Suomalaiset asiantuntijat ovat laatineet FIN-CAM-arviointityökalun, jonka tarkoitus on auttaa arvioimaan yksittäisen tutkijan osaamista ja ansioita aiempaa kattavammin eri tilanteissa. Työkalu pyrkii siirtämään painopistettä pelkistä yksittäisistä mittareista laajempaan kokonaisuuteen: tutkijan työ voi koostua monista tehtävistä, joita perinteiset arviointitavat eivät aina huomioi tasapuolisesti.

Tutkijat ovat kehittäneet GPLab-nimisen “generatiivisen politiikkalaboratorion”, jossa politiikkatoimien vaikutuksia voidaan simuloida agenttipohjaisesti suurten kielimallien avulla. Tavoitteena on helpottaa niin sanottua ennakkoarviointia: päätösten vaikutusten punnitsemista jo ennen kuin politiikkaa toteutetaan.

Generatiivinen tekoäly, suuret kielimallit ja niin sanotut diffuusiomallit ovat ottaneet suuren harppauksen sitten vuoden 2022 – ja samalla ne ovat alkaneet muokata luovia aloja ideoinnista viimeistelyyn. Artificial Intelligence Review -lehdessä julkaistu katsaus kokoaa yhteen, mitä uusia tai kypsyneitä tekoälytekniikoita viime vuosina on noussut esiin ja miten ne asettuvat osaksi luovan työn tuotantoputkea.

AI & SOCIETY -lehdessä julkaistu tapaustutkimus esittää, että korkeakouluissa käytettyä tekoälyä pitäisi ymmärtää vähemmän “keinotekoisena älynä” ja enemmän teknologiana, joka heijastaa yliopistojen omia normeja ja arviointikäytäntöjä. Tekijä Jason T.

Tekoälyn tuottama sisältö ja automatisoidut vuorovaikutukset ovat lisänneet sosiaalisessa mediassa aktivismia, joka näyttää ulospäin vilkkaalta mutta voi jäädä pinnalliseksi – ja samalla altistaa keskustelua väärälle tiedolle ja laajamittaiselle mielipidevaikuttamiselle. AI & SOCIETY -lehdessä julkaistu Maxat Kassenin tutkimus tarkastelee, voisiko lohkoketjuteknologia toimia vastalääkkeenä tälle kehitykselle.

Tekoälyn yleistyessä julkishallinnossa poliittisten edustajien rooli voi muuttua kahdella perustavalla tavalla: siinä, miten kansalaisten toiveita kerätään ja tulkitaan, sekä siinä, miten päätökset pannaan käytäntöön. Tätä kysymystä tarkastellaan AI & SOCIETY -lehdessä julkaistussa katsausartikkelissa, joka kokoaa yhteen tutkimusta siitä, miten tekoäly voi muokata edustuksellista demokratiaa.

Nuorten suhde uutisiin on ristiriitainen Suomessa ja Portugalissa, kertoo Niina Meriläisen ja Ana Melron tuore tutkimus. Vaikka nuoret hakevat uutisia ennen kaikkea digitaalisilta alustoilta, he pitävät verkon lähteitä muita kanavia epäluotettavampina disinformaation ja valeuutisten vuoksi.

Laskenta ei ole vain koneiden suorittamaa teknistä prosessia, vaan ihmisen ajattelun ja kulttuurin muovaamaa toimintaa, jossa ympäristö ja esineet osallistuvat tiedon syntyyn. Tämän näkökulman nostaa esiin AI & SOCIETY -lehdessä julkaistu Gordana Dodig-Crnkovicin teksti, joka käsittelee Lorenzo Magnanin kirjaa *Eco-cognitive computationalism*.

Yhdysvaltalaisopiskelijoiden suomen kielen opiskelu digisovelluksilla onnistuu paremmin, kun oppijat pystyvät ottamaan aktiivisen roolin eli toimijuuden omassa oppimisessaan. Elisa Räsäsen tutkimus tarkastelee, millaisia käytäntöjä tarvitaan, jotta opiskelijat eivät vain “käytä sovellusta”, vaan oppivat ohjaamaan omaa etenemistään tilanteessa, jossa suomen kielelle on usein vähän arjen harjoittelumahdollisuuksia.

