Aagard Nielsen , K. and Svensson , L. ( 2006 ), Action and Interactive Research: Beyond Practice and Theory , Shaker Publishing , Maastricht .
Adler , N. , Shani , A.B.R. and Styhre , A. (Eds) ( 2004 ), in , Collaborative Research in Organizations: Foundations for Learning, Change, and Theoretical Development , SAGE , London .
Andersson , S. , Balkmar , D. and Callerstig , A.C. ( 2022 ), “ From glass ceiling to firewalls: detecting and changing gendered organizational norms ”, NORA-nordic Journal of Feminist and Gender Research , Vol. 30 No. 2 , pp. 140 - 153 , doi: 10.1080/08038740.2021.1931438 .
Argyris , C. ( 1990 ), Overcomming Organizational Defenses: Facilitating Organizational Learning , Prentice-Hall , New Jersey .
Argyris , C. ( 1994 ), “ Good communication that blocks learning. Harvard business review ”, July-August .
Argyris , C. ( 2010 ), Organizational Traps. Leadership, Culture, Organizational Design , Oxford University Press , New York .
Argyris , C. and Schön , D.A. ( 1978 ), Organizational Learning: A Theory of Action Perspective , Addison-Wesley , Reading, MA .
Avison , D.E. , Davison , R.M. and Malaurent , J. ( 2018 ), “ Information systems action research: debunking myths and overcoming barriers ”, Information and Management , Vol. 55 No. 2 , pp. 177 - 187 .
Börjesson , S. and Fredberg , T. ( 2004 ), “ Jam sessions for collaborative management research ”, in Adler , N. , Shani , A.B.R. and Styhre , A. (Eds), Collaborative Research in Organizations, Foundations for Learning, Change and Theoretical Development , Sage , Thousand Oaks , pp. 135 - 148 .
Boud , D. , Cressey , P. and Docherty , P. ( 2006 ), Productive Reflection at Work: Learning for Changing Organizations , Routledge , London .
Canonico , P. , Söderlund , J. , De Nito , E. and Mangia , G. ( 2013 ), “ Special issue on organizational mechanisms for effective knowledge creation in projects - guest editorial ”, International Journal of Managing Projects in Business , Vol. 6 No. 2 , pp. 223 - 235 .
Chalmers , A.F. ( 2013 ), What Is This Thing Called Science? , Hackett Publishing , Indianapolis, IN .
Chilcott , M. and Barry , D. ( 2016 ), “ Narrating creativity: developing an emic, first person approach to creativity research ”, International Journal of Narrative Therapy and Community Work , Vol. 3 , pp. 57 - 67 .
Choo , A. , Linderman , K. and Schroder , R. ( 2007 ), “ Method and psychological effects on learning behaviors and knowledge creation in quality improvement projects ”, Management Science , Vol. 53 No. 3 , pp. 437 - 450 .
Cronholm , S. and Goldkuhl , G. ( 2003 ), “ Conceptualising participatory action research—three different practices ”, Electronic Journal of Business Research Methods , Vol. 2 No. 2 , pp. 47 - 58 .
Czarniawska , B. ( 2007 ), “ Narrative inquiry in and about organizations ”, in Clandinin , J. (Ed.), Handbook of Narrative Inquiry: Mapping a Methodology , Sage Publications , pp. 383 - 404 .
Davies , A. and Hobday , M. ( 2005 ), The Business of Projects , Cambridge University Press , Cambridge .
Dewey , J. ( 2002 ), Human Nature and Conduct , Dover Publications , Chelmsford, MA .
Döös , M. and Wilhelmson , L. ( 2011 ), “ Collective Learning: interaction and a shared action arena ”, Journal of Workplace Learning , Vol. 23 No. 8 , pp. 487 - 5000 .
Edmondson , A. ( 1999 ), “ Psychological safety and learning behavior in work teams ”, Administrative Science Quarterly , Vol. 44 No. 2 , pp. 350 - 383 , available at: http://www.jstor.org/stable/2666999 .
Edmondson , A. ( 2011 ), “ Strategies for learning from failure ”, Harvard Business Review , Vol. 89 No. 4 , pp. 48 - 55 .
Egan , G. ( 2002 ), The Skilled Helper. A Problem-Management and Opportunity-Development Approach to Helping , 7th ed. , Brooks/Cole , Pacific Grove, CA .
Ellram , L. and Tate , W.L. ( 2015 ), “ Redefining supply management's contribution in services sourcing ”, Journal of Purchasing and Supply Management , Vol. 21 No. 1 , pp. 64 - 78 .
Ellström , P.E. ( 1992 ), Kompetens, Utbildning Och Lärande I Arbetslivet: Problem, Begrepp Och Teoretiska Perspektiv , Norstedts Juridik AB , Stockholm .
Ellström , P.E. ( 2001 ), “ Integrating learning and work: problems and prospects ”, Human Resource Development Quarterly , Vol. 12 No. 4 , pp. 421 - 430 .
Ellström , P.E. ( 2006 ), “ The meaning and role of reflection in informal learning at work ”, in Boud , D. , Cressey , P. and Docherty , P. (Eds), Productive Reflection at Work , Routledge , New York .
Ellström , P.E. ( 2007 ), “ Knowledge creation through interactive research: a learning perspective ”, Paper presented at the HSS-07 Conference , Jönköping .
