AISB 2026 Symposium: Hype, Promise, and Speculation: AI Bubbles and the Replication Crisis in Computer Science

Symposium outline

In this symposium we intend to tackle complementary issues related to the likelihood of a replication crisis in computer science and computational methods, and an emerging AI bubble on the other. 

The replication crisis

The replication crisis has crossed multiple fields in science asking if results presented in published papers can be reproduced, repeated, and/or replicated. In their efforts to verify results various disciplines, including computer science, have already found that the answer for too many papers is “no” (Gundersen et al 2025, Cockburn et al 2020). In this symposium we look at the replication crisis as it pertains especially to computer science, whether within the discipline (cf. Cockburn et al 2020), or as applied to, or utilised in, other disciplines, such as computational modelling for neuroscience (Miłkowski et al 2018). 

There is also uncertainty about the extent to which ‘questionable research practices’ (QRPs) can be found in the above contexts. These can include manipulating data for statistically significant results (p-hacking), post hoc analysis to find statistically significant outcomes (p-fishing), or so as to present these as expected, i.e. ‘Hypothesising After the Results are Known’ (HARKing) (Cockburn et al 2020). Meanwhile, there are also proposals to address QRPs in computer science research, for instance through replication or the use of pre-study registered reports that include hypotheses and methods etc (Brown et al 2022).

AI bubbles

It’s clear that AI development is expanding substantially (Giattino et al 2023) , but the extent to which this growth is sustainable is unclear. Meanwhile, the possibility of this becoming another bubble, like those from the dot com boom and real estate, is clear (Carvão 2025). A bubble is a vague concept that captures where a process or commodity is valued or hyped beyond its intrinsic worth, typically in unsustainable ways. If contemporary expectations currently dominating the AI field do turn out to be a bubble we can expect further expansion, and then collapse, typically causing damage in the process. The economic damage of a collapse is already estimated by US commentators to rival the bursting of the dot-com bubble in 1990 and the financial crash of 2008 (Allyn 2025, Casselman and Ember 2025, Yip 2025). In the symposium we look beyond the speculation of AI stocks at the promises and reality of AI capabilities and what the effects of the potential bubble are.

In addition to the above are epistemic bubbles, which form around new or ‘popular’ ideas. ‘Epistemic bubbles’ may include ‘self-segregated’ networks of ‘like-minded people’ whose members are ‘liable to converge on and resist correction of false, misleading or unsupported claims’ (Sheeks 2023). These bubbles can in turn create ‘social epistemic’ structures which are similar to echo chambers, ‘in which other relevant voices have been actively discredited’ (Nguyen). In AI contexts, these epistemic bubbles might exclude voices who are critical of these technologies, or who doubt either its identity as AI, or its scope for positive impacts and change. Not least as ‘AI’ as a term brings greater expectations, including financial, compared with describing the technology in terms of its components and capacities, e.g. as LLMs, RAGs, DNNs, transformers, models, etc. Bubbles can also be created through the use of AI itself, for instance due to its scope for personalisation on media platforms, and agreeableness in GenAI chatbots, such that views of users are neither challenged nor developed.

References

Allyn, B. (November 2025). Here's why concerns about an AI bubble are bigger than ever. Published online at NPR. Retrieved from: https://www.npr.org/2025/11/23/nx-s1-5615410/ai-bubble-nvidia-openai-revenue-bust-data-centers

Brown, N. C., Marinus, E., & Hubbard Cheuoua, A. (2022, August). Launching registered report replications in computer science education research. In Proceedings of the 2022 ACM Conference on International Computing Education Research, Volume 1, 309-322.

Carvão, P. (August 2025). Is The AI Bubble Bursting? Lessons From The Dot-Com Era. Published online at Forbes.Retrieved from: https://www.forbes.com/sites/paulocarvao/2025/08/21/is-the-ai-bubble-bursting-lessons-from-the-dot-com-era/

Casselman, B. & Ember, S. (November 2025). The A.I. Boom Is Driving the Economy. What Happens if It Falters? Published online at NY Times. Retrieved from: https://www.nytimes.com/2025/11/22/business/the-ai-boom-economy.html

Cockburn, A., Dragicevic, P., Besançon, L., & Gutwin, C. (2020). Threats of a replication crisis in empirical computer science. Communications of the ACM, 63(8), 70-79.

