mistreatment. Granger tests suggest that the prevalence of prejudice-denoting terms in news outlets might be predictive of shifts in public perceptions of prejudice severity in society for some, but not all, types of prejudice. [ABSTRACT FROM AUTHOR]

This work describes a chronological (2000–2019) analysis of sentiment and emotion in 23 million headlines from 47 news media outlets popular in the United States. We use Transformer language models fine-tuned for detection of sentiment (positive, negative) and Ekman's six basic emotions (anger, disgust, fear, joy, sadness, surprise) plus neutral to automatically label the headlines. Results show an increase of sentiment negativity in headlines across written news media since the year 2000. Headlines from right-leaning news media have been, on average, consistently more negative than headlines from left-leaning outlets over the entire studied time period. The chronological analysis of headlines emotionality shows a growing proportion of headlines denoting anger, fear, disgust and sadness and a decrease in the prevalence of emotionally neutral headlines across the studied outlets over the 2000–2019 interval. The prevalence of headlines denoting anger appears to be higher, on average, in right-leaning news outlets than in left-leaning news media. [ABSTRACT FROM AUTHOR]

The term political extremism is commonly used to refer to political attitudes considered to be outside the ideological mainstream. This study leverages computational content analysis of big data to longitudinally examine (1970–2019) the prevalence of terms denoting far-right and far-left political extremism in more than 30 million written news and opinion articles from 54 news media outlets popular in the United States and the United Kingdom. We find that the usage of terms denoting right and left political extremism has been increasing across news media outlets in both countries. This trend is particularly stark for far-right-denoting terms, which have been growing in prevalence since at least 2008. Most U.S. and U.K. news media outlets tend to use far-right-denoting terms substantially more often than they use far-left-denoting terms. The rising prevalence in news media of terms denoting political extremism is strongly correlated with similar growing usage of terms denoting prejudice and social justice discourse. [ABSTRACT FROM AUTHOR]

Concerns about gender bias in word embedding models have captured substantial attention in the algorithmic bias research literature. Other bias types however have received lesser amounts of scrutiny. This work describes a large-scale analysis of sentiment associations in popular word embedding models along the lines of gender and ethnicity but also along the less frequently studied dimensions of socioeconomic status, age, physical appearance, sexual orientation, religious sentiment and political leanings. Consistent with previous scholarly literature, this work has found systemic bias against given names popular among African-Americans in most embedding models examined. Gender bias in embedding models however appears to be multifaceted and often reversed in polarity to what has been regularly reported. Interestingly, using the common operationalization of the term bias in the fairness literature, novel types of so far unreported bias types in word embedding models have also been identified. Specifically, the popular embedding models analyzed here display negative biases against middle and working-class socioeconomic status, male children, senior citizens, plain physical appearance and intellectual phenomena such as Islamic religious faith, non-religiosity and conservative political orientation. Reasons for the paradoxical underreporting of these bias types in the relevant literature are probably manifold but widely held blind spots when searching for algorithmic bias and a lack of widespread technical jargon to unambiguously describe a variety of algorithmic associations could conceivably be playing a role. The causal origins for the multiplicity of loaded associations attached to distinct demographic groups within embedding models are often unclear but the heterogeneity of said associations and their potential multifactorial roots raises doubts about the validity of grouping them all under the umbrella term bias. Richer and more fine-grained terminology as well as a more comprehensive exploration of the bias landscape could help the fairness epistemic community to characterize and neutralize algorithmic discrimination more efficiently. [ABSTRACT FROM AUTHOR]

The term diversity can be operationalized demographically (in terms of physical or external characteristics such as race, gender, ethnicity and nationality) or intellectually (in terms of mental phenomena such as viewpoints, beliefs, ideas and political opinion). This work examines the context in which the concept of diversity is used by 50 US elite universities in their online institutional presence. Distributional semantics theory is leveraged to quantify semantic similarity between linguistic items based on their distributional properties in a large sample of language data taken from universities online profiles. The language modelling is carried out using Word2vec, a state-of-the-art machine learning model widely used by the natural language processing community to create vector representations of words (i.e. word embeddings). The model uses a neural network trained to reconstruct the linguistic context of words in the training corpus. As a by-product of the training objective, word2vec embeds words into a learned vector space where words that share common contexts and thus semantic meaning according to the distributional hypotheses, are located in close proximity to one another. A quantitative analysis of cosine similarities between word vectors derived from the corpus of text retrieved from universities online institutional profiles shows that the diversity concept is much closer to demographic operationalisations of diversity such as race, gender, ethnicity or nationality than to intellectual ones such as viewpoints, values, beliefs or political orientation. That is, the universities studied tend to use the word diversity predominantly in its demographic denotation to refer to variety of external appearance instead of to variety of mental phenomena. This is significant in light of the severe lack of ideological diversity in universities across the US, with the vast majority of faculty leaning left of center. Universities emphasis on the usage of the term diversity to denote demographic subtypes of diversity could be indicative of a majority power structure in the Academy which tries to hinder the fostering of viewpoint diversity by steering diversity efforts towards demographic interpretations of the word. At the very least, the results of this work suggest that universities, as judged from the way they use language in their own online institutional profiles, prioritize demographic types of diversity around variety of external appearance cues over intellectual heterogeneity. [ABSTRACT FROM AUTHOR]