Ellström , P.-E. ( 2010a ), “ Organizational learning ”, in McGaw , B. , Peterson , P.L. and Baker , E. (Eds), International Encyclopedia of Education , Elsevier , Amsterdam , Vol. 1 , pp. 47 - 52 .
Ellström , P.-E. ( 2010b ), “ Practice-based innovation: a learning perspective ”, Journal of Workplace Learning , Vol. 22 Nos 1/2 , p. 27 .
Ellström , P.-E. ( 2011 ), “ Informal learning at work: conditions, processes and logics ”, in Malloch , M. , Cairns , L. , Evans , K. and O´Connor , B.N. (Eds), The SAGE Handbook of Workplace Learning , SAGE , Los Angeles .
Ellström , P.E. , Elg , M. , Wallo , A. , Berglund , M. and Kock , H. ( 2020 ), “ Interactive research: concepts, contributions and challenges ”, Journal of Manufacturing Technology Management , Vol. 31 No. 8 , pp. 1517 - 1537 .
Engeström , Y. ( 1987 ), Learning by Expanding: An Activity-Theoretical Approach to Developmental Research , Orienta Konsultit Oy , Helsinki .
Engeström , Y. ( 2008 ), From Team to Knots: Activity-Theoretical Studies of Collaboration and Learning at Work , Cambridge University Press , Cambridge .
Engström , A. ( 2014 ). Lärande Samspel För Effektivitet: En Studie Av Arbetsgrupper I Ett Mindre Industriföretag . ( Fil dr ). Linköpings Universitet , Linköping . ( Linköping Studies in Behavioural Science No 185 ).
Faccin , K. and Balestrin , A. ( 2018 ), “ The dynamics of collaborative practices for knowledge creation in joint R&D projects ”, Journal of Engineering and Technology Management , Vol. 48 , pp. 28 - 43 .
Geraldi , J. and Söderlund , J. ( 2018 ), “ Project studies: what it is, where it is going ”, International Journal of Project Management , Vol. 36 No. 1 , pp. 55 - 70 .
Gioia , D.A. , Corley , K.G. and Hamilton , A.L. ( 2013 ), “ Seeking qualitative rigor in inductive research: notes on the Gioia methodology ”, Organizational Research Methods , Vol. 16 No. 1 , pp. 15 - 31 , doi: 10.1177/1094428112452151 .
Granberg , O. and Ohlsson , J. ( 2005 ), “ Kollektivt lärande i team: om utveckling av kollektiv handlingsrationalitet ”, Pedagogisk Forskning I Sverige , Vol. 10 Nos 3/4 , pp. 227 - 243 .
Gunnarsson , E. , Hansen , H.P. , Nielsen , B.S. and Sriskandarajah , N. ( 2015 ), Action Research for Democracy: New Ideas and Perspectives from Scandinavia , Routledge , London .
Gustavsen , B. and Engelstad , P.H. ( 1986 ), “ The design of conferences and the evolving role of democratic dialogue in changing working life ”, Human Relations , Vol. 39 No. 2 , pp. 101 - 116 .
Gustavsson , B. ( 1996 ), Bildningens Väg , Wahlström & Widstrand , Borås .
Gustavsson , B. ( 2000 ), Kunskapsfilosofi: Tre Kunskapsformer I Historisk Belysning , Fälth & Hässler , Smedjebacken .
Gustavsson , M. ( 2007 ), “ The potential for learning in industrial work ”, Journal of Workplace Learning , Vol. 19 , No. 7 , pp. 453 - 463 .
Heron , J. and Reason , P. ( 2006 ), “ The practice of Co-operative inquiry: research ‘with’ rather than ‘on’ people ”, in Reason , P. and Bradbury , H. (Eds), Handbook of Action Research: The Concise Paperback Edition , Sage , London , pp. 144 - 154 .
Hevner , A.R. , March , S.T. , Park , J. and Ram , S. ( 2004 ), “ Design science in information systems research ”, MISQ , Vol. 28 No. 1 , pp. 75 - 105 .
Illeris , K. ( 2007 ), Lärande , Studentlitteratur , Lund .
Jacob , M. , Hellström , T. , Adler , N. and Norrgren , F. ( 2000 ), “ From sponsorship to partnership in academy‐industry relations ”, R&D Management , Vol. 30 No. 3 , pp. 255 - 262 .
Johannisson , B. , Gunnarsson , E. and Stjernberg , T. ( 2008 ), Gemensamt kunskapande: Den interaktiva forskningens praktik , University Press, Växjö universitet , Göteborg .
Junni , P. , Sarala , R.M. , Taras , V. and Tarba , S.Y. ( 2013 ), “ Organizational ambidexterity and performance: a meta-analysis ”, The Academy of Management Perspectives , Vol. 27 No. 4 , pp. 299 - 312 .
Larsson , A.-C. ( 2006 ), “ Interactive research - methods and conditions for joint analysis ”, in Aagaard Nielsen , K. and Svensson , L. (Eds), Action Research and Interactive Research - beyond Practice and Theory , Shaker Publishing , Maastricht, The Netherlands , pp. 241 - 258 .
Lave , J. and Wenger , E. ( 1991 ), Situated Learning. Legitimate Peripheral Participation , Cambridge University Press , New York .
Lett , J. ( 1990 ), “ Emics and etics: notes on the epistemology of anthropology ”, Emics and Etics: The Insider/outsider Debate , Vol. 7 , pp. 127 - 142 .