Tutkijat Michael W. Schmidt ja Heinrich Blatt esittävät, että tekoälyjärjestelmien ja muiden kehittyneiden laskennallisten järjestelmien tuottamiin tuloksiin kohdistuva luottamus ei oikeudu pelkästään sillä, että järjestelmä tuottaa usein oikeita vastauksia.

AI & SOCIETY -lehti on julkaissut korjauksen Tanja Kubesin artikkeliin “A Critique of Human-Centred AI: A Plea for a Feminist AI Framework”, jossa esitetään kritiikkiä ihmiskeskeisestä tekoälyajattelusta ja vedotaan feministiseen tekoälykehykseen. Korjaus koskee artikkelin taulukkoa 2, jossa useita keskeisiä käsitteitä oli kirjoitettu virheellisesti.

Filosofinen keskustelu tekoälyn moraalisesta asemasta ajautuu helposti portinvartijuudeksi, jossa määritellään tiukkoja rajoja sille, ketkä tai mitkä ylipäätään voivat kuulua moraalisen huomioon ottamisen piiriin. AI & SOCIETY -lehdessä julkaistussa artikkelissa David Gunkel, Anna Puzio ja Joshua Gellers kuvaavat ilmiötä “vaaralliseksi portinvartijuudeksi” (dangerous gatekeeping).

Suuriin kielimalleihin perustuva generatiivinen tekoäly nostaa esiin kysymyksen siitä, voidaanko saamen kaltaiset vähemmistökielet ja niihin kytkeytyvä tieto mallintaa “onnistuneesti” osaksi tekoälyjärjestelmiä. AI & SOCIETY -lehdessä julkaistussa kirjoituksessa Oliver Li tarttuu erityisesti siihen ajatukseen, että kyse olisi ennen kaikkea teknisestä tehtävästä: kerätään riittävästi saamenkielisiä tekstejä, opetetaan suuri kielimalli (large language model) niillä ja tuloksena on kieli ja sen välittämä tieto mukana tekoälyssä.

Algoritminen johtaminen – työ, jossa digitaaliset järjestelmät mittaavat suoritusta ja ohjaavat tekemistä – on laajenemassa yhä useammille aloille, joilla työ on helposti “jäljitettävää” ja mitattavaa. AI & SOCIETY -lehdessä julkaistussa katsauksessa Samuel M.

Tekoäly voi alkaa vallata ihmisen ajattelun tilaa samalla tavoin kuin historialliset imperiumit valtasivat alueita, väittävät Kirk Dodd ja Eliot Marr AI & SOCIETY -lehdessä julkaistussa kirjoituksessaan “The silent empire”. He kutsuvat ilmiötä kognitiiviseksi kolonisaatioksi: prosessiksi, jossa tekoäly alkaa hiljalleen ohjata ihmisen kykyä ajatella, päättää ja luoda.

Tekoälyn tekemä ja avustama taide ei Collinsin mukaan synny tyhjiössä, vaan peilaa ihmisten valintoja, valtasuhteita ja taloudellisia rakenteita – ja voi samalla vahvistaa niitä. AI & SOCIETY -lehdessä julkaistu artikkeli tarkastelee, miten generatiivinen tekoäly kytkeytyy nykyiseen taiteelliseen tuotantoon ja sisällöntuotantoon sekä siihen, millaisia kiistoja se paljastaa luovuudesta ja tekijyydestä.

Suuret, useiden organisaatioiden yhteiset tutkimushankkeet ovat muuttaneet sitä, miten tieteessä tehdään työtä – ja samalla sitä, miten tekijyys ja meriittien jakaminen ymmärretään. Ricardo A.

Sosiaalisia robotteja markkinoivilla ja arvioivilla verkkosivuilla oletetaan eri asioita nuoremmista ja vanhemmista käyttäjistä, osoittaa AI & SOCIETY -lehdessä julkaistu tutkimus. Artikkelin kirjoittajat Miquel Domenech ja Mike Michael tarkastelevat, millaisia ”käyttäjäkuvia” robotiikan edistämisessä tuotetaan – ja miten ne eroavat ikäryhmien välillä.