Giattino, C., Mathieu, E., Samborska, V., & Roser, M. (2023) Artificial Intelligence Published online at OurWorldinData.org. Retrieved from: 'https://ourworldindata.org/artificial-intelligence'

Gundersen, O.E., Cappelen, O., Mølnå, M. and Nilsen, N.G. (2025). The Unreasonable Effectiveness of Open Science in AI: A Replication Study. Proceedings of the AAAI Conference on Artificial Intelligence. 39, 25 (Apr. 2025), 26211-26219. DOI:https://doi.org/10.1609/aaai.v39i25.34818.

Miłkowski, M., Hensel, W. M., & Hohol, M. (2018). Replicability or reproducibility? On the replication crisis in computational neuroscience and sharing only relevant detail. Journal of computational neuroscience, 45(3), 163-172.

Sheeks, M. (2023). The Myth of the Good Epistemic Bubble. Episteme, 20(3), 685–700. https://doi.org/10.1017/epi.2022.52

Nguyen, C. Thi. (2020). Echo Chambers and Epistemic Bubbles. Episteme 17 (2): 141–61.https://doi.org/10.1017/epi.2018.32.

Yip J. (October 2025) Are we in an AI bubble? Here’s what analysts and experts are saying Published online at cnbc.com. Retrieved from: https://www.cnbc.com/2025/10/21/are-we-in-an-ai-bubble.html

Submissions

We invite papers from a wide range of disciplines, including computer science, AI, Machine Learning, Natural Language Processing, Explainable AI, philosophy, behavioural sciences, psychology, social sciences, and those working with computational models, e.g. in finance. 

We welcome a broad variety of topics, including but not limited to:

• Machine learning (e.g. modelling, AI)
• Large language models
• Neural networks
• Deep learning
• Explainable AI
• Decision trees
• Replication crisis
• AI bubble(s)

Example research questions: 

• What kinds of impacts are computational methods having on science, e.g. machine learning methods, statistical analysis?
• How do computer science methods harm or help the replicability of research?
• Is research in computer science replicable?
• Does the name ‘Artificial Intelligence’ have an effect on what is expected of AI?
• Are current valuations (financial, social etc) of AI realistic?
• Is there an AI bubble in science?
• Related bubbles that might be relevant to these topics, e.g. is big data also a bubble?

Submission: Extended abstracts of 500 words (maximum, excluding references) to Easychair: https://easychair.org/conferences/?conf=aibc2026

Submission timeline

Organising Committee

• Y. J. Erden (University of Twente)
• Kiona Bijker (University of Twente)
• Katleen Gabriels (Maastricht University)
• Martin Lentschat (Université Toulouse)
• Doina Bucur (University of Twente)

Programme Committee 

• Maren Behrensen (Philosophy, University of Twente)
• Marcus Gerhold (Computer Science, University of Twente)
• Susannah E. Glickman (History, Stony Brook University)
• Adam Henschke (Philosophy, University of Twente)
• Saana Jukola (Philosophy, University of Twente)
• Miles MacLeod (Philosophy, University of Twente)
• Cyrus C. M. Mody (STS, Maastricht University)
• Yagmur Ozturk (Grenoble Informatics Laboratory (LIG), Université Grenoble Alpes)
• Stephen Rainey (Philosophy, TU Delft)
• Danielle Shanley (Philosophy, Maastricht Univertisy)
• Nicola Strisciuglio (Computer Science, University of Twente)
• Rob Wortham (Dept of Electronic and Electrical Engineering, University of Bath)

Schedule

The symposium talks will take place in Future Technologies Lab.
Abstracts can be found under the schedule. PDF versions with references and footnotes for each abstract can be found through the title links in the schedule.

