Easy to Learn
Discovering state-of-the-art reinforcement learning algorithms
Junhyuk Oh et al.
Nature, forthcoming
Abstract:
Humans and other animals use powerful reinforcement learning (RL) mechanisms that have been discovered by evolution over many generations of trial and error. By contrast, artificial agents typically learn using hand-crafted learning rules. Despite decades of interest, the goal of autonomously discovering powerful RL algorithms has proven elusive. In this work, we show that it is possible for machines to discover a state-of-the-art RL rule that outperforms manually-designed rules. This was achieved by meta-learning from the cumulative experiences of a population of agents across a large number of complex environments. Specifically, our method discovers the RL rule by which the agent's policy and predictions are updated. In our large-scale experiments, the discovered rule surpassed all existing rules on the well-established Atari benchmark and outperformed a number of state-of-the-art RL algorithms on challenging benchmarks that it had not seen during discovery. Our findings suggest that the RL algorithms required for advanced artificial intelligence may soon be automatically discovered from the experiences of agents, rather than manually designed.
Revealing emergent human-like conceptual representations from language prediction
Ningyu Xu et al.
Proceedings of the National Academy of Sciences, 4 November 2025
Abstract:
People acquire concepts through rich physical and social experiences and use them to understand and navigate the world. In contrast, large language models (LLMs), trained solely through next-token prediction on text, exhibit strikingly human-like behaviors. Are these models developing concepts akin to those of humans? If so, how are such concepts represented, organized, and related to behavior? Here, we address these questions by investigating the representations formed by LLMs during an in-context concept inference task. We found that LLMs can flexibly derive concepts from linguistic descriptions in relation to contextual cues about other concepts. The derived representations converge toward a shared, context-independent structure, and alignment with this structure reliably predicts model performance across various understanding and reasoning tasks. Moreover, the convergent representations effectively capture human behavioral judgments and closely align with neural activity patterns in the human brain, providing evidence for biological plausibility. Together, these findings establish that structured, human-like conceptual representations can emerge purely from language prediction without real-world grounding, highlighting the role of conceptual structure in understanding intelligent behavior. More broadly, our work suggests that LLMs offer a tangible window into the nature of human concepts and lays the groundwork for advancing alignment between artificial and human intelligence.
The cost of saving: How photos and screenshots impair memory
Rebecca Lurie, Sophia Fabrizio & Deanne Westerman
Memory & Cognition, October 2025, Pages 2301-2311
Abstract:
The photo impairment effect refers to worse memory for experiences that are photographed compared with those that are not. One proposed explanation for this effect is that photo-taking divides attention between the event and the actions required for photography. However, the results of the present study challenge this account. Specifically, we found that the magnitude of the photo impairment effect did not increase with task complexity, undermining the idea that divided attention is the primary cause. Across three experiments, participants viewed art presented on a computer or their own smartphones and either photographed or took screenshots of the images. Memory was consistently worse for images saved by any of these methods. Notably, screenshotting had particularly detrimental effects on memory, despite being the least complex saving method. Furthermore, the impairment did not vary based on whether participants used a familiar device (their own smartphone) or an unfamiliar, experimenter-provided camera. These findings suggest that divided attention alone cannot account for the photo impairment effect.
Detecting Deepfakes Through Emotion?: Facial Expression and Emotional Contagion as Dual Indicators of Deepfake Credibility
Jiyoung Lee & Kevin John
Applied Cognitive Psychology, November/December 2025
Abstract:
This study examines the differences in emotional expressions between face-swap deepfake videos featuring real human subjects and their authentic counterparts and investigates how these discrepancies influence viewers' emotional reactions and perceptions of video credibility. Using a two-step approach, we first applied computer-based facial expression analysis to compare emotional displays between deepfake and authentic videos. Next, guided by emotional contagion and emotion-as-information frameworks, we conducted an audience response analysis to assess how displayed emotions in the two types of videos transfer to viewers' emotional experiences and their subsequent assessments of the videos. Results indicate that deepfakes generally exhibit lower overall and negative emotions compared to authentic counterparts. Notably, audiences' reduced emotional responses to deepfakes were associated with higher perceived credibility. These findings underscore the importance of emotion-based signals for detecting fabricated videos and highlight the relationship between viewers' emotional responses and their perceived trust in AI-generated content.