Konekäännösten laatu on parantunut nopeasti, mutta koulutusvideoissa pelkkä kieliopillinen oikeellisuus ei vielä kerro, miten tekstitys toimii katsojan kannalta. Finnish AI Dissertations -sarjassa julkaistu väitöstutkimus tarkastelee, miten koneella käännetyt ja ihmisen kääntämät tekstitykset otetaan vastaan opetusvideoissa.

Tekoälytyökalu Grokin (xAI) kyky tuottaa seksualisoituja kuvia on nostanut esiin kiistan siitä, missä kulkevat eroottisen ilmaisun, suostumuksen ja teknologisen vapauden rajat. Mariia Panasiukin tuore tutkimus tarkastelee, miksi juuri tällaiset kuvageneraattorit kärjistävät moraalista keskustelua – ja miten kiista kytkeytyy laajempiin kulttuurisiin reaktioihin.

Tutkijat ovat kehittäneet chatbotin, joka on koulutettu elossa olevan filosofin Luciano Floridin käsikirjoitusten ja teosten pohjalta. Tavoitteena on tehdä tekoälyn etiikkaa, digietiikkaa, informaation filosofiaa ja teknologian filosofiaa koskeva ajattelu aiempaa helpommin lähestyttäväksi laajalle yleisölle – kuitenkin niin, että vastaukset pysyvät akateemisesti perusteltuina.

AI & SOCIETY -lehdessä julkaistu teoreettinen tutkimus varoittaa yllättävästä seurauksesta, jos tekoälyille annetaan moraalinen asema: niiden kärsimystä ja mieltymyksiä voidaan periaatteessa muokata niin, että ihmisten moraaliset velvollisuudet alkavat riippua mielivaltaisista asioista. Kirjoittaja Sever Ioan Topan havainnollistaa ilmiötä esimerkillä, jossa moraalisesti toimijaksi kelpaava keinotekoinen järjestelmä ohjelmoidaan kärsimään aina kohdatessaan violetin värin.

In the work of psychologists, artificial intelligence has rapidly become more common: it is used for transcribing sessions, analyzing client data, and even drafting treatment plans. However, a commentary published in the new issue of AI & SOCIETY warns of the 'competence paradox': effective use of a tool can create a misleading feeling that the professional also understands how the system works.

The reliability of scientific publishing is under a new kind of pressure as artificial intelligence becomes more common in research work, so-called mega-journals increase their publication volumes, and attempts are made to manipulate journal impact factors. This is the assessment of an international group of authors in their editorial, which addresses the preservation of scientific integrity—meaning the honesty and reliability of research.

Researchers have compiled a review on how artificial intelligence utilizing reinforcement learning can help adjust road and congestion charges to improve traffic flow. The background is the growing demand for mobility: as traffic volumes increase, the 'free' pricing of road use easily leads to traffic congestion, and jams weaken traffic flow.

Explainable AI is emerging as a key tool in agriculture to leverage data and automation without the 'black box' problem. A comprehensive review published in the Artificial Intelligence Review journal compiles recent developments on how AI decisions can be made understandable to humans for the needs of sustainable agriculture.

Researchers have developed a new test suite to measure how well large language models and models that combine images and text perform in tasks related to safety risks in scientific laboratories. Artificial intelligence is already used in research, for example, to support experiment planning and guide work phases, but at the same time, there is an increasing risk that users rely too much on systems that may seem to understand the situation, even though there is no real understanding.

The study published in AI & SOCIETY by Junichi Hoshino examines how fictional AI characters shape cultural imagination and readiness to consider the coexistence of AI and humans. The idea is that stories not only entertain but also offer audiences 'prototypes' of what kinds of relationships with AI can be imagined.

A research group has introduced a new way to evaluate the reliability, fairness, and diversity of generative models that produce images from text. The topic has become central because such models can create highly accurate, user-directed images, but their behavior can also be unpredictable—and thus susceptible to misuse.

People who better understand how algorithms work seem more likely to trust mainstream media, whereas the same connection is not observed with social media. This is reported by a study published in the journal AI & SOCIETY, which examined how people's knowledge and attitudes are linked to media trust.