Session 1 abstracts

Open Artifacts, Closed Research: How Shared Code Can Undermine Replicability
Adrian Gavornik, Katarína Marcinčinová, and Marek Havrila
The rapid developments in AI have contributed to a significant increase in the pace and volume of research in computer science. This growth has been accompanied by evolving publication practices enabling faster dissemination of results, increasingly relying on preprints and code sharing (Peng, 2011; Zhou et al., 2025; Cavenaghi et al., 2023). Although these practices are commonly intended to enhance transparency and reproducibility, we argue that they may produce unintended effects and undermine replicability. While reproducibility is treated as the cornerstone of current computer science research, the more fundamental question is whether research results truly can be trusted. Furthermore, the described tension between reproducibility and replicability opens up broader questions about the trustworthiness of computer science research, particularly in the context of trustworthy AI. If such systems are understood as socio-technical systems, whose reliability depends on technical, organizational, and epistemic practices, then the trustworthiness of AI cannot be separated from the trustworthiness of the scientific practices through which these systems are produced and validated.

We demonstrate how the re-use of code and evaluation methodologies in reproducibility studies facilitates the propagation of inaccuracies, including logical and implementation errors. We show these effects on a case study of a coherent line of five follow-up publications on multi-objective recommender systems (Xin Xin et al., 2025; Stamenkovic et. al., 2022; Paparella et al., 2023; Labarca Silva et al., 2024; Rajapakse and Jannach, 2025). We suggest that our observations are not just another example of questionable research practice or coincidental errors. Instead, they highlight structural vulnerabilities in current experimental practices that can be better understood in light of the two general tensions present in contemporary computer science research.

First tension points to the intricate relationship between reproducibility and replicability (Plesser, 2018; Raff et al., 2025). While both contribute to the reliability of the research and reduce accidental errors, randomness, or methodological flaws, reproducibility involves re-running the original code and data, whereas replicability requires an independent reconstruction of the model or method. When artifacts such as code are not shared, a study is not reproducible; however, it can still be replicable. Thus, reproducibility is not a prerequisite for replication. In this case study, we demonstrate that the availability of easy-to-use artifacts may, however, discourage genuine replication. Code reuse provides real benefits in terms of time and resource savings, while simultaneously creating an illusory assurance that errors and mistakes can be prevented by refraining from implementing the method from scratch.

Second, in the context of broader discussions on the role of reproducibility in scientific research as such (Fidler and Wilcox, 2026), we suggest that practices intended to promote transparency, such as shared code, datasets, and evaluation pipelines, can function as Latourian black-box mechanisms. As Latour (1987) argues in his laboratory studies on scientific practices, black-boxing occurs when a system works reliably enough that its internal assumptions are no longer questioned. In this sense, once the artifacts, such as shared code, produce seemingly stable and publishable outputs, their internal assumptions are no longer questioned and verified. We observed that an agreement and stabilized knowledge emerged not from independent validation, but from alignment with the same erroneous artifact. This creates an illusion of improved capabilities of multistakeholder recommender systems and scientific progress in the field as such.

The discussed tensions become especially relevant in the context of Trustworthy AI, where principles such as reliability, robustness, transparency, or accountability are often emphasized (see existing documents, such as the Ethics Guidelines for Trustworthy AI by European Commission & Directorate-General for Communications Networks, 2019; Fjeld et al., 2020). However, they often focus on the AI systems themselves, paying less attention to the scientific practices through which such systems are developed and evaluated. In other words, the trustworthiness of AI cannot be meaningfully separated from the trustworthiness of the research practices that produce it. What is required is more than just improving algorithmic or evaluation metrics, but also critically examining the practices and norms of contemporary computer science research, including code reuse, reproducibility, and experimental validation.

Bubbles, crises, and harms: the promises of GenAI
Y J Erden
This paper examines the rise of GenAI in the context of the replication crisis and suggests we are creating substantial harms and ignoring obvious risks. The ‘replication crisis’, which largely began in psychology, largely concerns difficulties reproducing or replicating a scientific study (Ioannidis 2005; Fanelli 2009; Ritchie 2020). These problems are not restricted to psychology however, and some have suggested there could be similar problems in empirical computer science (Cockburn et al. 2020).