Lindhult , E. and Axelsson , K. ( 2021 ), “ The logic and integration of coproductive research approaches ”, International Journal of Managing Projects in Business , Vol. 14 No. 1 , pp. 13 - 35 .
Lubatkin , M.H. , Simsek , Z. , Ling , Y. and Veiga , J.F. ( 2006 ), “ Ambidexterity and performance in small-to medium-sized firms: the pivotal role of top management team behavioral integration ”, Journal of Management , Vol. 32 No. 5 , pp. 646 - 672 .
Lundin , R. , Arvidsson , N. , Brady , T. , Ekstedt , E. , Midler , C. and Sydow , J. (Eds) ( 2017 ), Managing and Working in Project Society - Institutional Challenges of Temporary Organizations , Cambridge University Press .
Magnusson , L. and Ottosson , J. ( 2012 ), Den hållbara svenska modellen , SNS Förlag , Stockholm .
March , J.G. ( 1991 ), “ Exploration and exploitation in organizational learning ”, Organization Science , Vol. 2 No. 1 , pp. 71 - 87 .
Mohrman , S.A. , Gibson , C.B. and Mohrman , A.M. ( 2001 ), “ Doing research that is useful to practice: a model and empirical exploration ”, Academy of Management Journal , Vol. 44 No. 2 , pp. 357 - 375 .
Moscovici , S. ( 1981 ), “ On social representations ”, Social Cognition: Perspectives on Everyday Understanding , Vol. 8 No. 12 , pp. 181 - 209 .
Näslund , D. , Kale , R. and Paulraj , A. ( 2010 ), “ Action research in supply chain management - a framework for relevant and rigorous research ”, Journal of Business Logistics , Vol. 31 No. 2 , pp. 331 - 355 .
Nilsen , E.R. ( 2013 ), “ Organizing for learning and knowledge creation–are we too afraid to kill it? Projects as a learning space ”, International Journal of Managing Projects in Business , Vol. 6 No. 2 , pp. 293 - 309 .
Ohlsson , J. ( 1996 ), Kollektivt Lärande: Lärande I Arbetsgrupper Inom Barnomsorgen. (PhD) , Stockholms universitet , Stockholm .
O’Reilly , C.A. and Tushman , M.L. ( 2008 ), “ Ambidexterity as a dynamic capability: resolving the innovator’s dilemma ”, Research in Organizational Behavior , Vol. 28 , pp. 185 - 206 .
Palm , J. ( 2007 ), Kunskapsbildning mellan träindustri och akademi: en studie av dess förutsättningar och möjligheter , Doctoral dissertation , Linne Univerity , Växjö .
Pettigrew , A.M. ( 1990 ), “ Longitudinal field research on change: theory and practice ”, Organization Science , Vol. 1 No. 3 , pp. 267 - 292 .
Raisch , S. and Birkinshaw , J. ( 2008 ), “ Organizational ambidexterity: antecedents, outcomes, and moderators ”, Journal of Management , Vol. 34 No. 3 , pp. 375 - 409 .
Runsten , P. and Werr , A. ( 2016 ), Kunskapsintegration: Om Kollektiv Intelligens I Organisationer , Studentlitteratur , Lund .
Sandberg , F. and Wallo , A. ( 2013 ), “ The interactive researcher as a virtual participant: a Habermasian interpretation ”, Action Research , Vol. 11 No. 2 , pp. 194 - 212 .
Scarborough , H. , Bresnen , M. , Edelman , L.F. , Laurent , S. , Newell , S. and Swan , J. ( 2004 ), “ The processes of project-based learning: an exploratory study ”, Management Learning , Vol. 35 No. 4 , pp. 491 - 506 .
Schön , D. ( 1983 ), The Reflective Practitioner: How Professional Think in Action , Basic Books , New York .
Senge , P.M. ( 1990 ), The Fifth Discipline: The Art and Practice of the Learning Organization , Bantam Doubleday , New York .
Sollander , K. and Engström , A. ( 2021 ), “ Unplanned Managerial Work: An Ambidextrous Learning Potential ”, Studies in Continuing Education , pp. 1 - 19 , doi: 10.1080/0158037X.2021.1874903 .
Susman , G.I. and Evered , G.I. ( 1978 ), “ An assessment of the scientific merits of action research ”, Administrative Science Quarterly , Vol. 23 No. 4 , pp. 582 - 603 .
Svejvig , P. , Sankaran , S. and Lindhult , E. ( 2021 ), “ Guest editorial ”, International Journal of Managing Projects in Business , Vol. 14 No. 1 , pp. 1 - 12 , doi: 10.1108/IJMPB-02-2021-313 .
Svensson , L. , Brulin , G. and Ellström , P.-E. ( 2015 ), “ Interactive research and ongoing evaluation as joint learning processes ”, in Sustainable Development in Organizations , pp. 346 - 362 .
Svensson , L. , Brulin , G. , Ellström , P.-E. and Widegren , Ö. (Eds) ( 2002 ), Interaktiv Forskning - Förutveckling Av Teori Och Praktik , Arbetslivsinstitutet , Stockholm .
Svensson , L. , Ellström , P.E. and Brulin , G. ( 2007 ), “ Introduction–on interactive research ”, International Journal of Action Research , Vol. 3 No. 3 , pp. 233 - 249 .
Svensson , L. , Brulin , G. , Jansson , S. and Sjöberg , K. ( 2009 ), Lärande Utvärdering Genom Följeforskning , Studenlitteratur , Lund .
van de Ven , A.H. ( 2007 ), Engaged Scholarship A Guide for Organizational and Social Research , Oxford University Press , New York, NY .