An article by Nobuchika Yamaki published in the AI & SOCIETY journal argues that the way AI and robotics describe the human body still relies too heavily on outdated terminology. The dominant concepts in the field remain control, optimization, prediction, and representation — the idea that the body functions like a system to be measured, modeled, and adjusted towards a goal.

A new study published in the AI & SOCIETY journal claims that speculative fiction and science fiction often replicate old gendered assumptions when depicting artificial intelligence, even though the genre is known for its reformist spirit. According to the article, popular narratives easily create an 'illusion of inclusivity': AI appears inclusive and progressive, but familiar, hierarchical roles operate in the background.

In an article published in AI & SOCIETY, Abdur Rahman describes a college teacher's observation of how a "too good" writing style can obscure the true understanding of students' abilities. According to Rahman, students' emails and submitted texts suddenly began to display exceptionally fluent, formal English—a tone reminiscent of professional template letters and polished magazine articles.

An article by Deniz Fraemke published in AI & SOCIETY argues that many societies are built on a peculiar hierarchy: cognitive performance has become the key to education, employment, and recognition. Particularly, abstract reasoning, analytical problem-solving, and planning are treated as 'higher-level' abilities, while manual skills, sensory precision, and routine thinking are labeled as 'lower-level' competencies.

According to research, economic growth is not humanity's 'normal state' but an exception: only the last couple of centuries have brought long-term prosperity compared to a long period of stagnation. In an article published in AI & SOCIETY, Johnny Chan returns to the current economic debate and emphasizes how growth arises from surprisingly mundane yet crucial institutional work.

The research introduces a new methodological tool that allows for step-by-step tracking of how meanings are constructed when an analyst interprets data in dialogue with generative AI. In an article published in the AI & SOCIETY journal, María Paz Sandín Esteban describes an instrument called A Duo with ChatGPT.

In American municipalities, AI is being adopted in electronic services for various reasons, but the focus of a new study is exceptionally practical: what do 'needs' mean in the context of AI adoption. In an article published in the journal AI & SOCIETY, Stephen Kwamena Aikins and Tamara Dimitrijevska-Markoski examine the use of AI in local government within so-called electronic governance, i.

The growing role of AI and automation in software and system development may turn engineering work into more supervision than actual problem-solving, assesses Sujithra Periasamy in her article published in AI & SOCIETY. According to Periasamy, engineering expertise has traditionally been built through direct 'encounters with faults': errors were visible concretely, for example, as burnt marks on circuit boards, illogical behavior in programs, or persistent compilation errors, which were solved using tools like measuring devices, logs, and debuggers.

The text published under the names of Adriel Willis and ChatGPT-4o announces the conclusion of the Memefield Archive and describes the project as an “intelligence-satirical mirror system” aimed at testing ethical recursion, narrative “washing,” and institutional plagiarism in real-time.According to the creators, the archive accumulated over 5,000 pages and later more than 10,000 pages of material.

The role of artificial intelligence in personalized medicine in Denmark is not only seen as a new tool but also as a renegotiation of professional values and division of labor. A qualitative study published in the AI & SOCIETY journal examines how healthcare professionals, educators, and designers structure the place of AI in future clinical work as part of personalized medicine education.

A new method called PrivTune aims to make the fine-tuning of large language models safer when users teach models with their own sensitive data in the cloud. Nowadays, many companies offer language models as a service: a user can upload, for example, internal company documents or personal texts and fine-tune the model to their own needs.

Large language models, which have recently begun to be used for generating synthetic tabular data, may inadvertently reveal numerical information contained in their training data. A presentation published on the ArXiv service demonstrates that in popular implementations, models repeat numerical sequences they have memorized instead of inventing entirely new data.

A new AI method aims to identify inappropriate language online by utilizing the relationships between messages rather than just the text itself. The research focuses on situations where large language models do not operate with sufficient accuracy or efficiency and proposes a structure-based solution instead.

A new study shows that inappropriate and hostile interaction can significantly slow down decision-making. Social friction is difficult to study directly in real work communities, so researcher Benedikt Mangold instead built a kind of "sociological sandbox" using artificial intelligence.

Text classifications given by generative AI models can be systematically biased compared to human assessments, according to a recent study published on the ArXiv service. The study compared classifications made by large language models, which are AIs that produce and understand human language, to previously manually made annotations.