Meanwhile the now familiar cycle of AI hype is once again peaking, with media outlets describing an ‘AI bubble’ comparable with the dot.com ‘boom and bust’ of the 1990s and 2000s (BBC, Guardian). This present bubble is largely driven by Generative artificial intelligence (GenAI) that relies on Large Language Models (LLMs), and attracts extensive funding as well as other resources (NY Times). These complex models are said to ‘generate high-quality, human-like material’, and in so doing produce ‘previously unseen synthetic content, in any form and to support any task’ (García-Peñalvo & Vázquez-Ingelmo 2023).

Yet these grand expectations are tempered by reports of fabrication, hallucination (LaGrandeur 2024), and of ‘accuracy collapse beyond certain complexities’ (Shojaee et al. 2025). Alongside which are questions about the ‘black box’ nature of the models, a lack of transparency about methods and data, plus difficulty in establishing where and how failure occurs (Barassi 2024). Despite this, we see a rise in the use of GenAI across diverse sectors. There is, for example, enthusiasm for these technologies in spheres where risks to livelihoods are nevertheless high, such as in education (Lee and Low 2024), human resources (Nyberg et al. 2025), and the arts (Epstein 2023).

This paper addresses these issues by first drawing comparisons to the development of GenAI with the replication crisis in other disciplines. In so doing, the paper offers reasons to treat GenAI with significantly more scepticism, not only about its scope, but also regarding methodological validity, the foundations for the models, and the (training) data on which they rely. Next the paper examines case studies in contexts where there is considerable scope for harm, both direct and indirect, such as in contexts of mental health care (Solaiman 2024) and the arts (Jiang 2023). Finally, the paper argues that not only is GenAI found wanting practically, ethically and socially, it remains to be seen whether it even meets the criteria to be considered an ‘artificial intelligence’. With these arguments, the paper concludes that GenAI has multiple failures in definition, form, content, and application. Thus, if we want to make good use of these technologies, we need to do so with our eyes firmly open to their limitations.

Session 2 abstracts

Different language games, different embeddings? How word embeddings could show language games differ in science
Kiona Bijker
This research will cover theory behind using natural language processing (NLP) word embedding methods to detect differences in language games played in science. Wittgenstein uses ’language games’ to explain how words gain meaning through use. Among a group of people, usage of a word often follows certain, unwritten, rules. These rules can differ from those in another group, like house rules for a card game which vary based on the group playing and where they are. At my friend’s house playing a 10 in a game of Mau-Mau means everyone hands their cards to the person to their left. Meanwhile, at my aunt’s house a 10 means the current player may play another card. The difference in word usage can lead to a difference in meaning between those groups (§65-69 Wittgenstein 1989). For example, to one group of friends ’going to the city’ means they meet up and travel together, where another group meets at a cafe in the city. Both groups on their own know what they mean, but if someone from the first group joins someone from the second they may both think they were stood up. In science different disciplines can play their own language games. Similar to the friend groups, this can cause confusion when collaborating across disciplines. Unlike the friend groups, the language games of disciplines often play out in published articles. This can cause misunderstandings when someone reads across different disciplines and make the articles outside their own discipline less accessible (Ellaway 2021).

To those familiar with NLP ’meaning through use’ may already sound similar to the idea behind word embeddings. Word embeddings are mathematical representations of how a word is used in text. The exact link between the embedding and the word’s use depends on the method, but is often based on which words occur around the embedded term. While the embeddings of words may not cover the entire ’context’ of a language game (Skelac and Jandri´c 2020) I argue they can be used to detect differences between language games of different groups. In this case word embeddings can be used to detect differences between disciplines’ language games around that word. This work therefore offers a new view of embedding distance to show differences between language games. Distance between group’s word embeddings can then be used to help those crossing between the language games know which words may lead to miscommunications

