AI & Society, Год журнала: 2023, Номер 39(5), С. 2499 - 2506
Опубликована: Июль 12, 2023
Язык: Английский
SSRN Electronic Journal, Год журнала: 2024, Номер unknown
Опубликована: Янв. 1, 2024
Artificial intelligence (AI) now matches or outperforms human in an astonishing array of games, tests, and other cognitive tasks that involve high-level reasoning thinking. Many scholars argue that—due to bias bounded rationality—humans should (or will soon) be replaced by AI situations involving cognition strategic decision making. We disagree. In this paper we first trace the historical origins idea artificial as a form computation information processing. highlight problems with analogy between computers minds input-output devices, using large language models example. Human cognition—in important instances—is better conceptualized theorizing rather than data processing, prediction, even Bayesian updating. Our argument, when it comes cognition, is AI's data-based prediction different from theory-based causal logic. introduce belief-data (a)symmetries difference use "heavier-than-air flight" example our arguments. Theories provide mechanism for identifying new evidence, way "intervening" world, experimenting, problem solving. conclude discussion implications arguments making, including role human-AI hybrids might play process.
Язык: Английский
Процитировано
9
Communications Psychology, Год журнала: 2024, Номер 2(1)
Опубликована: Июнь 3, 2024
In the present study, we investigate and compare reasoning in large language models (LLMs) humans, using a selection of cognitive psychology tools traditionally dedicated to study (bounded) rationality. We presented human participants an array pretrained LLMs new variants classical experiments, cross-compared their performances. Our results showed that most included errors akin those frequently ascribed error-prone, heuristic-based reasoning. Notwithstanding this superficial similarity, in-depth comparison between humans indicated important differences with human-like reasoning, models' limitations disappearing almost entirely more recent LLMs' releases. Moreover, show while it is possible devise strategies induce better performance, machines are not equally responsive same prompting schemes. conclude by discussing epistemological implications challenges comparing machine behavior for both artificial intelligence psychology.
Язык: Английский
Процитировано
9
Nature Communications, Год журнала: 2024, Номер 15(1)
Опубликована: Июнь 29, 2024
Abstract When processing language, the brain is thought to deploy specialized computations construct meaning from complex linguistic structures. Recently, artificial neural networks based on Transformer architecture have revolutionized field of natural language processing. Transformers integrate contextual information across words via structured circuit computations. Prior work has focused internal representations (“embeddings”) generated by these circuits. In this paper, we instead analyze directly: deconstruct into functionally-specialized “transformations” that words. Using functional MRI data acquired while participants listened naturalistic stories, first verify transformations account for considerable variance in activity cortical network. We then demonstrate emergent performed individual, “attention heads” differentially predict specific regions. These heads fall along gradients corresponding different layers and context lengths a low-dimensional space.
Язык: Английский
Процитировано
9
Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Фев. 2, 2024
The role of social media in information dissemination and agenda-setting has significantly expanded recent years. By offering real-time interactions, online platforms have become invaluable tools for studying societal responses to significant events as they unfold. However, reactions external developments are influenced by various factors, including the nature event environment. This study examines dynamics public discourse on digital shed light this issue. We analyzed over 12 million posts news articles related two events: release ChatGPT 2022 global discussions about COVID-19 vaccines 2021. Data was collected from multiple platforms, Twitter, Facebook, Instagram, Reddit, YouTube, GDELT. employed topic modeling techniques uncover distinct thematic emphases each platform, which reflect their specific features target audiences. Additionally, sentiment analysis revealed perceptions regarding topics studied. Lastly, we compared evolution engagement across unveiling unique patterns same topic. Notably, spread more rapidly due immediacy subject, while ChatGPT, despite its technological importance, propagated gradually.
Язык: Английский
Процитировано
8
Science Advances, Год журнала: 2024, Номер 10(21)
Опубликована: Май 23, 2024
Current large language models (LLMs) rely on word prediction as their backbone pretraining task. Although is an important mechanism underlying processing, human comprehension occurs at multiple levels, involving the integration of words and sentences to achieve a full understanding discourse. This study by using next sentence (NSP) task investigate mechanisms discourse-level comprehension. We show that NSP enhanced model’s alignment with brain data especially in right hemisphere demand network, highlighting contributions nonclassical regions high-level understanding. Our results also suggest can enable model better capture performance encode contextual information. demonstrates inclusion diverse learning objectives leads more human-like representations, investigating neurocognitive plausibility tasks LLMs shed light outstanding questions neuroscience.