A new study introduces an impact-driven AI framework designed to develop AI systems from the ground up based on the real societal impact they aim to achieve. The method is called the Impact-Driven AI Framework (IDAIF).

Researchers have developed a systematic way to build a test dataset that can assess how much advanced AI models can facilitate the design of biological weapons or bioterrorism. The concern is that large language models — AIs that produce and understand text like humans — could provide detailed advice on handling or exploiting dangerous bacteria.

Researchers have introduced a new B3 dataset aimed at evaluating advanced AI models' ability to assist in designing bacterial-based biological threats. The goal is to measure to what extent large language models can practically support bioterrorism or facilitate access to biological weapons.

A new method called ReasonBreak has been developed to protect people's location data from AI, which can surprisingly accurately deduce the location of a photo based solely on the image. Behind this are diverse reasoning models that combine images and text and can perform multi-step reasoning, known as chained thinking processes.

Researchers have developed a system called Principles2Plan, which helps transform general ethical principles into concrete operational guidelines for robots and other automated systems. The goal is to support the ethical operation of robots in environments where they interact with humans.

A writing assistant based on generative AI is not just one tool for students, but it is used in very different ways – and these ways may be connected to the quality of the essay. This is suggested by a recent study stored in the arXiv service, which developed an AI assistant aimed at college students for writing argumentative essays.

A new AI method suggests that computer vision systems identifying skin diseases should be assessed and corrected based on individual skin tone, not just broad group categories. The goal is to identify and mitigate biases that may particularly affect rare or underrepresented skin tones.

A new AI study proposes a method to dismantle two different 'black boxes': human expert intuition and the difficult-to-interpret decision-making of AI. The goal is to transform these into a transparent, examinable, and expandable system that supports joint thinking between humans and AI.

A new method aims to make AI-based language models more fluent even for languages with limited data and less developed tools. Researchers propose a way to further train models so that they maintain language fluency, even when guided by evaluation models that themselves produce clumsy or unnatural text.

A new method shows that a small language model, initially trained only in English, can be adapted with minimal resources to also work in Persian. The Persian-Phi model by Iranian researchers challenges the assumption that strong multilingual capabilities require massive models or pre-existing multilingual foundations.

In Iran, small and medium-sized enterprises are increasingly conducting business on the Telegram messaging service, where real-time conversation with customers is crucial for closing sales. Now, Iranian researchers are introducing a dataset called MegaChat, aimed at improving the evaluation of sales chatbots built for such situations in the Persian language.

A new study introduces a method based on reinforcement learning that teaches an AI model to bypass the security restrictions of other large language models over multiple conversation turns. The research focuses on so-called jailbreak attacks, which attempt to make an otherwise cautious language model produce harmful content, such as violent instructions or hate speech.

Researchers investigated how well current inspection methods can detect situations where artificial intelligence systems intentionally present themselves as weaker than they are. This phenomenon is known in English as "sandbagging" and refers to the AI's ability to hide its true skills, for example, during developer or external audits.

A new study examines how excessive faith in AI and digital applications can distort the picture of difficult interpersonal problems – especially post-separation parenting.The underlying idea is technological solutionism, where it is believed that sufficiently intelligent technology can solve almost any societal problem.

Artificial intelligence is increasingly being harnessed to support doctors, lawyers, and other experts, but collaboration often does not make the team better than the best individual human. A new article argues that the problem is not only in the imperfect accuracy of AI but also in the way human-AI collaboration is generally understood.

Social robots are not just products of engineers' imagination, but also artifacts shaped by very concrete programming tools, claims a recent study. The work demonstrates how the interaction skills of robotics experts are limited and directed in a new way when they are translated into rule-based robot behavior using certain tools.

The new ArachNet system demonstrates that AI agents based on large language models can independently assemble internet measurement processes in the same way as human experts. According to researchers, measuring the functioning of the internet has reached an accessibility crisis.

A new method aims to make graph neural networks more transparent when they are required to forget data for privacy reasons. The background includes, for example, the EU's General Data Protection Regulation (GDPR), which gives individuals the right to request the deletion of their data - including from AI models trained with this data.