AI ethics is an epistemic bubble that must burst
Mary Lockwood
AI ethics promises fairness, accountability, and transparency, yet cannot deliver them. Contemporary AI ethics is an epistemic bubble in which the conditions required to realise these goals do not exist within its own structure. Its goals are admirable, but those it claims to protect have little meaningful role in defining the systems and criteria by which ethical compliance is determined (Birhane et al., 2022). Similar self-reinforcing tendencies have been identified in technological discourse and innovation-driven systems more broadly (Vinsel and Russell, 2020; Gertz, 2024). AI ethics is therefore asking the wrong question. Rather than pursuing impossible neutrality, it must interrogate whose knowledge builds systems and whose is excluded.

This paper introduces Equity Bias to explain how institutional AI ethics frameworks reproduce exclusion. It occurs through the selective incorporation of knowledge compatible with existing power structures, whilst marginalising forms of knowledge that challenge them. It provides a philosophical and practical framework for identifying how epistemic exclusion is embedded across the AI development lifecycle. It also shows how ethical governance processes can appear corrective whilst remaining structurally reinforcing.

The consequences are measurable. Hundreds of AI ethics guidelines now exist globally (Corrêa et al., 2023), yet automated decision-making in areas such as housing continues to produce discriminatory outcomes (Cheng et al., 2024). Those affected often have little meaningful recourse (Alon-Barkat et al., 2025). Harm persists not despite the ethics apparatus, but beneath its cover. The bubble not only fails but blocks the conditions under which genuine protection is possible.

Responses to ethical failures often follow a pattern which includes more guidelines and continued consultation with the same expert communities (Maclure and Morin-Martel, 2025). Equity Bias, when applied to AI ethics, reveals this pattern as epistemic replication: the reproduction of the same epistemic commitments regardless of outcome or harm. This invites comparison with bioethics, a field that has similarly grappled with questions of epistemic authority and representation (Hofmann, 2023).

Applied recursively, Equity Bias makes visible what the epistemic bubble conceals. Internal reform cannot resolve exclusions produced by the field’s own structures. Bursting the bubble requires creating conditions in which multiple knowledge systems can contest and inform both AI development and the ethics field itself. The question is not whether AI ethics must change, but who gets to determine what it changes into.

Call for Abstracts

Please send any questions to Y. J. Erden (University of Twente):

1-2 July 2026

AISB 2026, University of Sussex, UK, https://aisb.org.uk/

Keynote Speaker: Anil Seth, Professor of Cognitive and Computational Neuroscience, University of Sussex

Day of celebration: life and work of Prof Margaret Boden, 30 June. 

Attendance at the day of celebration is free for attendees of the AISB Convention. 

Symposium outline

In this symposium we intend to tackle complementary issues related to the likelihood of a replication crisis in computer science and computational methods, and an emerging AI bubble. 

Symposium website: https://aisb.org.uk/aisb-2026-symposium-hype-promise-and-speculation

Submission: Extended abstracts of 500 words (maximum, excluding references) to Easychair: https://easychair.org/conferences/?conf=aibc2026

Deadline: 23 March 2026

OVERVIEW: 

The replication crisis

The replication crisis has crossed multiple fields in science asking if results presented in published papers can be reproduced, repeated, and/or replicated. In their efforts to verify results various disciplines, including computer science, have already found that the answer for too many papers is “no” (Gundersen et al 2025, Cockburn et al 2020). In this symposium we look at the replication crisis as it pertains especially to computer science, whether within the discipline (cf. Cockburn et al 2020), or as applied to, or utilised in, other disciplines, such as computational modelling for neuroscience (Miłkowski et al 2018). There is also uncertainty about the extent to which ‘questionable research practices’ (QRPs) can be found in the above contexts. These can include manipulating data for statistically significant results (p-hacking), post hoc analysis to find statistically significant outcomes (p-fishing), or to present these as expected, i.e. ‘Hypothesising After the Results are Known’ (HARKing) (Cockburn et al 2020). Meanwhile, there are also proposals to address QRPs in computer science research, for instance through replication or the use of pre-study registered reports that include hypotheses and methods etc (Brown et al 2022).