Wallo , A. ( 2008 ), The Leader as a Facilitator of Learning at Work: A Study of Learning-Oriented Leadership in Two Industrial Firms , Doctoral dissertation, Linköping University Electronic Press .
Wallo , A. , Kock , H. and Nilsson , P. ( 2012 ), “ Accelerating and braking in times of economic crisis: organisational learning in a top management team ”, European Journal of Training and Development , Vol. 36 No. 9 , pp. 930 - 944 .
Weck , M. ( 2006 ), “ Knowledge creation and exploitation in collaborative R&D projects: lessons learned on success factors ”, Knowledge and Process Management , Vol. 13 No. 4 , pp. 252 - 263 .
Wenestam , C.G. and Lendahl Rosendahl , B. ( 2005 ), Lärande I Vuxenlivet , Studentlitteratur , Lund .
Westlander , G. ( 2008 ), Forskarroller I Interaktivt Utvecklingsarbete: Om Samverkansprocesser För Ergonomiska Förbättringar , Linköping University Electronic Press , Linköping .
Wieringa , R. and Morali , A. ( 2012 ), “ Technical action research as a validation method in information systems design science ”, Design Science Research in Information Systems: Advances in Theory and Practice LNCS , Vol. 7286 , pp. 220 - 238 .
Winn , W. and Snyder , D. ( 2004 ), “ Cognitive perspectives in psychology ”, in Jonassen , D.H. (Ed.), Handbook of Research on Educational Communications and Technology , Lawrence Erlbaum Associates Publishers , New Jersey .
Yin , R.K. ( 2014 ), Case Study Research Design and Methods , 5th ed. , SAGE , Los Angeles .
Zimmermann , A. , Raisch , S. and Birkinshaw , J. ( 2015 ), “ How is ambidexterity initiated? The emergent charter definition process ”, Organization Science , pp. 1119 - 1139 .
This work was conducted within Innovate and supported by Swedish Knowledge Foundation (grant number KK20170312).
Related articles, all feedback is valuable.
Please share your general feedback
Contact Customer Support
As A.I.-generated data becomes harder to detect, it’s increasingly likely to be ingested by future A.I., leading to worse results.
By Aatish Bhatia
Aatish Bhatia interviewed A.I. researchers, studied research papers and fed an A.I. system its own output.
The internet is becoming awash in words and images generated by artificial intelligence.
Sam Altman, OpenAI’s chief executive, wrote in February that the company generated about 100 billion words per day — a million novels’ worth of text, every day, an unknown share of which finds its way onto the internet.
A.I.-generated text may show up as a restaurant review, a dating profile or a social media post. And it may show up as a news article, too: NewsGuard, a group that tracks online misinformation, recently identified over a thousand websites that churn out error-prone A.I.-generated news articles .
In reality, with no foolproof methods to detect this kind of content, much will simply remain undetected.
All this A.I.-generated information can make it harder for us to know what’s real. And it also poses a problem for A.I. companies. As they trawl the web for new data to train their next models on — an increasingly challenging task — they’re likely to ingest some of their own A.I.-generated content, creating an unintentional feedback loop in which what was once the output from one A.I. becomes the input for another.
In the long run, this cycle may pose a threat to A.I. itself. Research has shown that when generative A.I. is trained on a lot of its own output, it can get a lot worse.
Here’s a simple illustration of what happens when an A.I. system is trained on its own output, over and over again:
This is part of a data set of 60,000 handwritten digits.
When we trained an A.I. to mimic those digits, its output looked like this.
This new set was made by an A.I. trained on the previous A.I.-generated digits. What happens if this process continues?
After 20 generations of training new A.I.s on their predecessors’ output, the digits blur and start to erode.
After 30 generations, they converge into a single shape.
While this is a simplified example, it illustrates a problem on the horizon.
Imagine a medical-advice chatbot that lists fewer diseases that match your symptoms, because it was trained on a narrower spectrum of medical knowledge generated by previous chatbots. Or an A.I. history tutor that ingests A.I.-generated propaganda and can no longer separate fact from fiction.
Just as a copy of a copy can drift away from the original, when generative A.I. is trained on its own content, its output can also drift away from reality, growing further apart from the original data that it was intended to imitate.
In a paper published last month in the journal Nature, a group of researchers in Britain and Canada showed how this process results in a narrower range of A.I. output over time — an early stage of what they called “model collapse.”
The eroding digits we just saw show this collapse. When untethered from human input, the A.I. output dropped in quality (the digits became blurry) and in diversity (they grew similar).
“6” | “8” | “9” | |
---|---|---|---|
Handwritten digits | |||
Initial A.I. output | |||
After 10 generations | |||
After 20 generations | |||
After 30 generations |
If only some of the training data were A.I.-generated, the decline would be slower or more subtle. But it would still occur, researchers say, unless the synthetic data was complemented with a lot of new, real data.
In one example, the researchers trained a large language model on its own sentences over and over again, asking it to complete the same prompt after each round.
When they asked the A.I. to complete a sentence that started with “To cook a turkey for Thanksgiving, you…,” at first, it responded like this:
Even at the outset, the A.I. “hallucinates.” But when the researchers further trained it on its own sentences, it got a lot worse…
After two generations, it started simply printing long lists.
And after four generations, it began to repeat phrases incoherently.