Human-computer interaction systems that bypass manual control can be beneficial for many use cases, including users with severe motor disability. We investigated pupillometry (inferring mental activity via dilations of the pupil) as an interaction method because it is noninvasive, easy to analyse, and increasingly available for practical development. In 3 experiments we investigated the efficacy of using pupillometry to detect imaginary motor movements of the hand. In Experiment 1 we demonstrated that, on average, the pupillary response is greater when the participant is imagining a hand-grasping motion, as compared with the control condition. In Experiment 2 we investigated how imaginary hand-grasping affects the pupillary response over time. In Experiment 3 we employed a simple classifier to demonstrate single-trial detection of imagined motor events using pupillometry. Using the mean pupil diameter of a single trial, accuracy rates as high as 71.25%, were achieved. Implications for the development of a pupillometry-based switch and future directions are discussed. [ABSTRACT FROM AUTHOR]

We are developing a novel system to improve arm function in stroke patients who have no, or only residual upper limb movement. Such a system fills an important gap in treatment options for people with little-to-no upper limb movement after stroke, and for whom regular treatments often are unsuitable. The system provides real-time visual and proprioceptive feedback of the arm plus the ability for participants to steer the movement direction of the arm through an assistive movement platform. The patient controls the system by simply looking at stimuli and engaging in motor imagery. The patient gaze is monitored with an eye tracker and motor output intentions are monitored with an EEG-based brain computer interface. Stimuli are presented as games in order to create a motivating rehabilitation environment. In this paper we discuss our motivation and design of the system. [ABSTRACT FROM AUTHOR]

Combining signals from traditional electroencephalography with those of novel physiological measures such as pupil dilation could enable more accurate and robust real-time monitoring of cognitive workload. [ABSTRACT FROM AUTHOR]

Non-contact measurements of cardiac pulse can provide robust measurement of heart rate (HR) without the annoyance of attaching electrodes to the body. In this paper we explore a novel and reliable method to carry out videobased HR estimation and propose various performance improvement over existing approaches. The investigated method uses Independent Component Analysis (ICA) to detect the underlying HR signal from a mixed source signal present in the RGB channels of the image. The original ICA algorithm was implemented and several modifications were explored in order to determine which one could be optimal for accurate HR estimation. Using statistical analysis, we compared the cardiac pulse rate estimation from the different methods under comparison on the extracted videos to a commercially available oximeter. We found that some of these methods are quite effective and efficient in terms of improving accuracy and latency of the system. We have made the code of our algorithms openly available to the scientific community so that other researchers can explore how to integrate video-based HR monitoring in novel health technology applications. We conclude by noting that recent advances in video-based HR monitoring permit computers to be aware of a user's psychophysiological status in real time. [ABSTRACT FROM AUTHOR]

For individuals with high degrees of motor disability or locked-in syndrome, it is impractical or impossible to use mechanical switches to interact with electronic devices. Brain computer interfaces (BCIs) can use motor imagery to detect interaction intention from users but lack the accuracy of mechanical switches. Hence, there exists a strong need to improve the accuracy of EEG-based motor imagery BCIs attempting to implement an on/off switch. Here, we investigate how monitoring the pupil diameter of a person as a psycho-physiological parameter in addition to traditional EEG channels can improve the classification accuracy of a switch-like BCI. We have recently noticed in our lab (work not yet published) how motor imagery is associated with increases in pupil diameter when compared to a control rest condition. The pupil diameter parameter is easily accessible through video oculography since most gaze tracking systems report pupil diameter invariant to head position. We performed a user study with 30 participants using a typical EEG based motor imagery BCI. We used common spatial patterns to separate motor imagery, signaling movement intention, from a rest control condition. By monitoring the pupil diameter of the user and using this parameter as an additional feature, we show that the performance of the classifier trying to discriminate motor imagery from a control condition improves over the traditional approach using just EEG derived features. Given the limitations of EEG to construct highly robust and reliable BCIs, we postulate that multi-modal approaches, such as the one presented here that monitor several psycho-physiological parameters, can be a successful strategy in making BCIs more accurate and less vulnerable to constraints such as requirements for long training sessions or high signal to noise ratio of electrode channels. [ABSTRACT FROM AUTHOR]