AI bubbles

It’s clear that AI development is expanding substantially (Giattino et al 2023) , but the extent to which this growth is sustainable is unclear. Meanwhile, the possibility of this becoming another bubble, like those from the dot com boom and real estate, is clear (Carvão 2025). A bubble is a vague concept that captures where a process or commodity is valued or hyped beyond its intrinsic worth, typically in unsustainable ways. If contemporary expectations currently dominating the AI field do turn out to be a bubble we can expect further expansion, and then collapse, typically causing damage in the process. The economic damage of a collapse is already estimated by US commentators to rival the bursting of the dot-com bubble in 1990 and the financial crash of 2008 (Allyn 2025, Casselman and Ember 2025, Yip 2025). In the symposium we look beyond the speculation of AI stocks at the promises and reality of AI capabilities and what the effects of the potential bubble are. In addition to the above are epistemic bubbles, which form around new or ‘popular’ ideas. ‘Epistemic bubbles’ may include ‘self-segregated’ networks of ‘like-minded people’ whose members are ‘liable to converge on and resist correction of false, misleading or unsupported claims’ (Sheeks 2023). These bubbles can in turn create ‘social epistemic’ structures which are similar to echo chambers, ‘in which other relevant voices have been actively discredited’ (Nguyen 2020). In AI contexts, these epistemic bubbles might exclude voices who are critical of these technologies, or who doubt either its identity as AI, or its scope for positive impacts and change. Not least as ‘AI’ as a term brings greater expectations, including financial, compared with describing the technology in terms of its components and capacities, e.g. as LLMs, RAGs, DNNs, transformers, models, etc. Epistemic bubbles can also be created through the use of AI itself, for instance due to its scope for personalisation on media platforms, and agreeableness in GenAI chatbots, such that views of users are neither challenged nor developed.

TOPICS OF INTEREST 

We invite papers from a wide range of disciplines, including: computer science, AI, Machine Learning, Natural Language Processing, Explainable AI, philosophy, behavioural sciences, psychology, social sciences, and those working with computational models, e.g. in finance. 

We welcome a broad variety of topics, including but not limited to:

• Machine learning (e.g. modelling, AI)
• Large language models
• Neural networks
• Deep learning
• Explainable AI
• Decision trees
• Replication crisis
• AI bubble(s)

Example research questions: 

• What kinds of impacts are computational methods having on science, e.g. machine learning methods, statistical analysis?
• How do computer science methods harm or help the replicability of research?
• Is research in computer science replicable?
• Does the name ‘Artificial Intelligence’ have an effect on what is expected of AI?
• Are current valuations (financial, social etc) of AI realistic?
• Is there an AI bubble in science?
• Related bubbles that might be relevant to these topics, e.g. is big data also a bubble?

SUBMISSION AND PUBLICATION DETAILS 

Submission: Extended abstracts of 500 words (maximum, excluding references) to Easychair: https://easychair.org/conferences/?conf=aibc2026

Deadlines: 

• Abstract submission deadline: 23 March 2026
• Notification of acceptance/rejection decisions: 17 April 2026
• Final versions of accepted abstracts: 15 May 2026
• Conference: 1 to 2 July 2026 [symposium date tbc] 


SYMPOSIUM ORGANISERS: 

Organising Committee

• Y. J. Erden (University of Twente) y.j.erden@utwente.nl
• Kiona Bijker (University of Twente) k.bijker@student.utwente.nl
• Katleen Gabriels (Maastricht University) k.gabriels@maastrichtuniversity.nl
• Martin Lentschat (Université Toulouse) martin.lentschat@univ-tlse2.fr 
• Doina Bucur (University of Twente) d.bucur@utwente.nl