Forecasting Social Science: Evidence from 100 Projects
Stefano DellaVigna & Eva Vivalt
NBER Working Paper, November 2025
Abstract:
Forecasts about research findings affect critical scientific decisions, such as what treatments an R&D lab invests in, or which papers a researcher decides to write. But what do we know about the accuracy of these forecasts? We analyze a unique data set of all 100 projects posted on the Social Science Prediction Platform from 2020 to 2024, which received 53,298 forecasts in total, including 66 projects for which we also have results. We show that forecasters, on average, over-estimate treatment effects; however, the average forecast is quite predictive of the actual treatment effect. We also examine differences in accuracy across forecasters. Academics have a slightly higher accuracy than non-academics, but expertise in a field does not increase accuracy. A panel of motivated repeat forecasters has higher accuracy, but this does not extend more broadly to all repeat forecasters. Confidence in the accuracy of one's forecasts is perversely associated with lower accuracy. We also document substantial cross-study correlation in accuracy among forecasters and identify a group of "superforecasters". Finally, we relate our findings to results in the literature as well as to expert forecasts.
Super-recognizers sample visual information of superior computational value for facial recognition
James Dunn et al.
Proceedings of the Royal Society: Biological Sciences, November 2025
Abstract:
Super-recognizers -- individuals with exceptionally high face recognition abilities -- are a key exemplar of biological visual expertise. Recent eye-tracking evidence suggests that their expertise may be driven by exploratory viewing behaviour during learning, but it remains unclear whether this perceptual sampling is functional for face identity processing. Here, we develop a novel approach to quantify the computational value of face information samples and test the utility of information sampling in super-recognizers. Using measurements of eye gaze behaviour, we reconstructed the retinal information that participants acquired while learning new faces. We then evaluated the computational value of this information for face identity processing using nine deep neural networks (DNNs) optimized for this task. Identity matching accuracy improved across all DNNs when using visual information sampled by super-recognizers compared with typical viewers. Interestingly, this advantage could not be explained by the greater quantity of information alone, and so differences in both the quantity and quality of face information encoded on the retina contribute to individual differences in face processing ability. These findings support accounts of visual expertise that emphasize attentional mechanisms and the role of active visual exploration in learning.
Brawn and Brainpower: Acute Resistance Exercise Improves Behavioral and Neuroelectric Measures of Executive Function
Nicholas Baumgartner et al.
Psychophysiology, November 2025
Abstract:
Acute resistance exercise (RE) is emerging as a promising strategy to improve executive function, yet the underlying mechanisms remain unclear. In this study, 121 participants (aged 18–50) were randomly assigned to either a moderate-intensity RE or rest intervention using a between-subjects design. We examined the effects of acute RE on behavioral and neuroelectric measures of executive function during a modified Flanker task and an N-back task, and collected lactate and blood pressure to explore possible physiological mechanisms. Results showed that following acute RE, blood lactate (d = 2.06) and systolic blood pressure (d = 0.99) significantly increased. Improvements in executive function were similarly observed following acute RE, including faster processing speed during inhibitory control (d = 0.37) and working memory (d = 0.46), and decreased P3 latency during inhibitory control (d = 0.41). Moreover, exploratory mediation analyses revealed that systolic blood pressure mediated the effect of RE on response time during both the Flanker (−10.73 ± 4.99, 95% CI = −21.50 to −1.52) and N-back (−28.20 ± 12.38, 95% CI = −53.22 to −5.37) tasks. Overall, these findings provide evidence that acute RE enhances neuroelectric and behavioral markers of inhibitory control and working memory performance and highlight the potential for systolic pressure as a mechanism through which acute RE influences cognition.
A Single Bout of Intermittent Hypoxia Increases Cerebral Blood Flow and Supports an Executive Function Benefit
Denait Haile et al.