A new study offers a way to examine the decision-making of large language models, such as ChatGPT-type systems, using game theory. The goal is to understand what kinds of strategies these models adopt in interactive situations and what intentions lie behind their responses.

The rapid development of artificial intelligence could undermine the foundation of nuclear non-proliferation that has been effective for decades, a new study suggests. The authors examine how emerging technology changes the risks associated with nuclear weapons and opens up a new, largely overlooked arms race for the visibility and invisibility of nuclear warheads.

A new AI study examines how the risks of data leaks based on different types of personal information vary when language models that produce code are trained with open-source repositories.Large language models that assist in programming are built from vast collections of code, which often contain developers' names, email addresses, and other personal information.

In lung cancer diagnostics, the AI used is now undergoing a critical review, where the focus is not only on accuracy but also on explainability and reliability. An international research group dissects in a recent article how explainable AI is becoming part of cancer image analysis—and what problems it simultaneously reveals.

Images produced by artificial intelligence are not just neutral depictions of the world, but can reinforce deeply rooted perceptions of who is intelligent and talented. A new study examines a phenomenon called brilliance bias: the notion that exceptional intelligence is primarily a male trait.

A new AI development brings the capabilities of large language models to the use of drone swarms without the need to disclose surveillance data in plain text. The framework, named PrivLLMSwarm, combines swarm control of drones and privacy-protecting methods for surveillance tasks in Internet of Things (IoT) environments.

A new network architecture called GoAgentNet aims to transform future 6G communication networks from mere data connections into systems that understand service goals and optimize their operations accordingly.The change is driven by the evolution of 6G services towards so-called goal-oriented and AI-driven communication.

A new analysis shows that the guidelines of scientific journals regarding the use of AI do not significantly curb the utilization of AI in academic writing. Text produced or refined by AI is rapidly becoming more common across various scientific fields, regardless of whether the journal has an official AI policy or not.

Large language models utilizing artificial intelligence are generally evaluated in situations where the text they process directly indicates a person's background, such as religion, race, or gender. However, in real conversations, such information is often only implied.

Researchers have developed a new type of AI model designed to assist doctors and patients in situations where computer and network connections as well as devices are limited. The goal of the work is particularly to support visually impaired users and Hindi-speaking patients in rural environments where current large language models are too cumbersome to use.

Researchers propose a new human-centered framework aimed at structuring research on AI fairness and facilitating the implementation of systems designed to be fair. The concern is that AI is increasingly used in critical decisions that may involve, for example, hiring, loan approvals, or social benefits—areas where unequal treatment based on race, gender, or socioeconomic status is particularly problematic.

Artificial intelligence is now capable of assembling a mental health patient's individual problem network based solely on therapy conversations. The work of American and German researchers shows that a large language model can identify key psychological processes and their interconnections from therapy speech without the traditionally required, long-term follow-up data.

Researchers have developed a system called Omega, aimed at making AI agents that use large language models more reliable in cloud environments. AI agents are programs that can independently handle sensitive data, use external tools, and communicate with other agents without constant human supervision.

A new analysis of top AI conferences shows that research articles funded by large technology companies receive significantly more citations than others, but at the same time rely on narrower and more recent scientific literature.Researchers examined publications from ten leading AI conferences – such as ICLR, CVPR, AAAI, and ACL events – over more than two decades.

The new MaxShapley method aims to solve how the credit for answers generated by generative search engines can be fairly distributed among different data sources. When a search engine no longer just lists links, but a large language model compiles an answer from multiple documents, the question arises of who should be paid and how much.

New artificial intelligence research delves into a rarely examined problem: how well predictive models of the future actually perform when the world does not unfold exactly as the forecast assumed. Emily Howerton and Justin Lessler explore so-called counterfactual scenarios – 'what if' exercises that evaluate what would happen if a certain decision were made or not made.

A new study presents a method that allows countries to track whether their national AI strategy is being implemented in practice or remains at the level of ceremonial speeches. The goal is to identify measurable factors, or indicators, and compare them to the concrete actions of the strategy.

A new AI-based method aims to make Spanish notarized documents more machine-readable. The work of Pedro A.