Previous scholarly literature has documented a pronounced increase in the prevalence of prejudice-denoting terms in American news media content. Some have referred to this shift in journalistic discourse and related public opinion trends signaling increasing perceptions of prejudice severity in U.S. society as The Great Awokening. This work analyzes whether the increasing prevalence of prejudice themes in American news media outlets has been replicated in the news media ecosystem of a Spanish-speaking country. Thus, we computationally analyzed the prevalence of words denoting prejudice in five million news and opinion articles written between 1976 and 2019 and published in three of the most widely read newspapers in Spain: El País, El Mundo and ABC. We report that within the studied time period, the frequency of terms that denote specific prejudice types related to gender, ethnicity, sexuality and religious orientation has also substantially increased across the analyzed Spanish news media outlets. There are, however, some notable distinctions in the long-term usage dynamics of prejudice-denoting terms between the leading Spanish newspaper of record, El País, and its U.S. counterpart, The New York Times. [ABSTRACT FROM AUTHOR]

Ascertaining when and where genes are expressed is of crucial importance to understanding or predicting the physiological role of genes and proteins and how they interact to form the complex networks that underlie organ development and function. It is, therefore, crucial to determine on a genome-wide level, the spatio-temporal gene expression profiles at cellular resolution. This information is provided by colorimetric RNA in situ hybridization that can elucidate expression of genes in their native context and does so at cellular resolution. We generated what is to our knowledge the first genomewide transcriptome atlas by RNA in situ hybridization of an entire mammalian organism, the developing mouse at embryonic day 14.5. This digital transcriptome atlas, the Eurexpress atlas (http://www.eurexpress.org), consists of a searchable database of annotated images that can be interactively viewed. We generated anatomy-based expression profiles for over 18,000 coding genes and over 400 microRNAs. We identified 1,002 tissue-specific genes that are a source of novel tissue-specific markers for 37 different anatomical structures. The quality and the resolution of the data revealed novel molecular domains for several developing structures, such as the telencephalon, a novel organization for the hypothalamus, and insight on the Want network involved in renal epithelial differentiation during kidney development. The digital transcriptome atlas is a powerful resource to determine co-expression of genes, to identify cell populations and lineages, and to identify functional associations between genes relevant to development and disease. [ABSTRACT FROM AUTHOR]

The COVID-19 pandemic has been one of the most disruptive and painful phenomena of the last few decades. As of July 2021, the origins of the SARS-CoV-2 virus that caused the outbreak remain a mystery. This work analyzes the prevalence in news media articles of two popular hypotheses about SARS-CoV-2 virus origins: the natural emergence and the lab-leak hypotheses. Our results show that for most of 2020, the natural emergence hypothesis was favored in news media content while the lab-leak hypothesis was largely absent. However, something changed around May 2021 that caused the prevalence of the lab-leak hypothesis to substantially increase in news media discourse. This shift has not been uniformed across media organizations but instead has manifested itself more acutely in some outlets than others. Our structural break analysis of daily news media usage of terms related to the laboratory escape hypothesis provides hints about potential sources for this sudden shift in the prevalence of the lab-leak hypothesis in prestigious news media. [ABSTRACT FROM AUTHOR]

The introduction of next-generation technologies to the maritime shipping industry, including Portable Pilotage Units, Remote Pilotage, advanced situation awareness aids, and Autonomous Shipping, creates an urgent need to understand operator workload during Bridge Team operations, and co-operations with shore based personnel. In this paper we analyse mental workload of maritime Captains, Pilots and Tug Masters during standard and emergency scenarios, using traditional measures (SWAT, ISA), communications analysis, and the collection of simultaneous electro-dermal activity (EDA) of team members. Results indicate that the EDA measure overcomes some of the problems with paper-based techniques, and has excellent temporal resolution for emergency events. Implications for testing of novel technologies are discussed. [ABSTRACT FROM AUTHOR]