Psychophysiology, October 2025
Abstract:
Alternating between brief normoxic and hypoxic intervals (i.e., intermittent hypoxia: IH) increases cerebrovascular dilation, cerebral blood flow (CBF), and O2 extraction. Some work has shown that the physiological adaptations arising from multiple IH sessions improve brain health and executive function (EF) -- a finding linked to a post-intervention improvement in cortical hemodynamics. Here, we provide a first demonstration of whether the physiological changes associated with a single IH session provide a transient post-intervention EF benefit. Healthy young adults (N = 24) completed an IH protocol entailing 12 alternating 5-min normoxic (PETO2 = 100 mmHg) and hypoxic (PETO2 = 50 mmHg) intervals that were normocapnic and isocapnic, and on a separate day completed a time-matched normoxic control protocol. Prior to (T0), and immediately (T1) and 30 min (T2) following each protocol, EF was assessed via the antisaccade task. Antisaccades require a goal-directed eye movement (i.e., saccade) mirror-symmetrical to a target and provide the resolution to detect subtle EF changes. As expected, hypoxic intervals decreased arterial and cerebral tissue O2 saturation and increased CBF as estimated via near-infrared spectroscopy and transcranial Doppler ultrasound (ps < 0.001). In turn, antisaccade reaction times (RT) did not differ between T0 and T1 (p = 0.29); however, at T2 a reliable RT reduction was observed (p = 0.004). Notably, cortical hemodynamic changes during the hypoxic intervals did not correlate with the antisaccade RT benefit observed at T2 (ps > 0.17). Thus, a single bout of IH provided a transient post-intervention EF “boost” that was not linked to a unitary physiological adaptation to a reduced O2 environment.
The Power of Technical Language: Does Jargon Use Influence the Credibility of Misinformation?
Tanisha Berrios, Dun-Ya Hu & Jyotsna Vaid
Applied Cognitive Psychology, November/December 2025
Abstract:
Across two studies, we examined the impact of different levels of jargon on the perceived credibility of texts containing reliable or unreliable information. Study 1 found that higher levels of jargon were associated with increased credibility ratings for unreliable texts. Study 2 showed that higher levels of jargon were also associated with lower processing fluency for both reliable and unreliable texts. Additionally, the high level of jargon was associated with higher credibility ratings compared to text containing no jargon, and those high in conspiracy mentality were more influenced by the amount of jargon when making their credibility ratings. Jargon had a direct effect on increasing credibility, but also indirectly decreased credibility through a decrease in processing fluency. Together, these findings suggest that the presence of technical terms enhances the credibility of even unreliable text, and that individuals with conspiracy mentality leanings are more susceptible to this influence of jargon.
Sensory multi-brain stimulation enhances dyadic cooperative behavior
Ivo Leiva-Cisterna et al.
Social Cognitive and Affective Neuroscience, October 2025
Abstract:
Hyperscanning research suggests that interbrain synchronization supports the regulation of social behavior. However, the evidence is predominantly correlational, leaving a gap for epiphenomenal accounts, where synchrony merely represents concurrent stimulus processing rather than a mechanism relevant to interpersonal interactions. Here, we demonstrate that interbrain synchrony causally drives cooperative success, as evidenced by non-invasive stimulation enhancing coupling and subsequently improving performance in a concurrent interdependent cooperation task. We applied dual-sensory entrainment at 16 Hz and 40 Hz to dyads and compared their performance with non-entrained control dyads performing the same cooperation task. We found that dual stimulation improved interbrain synchrony at the targeted frequencies relative to controls, with 16 Hz entrainment producing the most prominent effect. Strikingly, sensory entrainment facilitated sustained behavioral coupling, allowing partners to maintain coordination over extended periods. Notably, these effects are contingent on improved response coordination, indicating the importance of interbrain coupling for facilitating coordination and demonstrating causally that partner neural attunement is necessary to produce effective joint behavior. Thus, our study supports the concept that interbrain synchrony represents a neural mechanism with functional specificity in social interactions.