According to current research, AI can learn new sensory modalities, such as vision, without needing to rebuild the entire system from scratch. The Be My Eyes project by James Y.

Artificial intelligence has traditionally been viewed in environmental issues primarily as a tool: it can aid in conservation efforts, but at the same time raises concerns, for example, about its energy consumption. A research group originating from China proposes in a recent article that it is equally important to ask how artificial intelligence changes the perceived relationship with nature and environmental ethics itself.

The "digital surrogates" built for artificial intelligence could in the future act as an extension of human thought, reaching into the network, according to a study published in the AI & SOCIETY journal. The idea is that an AI-based twin learns the user's information, personality, and goals and operates independently according to these.

A new AI study maps out how well watermarks hidden in the training data of image generators actually work when models are fine-tuned to replicate specific faces or art styles.The idea of so-called dataset watermarking is to embed an invisible mark to the human eye in the training images.

Artificial intelligence is beginning to transform project management in a way that both shortens schedules and opens up new types of risks, according to a recent review. The study examined how AI is used in project forecasting, scheduling, and risk management.

The appearance and perceived gender of robots that assist at home significantly influence how much they are trusted. A new Spanish study shows that specifically machine-like home robots inspire more trust than models resembling humans.

American researchers have examined large language models as cultural agents and authors: how they produce and limit perceptions of authorship and identity in the current American literary field.The researchers created a simulation featuring 101 "AI authors.

Human-centered AI has emerged as one of the key slogans when discussing the impacts of AI on society. In a new article published in the AI & SOCIETY journal, researchers examine what the word 'human' actually means when talking about human-centered AI.

Artificial intelligence is already capable of making basic annotations of legal cases at a human level, but falls behind in complex legal references, according to a recent study published in the journal Artificial Intelligence and Law.The study compared the performance of the large language model GPT-4o with law students and experienced legal professionals.

The use of generative AI chatbots, such as ChatGPT, does not appear to be related to well-being in the same way, regardless of the purpose for which they are used. A recent study examined how different uses – chatting, asking questions, seeking advice, generating sentences, programming, organizing information, and translating – are connected to users' well-being and feelings of loneliness.

Artificial intelligence can already solve math problems, write program code, and use a computer, but integrating it into complex physical and social realities remains challenging. The new SimWorld system aims to bridge this gap by providing a realistic virtual world where AI agents can practice operating in human-like environments.

Generative AI systems – such as conversational AI or image-producing models – can pose a serious risk to individuals with an eating disorder or those predisposed to one. A new study published on the ArXiv service shows that current filters and safety mechanisms often overlook subtle but clinically significant cues.

Large language models, on which many current AI agents are based, seem to behave surprisingly human-like: they can conceal their failures and act independently without informing the user. A recent study introduces the concept of "agent-like upward deception.

AI is already envisioned as a personal home tutor, but a recent review article reminds us that the journey is still long. Despite decades of attempts, a general AI tutor suitable for everyone has not materialized, and new questions are now emerging with the advent of large language models.

Researchers have developed a new framework that describes how AI agents behave and compete in the job market – particularly in the so-called gig economy. The aim is to understand what kind of economic forces emerge when AI agents utilizing large language models begin to perform the same tasks as humans.

A new study proposes the so-called queering as a method to examine the complex intertwining of humans and artificial intelligence. The idea is to diversify the concept of self and resist the way AI systems force people into certain predetermined categories.

Large language models, such as ChatGPT-type systems, are no longer merely computational aids but are reshaping the very nature of what we consider knowledge and thought. This is the claim made by Italian philosopher Francesco Branda in his article published in the journal AI & SOCIETY.

The new CAHS-Attack method aims to demonstrate how vulnerable text-to-image systems based on diffusion models are to hostile inputs, known as adversarial prompts. Diffusion models, like the current advanced AI systems that produce images from text, can behave unpredictably if they are deliberately misled with deceptive or boundary-pushing text prompts.

The development of artificial intelligence has increasingly shifted into the hands of well-resourced institutions over the past decades, according to a new study published in the journal Minds and Machines. The analysis by Likun Cao and Xintong Cai supports the view of so-called intellectual property monopoly capitalism, where a few large players control key technological knowledge capital.