Programme Committee 

• Maren Behrensen (Philosophy, University of Twente)
• Marcus Gerhold (Computer Science, University of Twente)
• Susannah E. Glickman (History, Stony Brook University)
• Adam Henschke (Philosophy, University of Twente)
• Saana Jukola (Philosophy, University of Twente)
• Miles MacLeod (Philosophy, University of Twente)
• Cyrus C. M. Mody (STS, Maastricht University)
• Yagmur Ozturk (Grenoble Informatics Laboratory (LIG), Université Grenoble Alpes)
• Stephen Rainey (Philosophy, TU Delft)
• Danielle Shanley (Philosophy, Maastricht Univertisy)
• Nicola Strisciuglio (Computer Science, University of Twente)
• Rob Wortham (Dept of Electronic and Electrical Engineering, University of Bath)

About the AISB: https://aisb.org.uk/

The Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB) is the largest Artificial Intelligence Society in the United Kingdom. Founded in 1964, the society has an international membership from academia and industry, with a serious interest in Artificial Intelligence, Cognitive Science and related areas. It is a member of the European Coordinating Committee for Artificial Intelligence. The AISB Convention typically consists of a set of co-located symposia on a wide-range of topics in AI and the simulation of behaviour; there are often also a number of plenary lectures, and other events such as public engagement sessions, and historical/artistic exhibitions. The symposium model allows for the community to decide what the current topics of interest are and the direction that the field is heading. The event is central to the AISB and its mandate of promoting AI research, and in providing early career researchers and students a supportive environment in which to discuss their research. 

References

Allyn, B. (November 2025). Here's why concerns about an AI bubble are bigger than ever. Published online at NPR. Retrieved from: https://www.npr.org/2025/11/23/nx-s1-5615410/ai-bubble-nvidia-openai-revenue-bust-data-centers

Brown, N. C., Marinus, E., & Hubbard Cheuoua, A. (2022, August). Launching registered report replications in computer science education research. In Proceedings of the 2022 ACM Conference on International Computing Education Research, Volume 1, 309-322.

Carvão, P. (August 2025). Is The AI Bubble Bursting? Lessons From The Dot-Com Era. Published online at Forbes.Retrieved from: https://www.forbes.com/sites/paulocarvao/2025/08/21/is-the-ai-bubble-bursting-lessons-from-the-dot-com-era/

Casselman, B. & Ember, S. (November 2025). The A.I. Boom Is Driving the Economy. What Happens if It Falters? Published online at NY Times. Retrieved from: https://www.nytimes.com/2025/11/22/business/the-ai-boom-economy.html

Cockburn, A., Dragicevic, P., Besançon, L., & Gutwin, C. (2020). Threats of a replication crisis in empirical computer science. Communications of the ACM, 63(8), 70-79.

Giattino, C., Mathieu, E., Samborska, V., & Roser, M. (2023) Artificial Intelligence Published online at OurWorldinData.org. Retrieved from: 'https://ourworldindata.org/artificial-intelligence'

Gundersen, O.E., Cappelen, O., Mølnå, M. and Nilsen, N.G. 2025. The Unreasonable Effectiveness of Open Science in AI: A Replication Study. Proceedings of the AAAI Conference on Artificial Intelligence. 39, 25 (Apr. 2025), 26211-26219. DOI:https://doi.org/10.1609/aaai.v39i25.34818.

Miłkowski, M., Hensel, W. M., & Hohol, M. (2018). Replicability or reproducibility? On the replication crisis in computational neuroscience and sharing only relevant detail. Journal of computational neuroscience, 45(3), 163-172.

Sheeks, M. (2023). The Myth of the Good Epistemic Bubble. Episteme, 20(3), 685–700. https://doi.org/10.1017/epi.2022.52

Nguyen, C. Thi. 2020. “ECHO CHAMBERS AND EPISTEMIC BUBBLES.” Episteme 17 (2): 141–61.https://doi.org/10.1017/epi.2018.32.

Yip J. (October 2025) Are we in an AI bubble? Here’s what analysts and experts are saying Published online at cnbc.com. Retrieved from: https://www.cnbc.com/2025/10/21/are-we-in-an-ai-bubble.html