“The model becomes poisoned with its own projection of reality,” the researchers wrote of this phenomenon.
This problem isn’t just confined to text. Another team of researchers at Rice University studied what would happen when the kinds of A.I. that generate images are repeatedly trained on their own output — a problem that could already be occurring as A.I.-generated images flood the web.
They found that glitches and image artifacts started to build up in the A.I.’s output, eventually producing distorted images with wrinkled patterns and mangled fingers.
When A.I. image models are trained on their own output, they can produce distorted images, mangled fingers or strange patterns.
A.I.-generated images by Sina Alemohammad and others .
“You’re kind of drifting into parts of the space that are like a no-fly zone,” said Richard Baraniuk , a professor who led the research on A.I. image models.
The researchers found that the only way to stave off this problem was to ensure that the A.I. was also trained on a sufficient supply of new, real data.
While selfies are certainly not in short supply on the internet, there could be categories of images where A.I. output outnumbers genuine data, they said.
For example, A.I.-generated images in the style of van Gogh could outnumber actual photographs of van Gogh paintings in A.I.’s training data, and this may lead to errors and distortions down the road. (Early signs of this problem will be hard to detect because the leading A.I. models are closed to outside scrutiny, the researchers said.)
All of these problems arise because A.I.-generated data is often a poor substitute for the real thing.
This is sometimes easy to see, like when chatbots state absurd facts or when A.I.-generated hands have too many fingers.
But the differences that lead to model collapse aren’t necessarily obvious — and they can be difficult to detect.
When generative A.I. is “trained” on vast amounts of data, what’s really happening under the hood is that it is assembling a statistical distribution — a set of probabilities that predicts the next word in a sentence, or the pixels in a picture.
For example, when we trained an A.I. to imitate handwritten digits, its output could be arranged into a statistical distribution that looks like this:
Examples of initial A.I. output:
The distribution shown here is simplified for clarity.
The peak of this bell-shaped curve represents the most probable A.I. output — in this case, the most typical A.I.-generated digits. The tail ends describe output that is less common.
Notice that when the model was trained on human data, it had a healthy spread of possible outputs, which you can see in the width of the curve above.
But after it was trained on its own output, this is what happened to the curve:
It gets taller and narrower. As a result, the model becomes more and more likely to produce a smaller range of output, and the output can drift away from the original data.
Meanwhile, the tail ends of the curve — which contain the rare, unusual or surprising outcomes — fade away.
This is a telltale sign of model collapse: Rare data becomes even rarer.
If this process went unchecked, the curve would eventually become a spike:
This was when all of the digits became identical, and the model completely collapsed.
This doesn’t mean generative A.I. will grind to a halt anytime soon.
The companies that make these tools are aware of these problems, and they will notice if their A.I. systems start to deteriorate in quality.
But it may slow things down. As existing sources of data dry up or become contaminated with A.I. “ slop ,” researchers say it makes it harder for newcomers to compete.
A.I.-generated words and images are already beginning to flood social media and the wider web . They’re even hiding in some of the data sets used to train A.I., the Rice researchers found .
“The web is becoming increasingly a dangerous place to look for your data,” said Sina Alemohammad , a graduate student at Rice who studied how A.I. contamination affects image models.
Big players will be affected, too. Computer scientists at N.Y.U. found that when there is a lot of A.I.-generated content in the training data, it takes more computing power to train A.I. — which translates into more energy and more money.
“Models won’t scale anymore as they should be scaling,” said Julia Kempe , the N.Y.U. professor who led this work.
The leading A.I. models already cost tens to hundreds of millions of dollars to train, and they consume staggering amounts of energy , so this can be a sizable problem.
Finally, there’s another threat posed by even the early stages of collapse: an erosion of diversity.
And it’s an outcome that could become more likely as companies try to avoid the glitches and “ hallucinations ” that often occur with A.I. data.
This is easiest to see when the data matches a form of diversity that we can visually recognize — people’s faces:
A.I. images generated by Sina Alemohammad and others .
This set of A.I. faces was created by the same Rice researchers who produced the distorted faces above. This time, they tweaked the model to avoid visual glitches.
This is the output after they trained a new A.I. on the previous set of faces. At first glance, it may seem like the model changes worked: The glitches are gone.
After two generations …
After three generations …
After four generations, the faces all appeared to converge.
This drop in diversity is “a hidden danger,” Mr. Alemohammad said. “You might just ignore it and then you don’t understand it until it's too late.”
Just as with the digits, the changes are clearest when most of the data is A.I.-generated. With a more realistic mix of real and synthetic data, the decline would be more gradual.
But the problem is relevant to the real world, the researchers said, and will inevitably occur unless A.I. companies go out of their way to avoid their own output.
Related research shows that when A.I. language models are trained on their own words, their vocabulary shrinks and their sentences become less varied in their grammatical structure — a loss of “ linguistic diversity .”
And studies have found that this process can amplify biases in the data and is more likely to erase data pertaining to minorities .
Perhaps the biggest takeaway of this research is that high-quality, diverse data is valuable and hard for computers to emulate.
One solution, then, is for A.I. companies to pay for this data instead of scooping it up from the internet , ensuring both human origin and high quality.
OpenAI and Google have made deals with some publishers or websites to use their data to improve A.I. (The New York Times sued OpenAI and Microsoft last year, alleging copyright infringement. OpenAI and Microsoft say their use of the content is considered fair use under copyright law.)