Язык: Английский
Процитировано
8
Behavior Research Methods, Год журнала: 2024, Номер 56(6), С. 6082 - 6100
Опубликована: Янв. 23, 2024
Research on language and cognition relies extensively psycholinguistic datasets or "norms". These contain judgments of lexical properties like concreteness age acquisition, can be used to norm experimental stimuli, discover empirical relationships in the lexicon, stress-test computational models. However, collecting human at scale is both time-consuming expensive. This issue compounded for multi-dimensional norms those incorporating context. The current work asks whether large models (LLMs) leveraged augment creation large, English. I use GPT-4 collect multiple kinds semantic (e.g., word similarity, contextualized sensorimotor associations, iconicity) English words compare these against "gold standard". For each dataset, find that GPT-4's are positively correlated with judgments, some cases rivaling even exceeding average inter-annotator agreement displayed by humans. then identify several ways which LLM-generated differ from human-generated systematically. also perform "substitution analyses", demonstrate replacing a statistical model does not change sign parameter estimates (though select cases, there significant changes their magnitude). conclude discussing considerations limitations associated general, including concerns data contamination, choice LLM, external validity, construct quality. Additionally, all (over 30,000 total) made available online further analysis.
Язык: Английский
Процитировано
7
Elsevier eBooks, Год журнала: 2024, Номер unknown, С. 195 - 210
Опубликована: Янв. 1, 2024
Язык: Английский
Процитировано
6
The Science of The Total Environment, Год журнала: 2024, Номер 953, С. 175971 - 175971
Опубликована: Сен. 3, 2024
Язык: Английский
Процитировано
6
Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Май 11, 2024
Abstract Large language models (LLMs), like ChatGPT, Google’s Bard, and Anthropic’s Claude, showcase remarkable natural processing capabilities. Evaluating their proficiency in specialized domains such as neurophysiology is crucial understanding utility research, education, clinical applications. This study aims to assess compare the effectiveness of Language Models (LLMs) answering questions both English Persian (Farsi) covering a range topics cognitive levels. Twenty four (general, sensory system, motor integrative) two levels (lower-order higher-order) were posed LLMs. Physiologists scored essay-style answers on scale 0–5 points. Statistical analysis compared scores across different model, language, topic, Performing qualitative identified reasoning gaps. In general, demonstrated good performance (mean score = 3.87/5), with no significant difference between or The was strongest system 4.41) while weakest observed integrative 3.35). Detailed uncovered deficiencies reasoning, discerning priorities, knowledge integrating. offers valuable insights into LLMs’ capabilities limitations field neurophysiology. demonstrate general but face challenges advanced integration. Targeted training could address gaps causal reasoning. As LLMs evolve, rigorous domain-specific assessments will be for evaluating advancements performance.
Язык: Английский
Процитировано
5
Frontiers in Robotics and AI, Год журнала: 2024, Номер 11
Опубликована: Июль 23, 2024
Understanding the emergence of symbol systems, especially language, requires construction a computational model that reproduces both developmental learning process in everyday life and evolutionary dynamics throughout history. This study introduces collective predictive coding (CPC) hypothesis, which emphasizes models interdependence between forming internal representations through physical interactions with environment sharing utilizing meanings social semiotic within system. The total system is theorized from perspective . hypothesis draws inspiration studies grounded probabilistic generative language games, including Metropolis–Hastings naming game. Thus, playing such games among agents distributed manner can be interpreted as decentralized Bayesian inference shared by multi-agent Moreover, this explores potential link CPC free-energy principle, positing adheres to society-wide principle. Furthermore, paper provides new explanation for why large appear possess knowledge about world based on experience, even though they have neither sensory organs nor bodies. reviews past approaches offers comprehensive survey related prior studies, presents discussion CPC-based generalizations. Future challenges cross-disciplinary research avenues are highlighted.
Язык: Английский
Процитировано
5