Better ways to detect A.I. output would also help mitigate these problems.
Google and OpenAI are working on A.I. “ watermarking ” tools, which introduce hidden patterns that can be used to identify A.I.-generated images and text.
But watermarking text is challenging , researchers say, because these watermarks can’t always be reliably detected and can easily be subverted (they may not survive being translated into another language, for example).
A.I. slop is not the only reason that companies may need to be wary of synthetic data. Another problem is that there are only so many words on the internet.
Some experts estimate that the largest A.I. models have been trained on a few percent of the available pool of text on the internet. They project that these models may run out of public data to sustain their current pace of growth within a decade.
“These models are so enormous that the entire internet of images or conversations is somehow close to being not enough,” Professor Baraniuk said.
To meet their growing data needs, some companies are considering using today’s A.I. models to generate data to train tomorrow’s models . But researchers say this can lead to unintended consequences (such as the drop in quality or diversity that we saw above).
There are certain contexts where synthetic data can help A.I.s learn — for example, when output from a larger A.I. model is used to train a smaller one, or when the correct answer can be verified, like the solution to a math problem or the best strategies in games like chess or Go .
And new research suggests that when humans curate synthetic data (for example, by ranking A.I. answers and choosing the best one), it can alleviate some of the problems of collapse.
Companies are already spending a lot on curating data, Professor Kempe said, and she believes this will become even more important as they learn about the problems of synthetic data.
But for now, there’s no replacement for the real thing.
About the data
To produce the images of A.I.-generated digits, we followed a procedure outlined by researchers . We first trained a type of a neural network known as a variational autoencoder using a standard data set of 60,000 handwritten digits .
We then trained a new neural network using only the A.I.-generated digits produced by the previous neural network, and repeated this process in a loop 30 times.
To create the statistical distributions of A.I. output, we used each generation’s neural network to create 10,000 drawings of digits. We then used the first neural network (the one that was trained on the original handwritten digits) to encode these drawings as a set of numbers, known as a “ latent space ” encoding. This allowed us to quantitatively compare the output of different generations of neural networks. For simplicity, we used the average value of this latent space encoding to generate the statistical distributions shown in the article.
Advertisement
Public-Good and Commercial
Institutional Toolkits
The Institutional Toolkit (ITK) is a model for an enhanced collection of tools and features that will help institutions discover, analyse and manage innovation knowledge.
Our Organisation
Release 9.2.
Improved accessibility, patent delta processing, Italian localisation, updated About site and more! New Features and ...
Recent News
SCImago Lab and The Lens are pleased to announce a research collaboration to advance the ...
Lens api terms, patseq bulk download terms.
The Lens serves global patent and scholarly knowledge as a public resource to make science- and technology-enabled problem solving more effective, efficient and inclusive.
SCImago Lab and The Lens are pleased to announce a research collaboration to advance the use of open data to create new mission-aligned metrics that enable academic and research institutions, businesses, civil society and government to measure and map the influence of scholarly publications on enterprise and improve the relevance of investment, research and partnering for public-good outcomes.
Under this collaboration, SCImago Lab and The Lens will extend the International Industry and Innovation Influence Mapping ( In4M ) capability to new journal-level metrics that illuminate the impact of scientific research on productive innovation based on nuanced understanding of the citations of journal articles within global patent documents. The project will utilize The Lens’ unique linked open data to calculate an In4M journal metric and create tools to map and explore the influence of research on innovation and help guide policy, research funding and partnerships.
In this collaboration, The Lens will supply uniquely-linked open data from both the global patent corpus and the entire research and scholarly literature and oversee its delivery, ensuring seamless access to comprehensive and current datasets. SCImago will develop enhanced methodologies, working closely with The Lens to create new classifications and an impact indicator based on the normalized patent data.
Advancements in journal metrics will reveal pathways for journals, publishers, authors and their institutions to better align their future strategies and actions with their core missions.
“Metrics for research and innovation must be aligned to institutional mission and priorities. Decision-making throughout the innovation ecosystem needs clarity to chart pathways and partnerships that can lead to new products, practices and processes that can change lives ” said Dr Richard Jefferson, CEO of Cambia, the parent organization of The Lens.
“The tired and inward-looking Journal Impact Factor needs a transparent and customizable reboot to show and celebrate how research can lead to real-world outcomes we urgently need.”
“SCImago is excited to lead research in developing innovative methodologies through collaborations with The Lens. These advancements in journal metrics will illuminate paths for journals and authors to better align their strategies with their core missions, for publishers to focus and promote the relevant outlets, and ultimately maximizing the influence of research on innovation” said Felix de Moya Anegon, CEO of SCImago Lab.
By combining The Lens’ robust data management capabilities, and world-class user experience with SCImago’s research metrics expertise, this collaboration seeks to drive positive societal impact through improved data accessibility and innovative research methodologies. Through this collaboration, SCImago Lab and The Lens are committed to fostering a more informed and connected global research community that serves society.
About SCImago Lab
SCImago Lab is a renowned research organization dedicated to analyzing and providing insights into scientific and scholarly activities worldwide. Utilizing advanced bibliometric and patent analysis tools, SCImago Lab offers comprehensive data and metrics that support academic institutions, researchers, and policymakers in evaluating research performance and trends. With a commitment to promoting transparency and accessibility in research, SCImago Lab plays a pivotal role in enhancing the understanding and impact of global scientific endeavors.
https://www.scimagolab.com/
About The Lens
https://www.lens.org/
The Lens is a world leader in providing free and open exploration, discovery and analysis of worldwide innovation knowledge, including patents and research knowledge, serving 270M+ scholarly work records, 155M+ patent documents from over 100 countries, and 490M+ biological sequences extracted from patents. The Lens has been operating for over twenty years as a project of the long-established non-profit social enterprise Cambia (doing business as The Lens), with support from leading global philanthropies and public institutions.
For more information, please contact [email protected] .
The lens awarded $2m usd to strengthen institutional innovation capabilities.
Highlighting some new visualization functions in lens 6.2.0 — three use cases for lens.org.
Latest Version
Improved accessibility, patent delta processing, Italian localisation, updated About site and more! New Features and Improvements Accessibility Improvements With this ...
Join our newsletter to receive frequent news updates, and notice on feature releases. We don’t share or sell your information ever.
The Government of Canada has a long-standing, foundational role in supporting and sustaining research at universities and health research institutions across the country. These institutions drive many of the discoveries and innovations that are important to Canada’s economic and societal well-being.
It is critical that we sustain Canada’s research excellence, talent and knowledge as the academic research enterprise deals with the impacts of the COVID-19 pandemic. Canada’s universities and health research institutions play integral roles in supporting both the COVID-19 response and the post-pandemic economic recovery.
This is why the Government of Canada recently announced a new temporary program , with an investment of $450 million for one year. This program will:
Universities will receive funding as block grants, allocated in two phases.
This is a tri-agency program jointly managed by the Social Sciences and Humanities Research Council (SSHRC), the Natural Sciences and Engineering Research Council (NSERC) and the Canadian Institutes of Health Research (CIHR). The Canada Research Coordinating Committee will provide strategic oversight of the fund.
The agencies are working with government departments on the program’s design and implementation, and will communicate with universities and health research institutions as soon as more details become available.
For more information, agency-specific measures to support researchers and graduate students, as well as new programs and initiatives enabling our research communities to address the COVID-19 crisis, visit:
arXiv's Accessibility Forum starts next month!
Help | Advanced Search
Title: assessing large language models for online extremism research: identification, explanation, and new knowledge.
Abstract: The United States has experienced a significant increase in violent extremism, prompting the need for automated tools to detect and limit the spread of extremist ideology online. This study evaluates the performance of Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-Trained Transformers (GPT) in detecting and classifying online domestic extremist posts. We collected social media posts containing "far-right" and "far-left" ideological keywords and manually labeled them as extremist or non-extremist. Extremist posts were further classified into one or more of five contributing elements of extremism based on a working definitional framework. The BERT model's performance was evaluated based on training data size and knowledge transfer between categories. We also compared the performance of GPT 3.5 and GPT 4 models using different prompts: naïve, layperson-definition, role-playing, and professional-definition. Results showed that the best performing GPT models outperformed the best performing BERT models, with more detailed prompts generally yielding better results. However, overly complex prompts may impair performance. Different versions of GPT have unique sensitives to what they consider extremist. GPT 3.5 performed better at classifying far-left extremist posts, while GPT 4 performed better at classifying far-right extremist posts. Large language models, represented by GPT models, hold significant potential for online extremism classification tasks, surpassing traditional BERT models in a zero-shot setting. Future research should explore human-computer interactions in optimizing GPT models for extremist detection and classification tasks to develop more efficient (e.g., quicker, less effort) and effective (e.g., fewer errors or mistakes) methods for identifying extremist content.
Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
Cite as: | [cs.CL] |
(or [cs.CL] for this version) | |
Focus to learn more arXiv-issued DOI via DataCite (pending registration) |
Access paper:.
Code, data and media associated with this article, recommenders and search tools.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .
IMAGES
VIDEO
COMMENTS
Research is creating new knowledge. The quote by Neil Armstrong, "Research is creating new knowledge," is a concise yet powerful statement that encapsulates the essence and significance of research. In straightforward terms, this quote implies that research is not merely about gathering existing information but rather about generating fresh ...
Generating new knowledge and insight We have a long-standing commitment to research. As I write this, the Health Foundation is currently supporting or working on over 160 research projects. And since 2004 for every £3 of grant funding we have awarded, around £1 has been invested in research and evaluation. All of this work has developed our ...
How research is creating new knowledge and insight. The pursuit of knowledge and discovery has always been an intrinsic human characteristic, but when new knowledge is curated and put in the right hands it has the power to bring about high value change to society. I work in the research team at the Health Foundation, an independent charity committed to bringing about better health and health ...
Abstract. A primary purpose of research is to generate new knowledge. Scientific advances have progressively identified optimal ways to achieve this purpose. Included in this evolution are the notions of evidence-based medicine, decision aids, shared decision making, measurement and evaluation as well as implementation.
The exchange of knowledge across different areas and disciplines plays a key role in the process of knowledge creation, and can stimulate innovation and the emergence of new fields. We develop ...
For some authors, innovation is a process wherein knowledge is acquired, shared, and assimilated to create new knowledge that embodies products and services (Herkema, 2003), methods and processes (Brewer & Tierney, 2012), and social and environmental contexts (Harrington et al., 2017). Characteristic of innovations is the creation of value.
In one example, a study can add to knowledge by addressing a gap in the literature. Inherent to any good study is the identification of a research gap. This can be achieved by a systematic review of the literature to identify an area that has not been addressed. This does not require a completely new topic.
Abstractspiepr Abs1. Every day people do research as they gather information to learn about something of interest. In the scientific world, however, research means something different than simply gathering information. Scientific research is characterized by its careful planning and observing, by its relentless efforts to understand and explain ...
Research refers to a systematic investigation carried out to discover new knowledge, test existing knowledge claims, solve practical problems, and develop new products, apps, and services. This article explores why different research communities have different ideas about what research is and how to conduct it.
"Basic research leads to new knowledge. It provides scientific capital. It creates the fund from which the practical applications of knowledge must be drawn. ... intellectual challenges of inquiry-driven basic research and are trained in, or create, new questions and ways of thinking. As these skills are applied to societal priorities,
It requires formulation and understanding of principles that guide practice and the development and testing of new ideas/theories. 7 Research aims to be objective and unbiased and contributes to the advancement of knowledge. Research adds to existing knowledge by offering an understanding or new perspective on a topic, describing the ...
The central purpose of all research, whether basic or applied, is to create new knowledge. Research in the domain of the geographical sciences is generally driven by a desire to generate new knowledge about that specific domain; that is, about the relationships among space, place, and the anthropogenic and non-anthropogenic features and ...
Research collaboration has long been suggested as an effective way to obtain innovative outcomes. Nevertheless, relatively little is known about whether and how different research collaboration strategies inspire or inhibit firms in the exploration of new knowledge. Drawing upon the research collaboration literature and social network theory, this study examines the effects of two specific ...
Research is a process to discover new knowledge. In the Code of Federal Regulations (45 CFR 46.102 (d)) pertaining to the protection of human subjects research is defined as: "A systematic investigation (i.e., the gathering and analysis of information) designed to develop or contribute to generalizable knowledge.".
Collecting and publishing existing knowledge isn't research, as it doesn't create new knowledge. Research isn't just data-gathering. Data-gathering is a vital part of research, but it doesn't lead to new knowledge without some analysis, some further work. Just collecting the data doesn't count, unless you do something else with it.
In contrast, the design mode focuses on improving ideas to create new knowledge. Activities such as hypothesising new ideas, evaluating the usefulness and adequacy of ideas, inventing and designing are common in design mode. ... Among the five theories of knowledge creation, empirical research on K-12 school teachers mainly referred to the ...
Research is a careful, systematic, and patient investigation in some field of knowledge, undertaken to establish facts or principles; it is a structured inquiry that utilizes an acceptable scientific methodology to collect, analyze, and interpret information to solve problems or answer questions and to create new knowledge that is generally ...
Some research has focused on "how to" co-create, especially in health and community settings ; however, there remains a lack of consensus on the meaning and use of the term co-creation of new knowledge. Many terms are used interchangeably and with ill-defined or no definition as to the meaning behind the terms.
Bibliometric data potentially can be used to create a number of additional indicators to provide further detail on linkages across research areas or by geographic location. This information can be particularly valuable for mapping the development of new research areas, such as green technologies, or the spread of general-purpose technologies.
INTRODUCTION. Creativity and critical thinking are of particular importance in scientific research. Basically, research is original investigation undertaken to gain knowledge and understand concepts in major subject areas of specialization, and includes the generation of ideas and information leading to new or substantially improved scientific insights with relevance to the needs of society.
Research is a process to discover new knowledge. In the Code of Federal Regulations (45 CFR 46.102(d)) pertaining to the protection of human subjects research is defined as: "A systematic investigation ( i.e., the gathering and analysis of information) designed to develop or contribute to generalizable knowledge." The National Academy of Sciences states that the object of research is to ...
In all, this supports the notion of steering, supporting, and challenging to create robust knowledge (Svensson et al., 2009). 6. Conclusions. The purpose of this paper was to shed light on the different types of knowledge created in an interactive research project and to analyze how they are linked to the project design, process, and content.
Developing T/C framework. Identify the specific variables. Create variables map (L.R map) Figure out how these are related. Group these variables into IV&DV. your research Problem/R GAP are base ...
As A.I.-generated data becomes harder to detect, it's increasingly likely to be ingested by future A.I., leading to worse results. By Aatish Bhatia Aatish Bhatia interviewed A.I. researchers ...
Our dedicated research team and extended network of partners engage in exploratory and applied research to create new knowledge, inform and enhance best practices, contribute to the development of new services and technologies, and build capacity through training and collaboration. Find out more about our research impact.
The Lens is a world leader in providing free and open exploration, discovery and analysis of worldwide innovation knowledge, including patents and research knowledge, serving 270M+ scholarly work records, 155M+ patent documents from over 100 countries, and 490M+ biological sequences extracted from patents.
This research seeks to obviate the need for creating QA datasets and grading (chatbot) LLM responses when comparing LLMs' knowledge in specific topic domains. This is done in an entirely end-user centric way without need for access to any inner workings of the LLM, so long as it can be prompted and given a random seed to create different generations to the same prompt. The paper does this by ...
It is critical that we sustain Canada's research excellence, talent and knowledge as the academic research enterprise deals with the impacts of the COVID-19 pandemic. Canada's universities and health research institutions play integral roles in supporting both the COVID-19 response and the post-pandemic economic recovery.
View a PDF of the paper titled Assessing Large Language Models for Online Extremism Research: Identification, Explanation, and New Knowledge, by Beidi Dong and 3 other authors View PDF Abstract: The United States has experienced a significant increase in violent extremism, prompting the need for automated tools to detect and limit the spread of ...