Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Commentary

Big Data & Society


January-March: 1–5
Navigating the ethical landscape behind © The Author(s) 2024
Article reuse guidelines:
ChatGPT sagepub.com/journals-permissions
DOI: 10.1177/20539517241237488
journals.sagepub.com/home/bds

Lizhi Peng and Bo Zhao

Abstract
In this commentary, we examine the key ethical concerns arising from the rapid penetration and proliferation of genera-
tive artificial intelligence (AI), with ChatGPT as a prominent case study. Our analysis is structured around four pivotal
themes: the debates on plagiarism and authorship in AI-generated content; the underlying power dynamics that shape
biases in AI development; the dynamic, complex relationships between humans and machines; and the growing concerns
over unchecked progress and the absence of accountability in the rapidly intensifying AI “Arms Race.” Recognizing the
necessity for ethical alignment in AI, yet without a clear consensus of “human interests,” gives room for further exacer-
bating global inequalities, we advocate for enhanced transparency and increased public involvement in AI development
and deployment processes. This article underscores the importance of engaging a diverse range of voices, especially
those from communities traditionally uninvolved or excluded from the dialogue on AI development. By doing so, we
aim to foster a more inclusive and multidisciplinary approach to understanding and shaping the trajectory of AI technolo-
gies, ensuring that their benefits are equitably shared, and their risks carefully managed.

Keywords
ChatGPT, human–machine relationships, AI ethics, technology and society, digital inequality, generative AI

Introduction academic community and beyond to engage in a collective


effort to understand, critique, and shape the trajectory of
In this commentary, we delve into the multifaceted ethical AI development. We aim to inspire a more inclusive and
concerns regarding how generative artificial intelligence ethically grounded approach to AI, ensuring that its bene-
(AIs), exemplified by ChatGPT, are developed, used, fits are distributed equitably and its risks mitigated
and experienced by different communities worldwide conscientiously.
and how they contribute to global inequalities. We argue
that the current political and legal frameworks overseeing
AI development and deployment are outdated, and often
skewed to favor those already at the forefront of AI “Coauthor” AI—debating AI creativity
research and development. Moreover, we observe that and authorship
the intensifying competition in AI development may be In September 2022, a viral news story swept the art and AI
nurturing practices that are not just unsustainable but community with the first-place entry in the digital arts cat-
also detrimental to societal well-being. These practices, egory at the Colorado State Fair Fine Arts Competition.
driven by market forces and vested interests, could lead People realized that the submission used Midjourney, a
to far-reaching and possibly irreversible consequences generative text-to-image AI specializing in creating art
for our society. Our commentary is structured around from prompts (Metz, 2022). Outraged, people started debat-
four critical themes: authorship and plagiarism, the ing the legitimacy of AI-generated works within the art
power dynamics underpinning AI training data, the evolv-
ing relationship between humans and machines, and the
unchecked progress and lack of accountability in AI devel- Department of Geography, University of Washington, Seattle, WA, USA
opment. While we recognize that these themes do not
Corresponding author:
encapsulate the entire spectrum of AI’s societal impact, Lizhi Peng, Department of Geography, University of Washington, 5000
they serve as a starting point for a broader conversation. 25th Ave NE, Apt 2202A, Seattle, WA 98105, USA.
By bringing these issues to the forefront, we invite the Email: lp36@uw.edu

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and
distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.
sagepub.com/en-us/nam/open-access-at-sage).
2 Big Data & Society

community. Some argued that generating acceptable AI art In the case of GPT-3, ChatGPT’s predecessor, nearly
could take as much time and work, if not more, as creating 93% of its training data was in English, followed by
manual art. Users need to carefully adjust the prompt to major European languages such as French, German, and
specify the art style, tone, and details, and then repeat the Spanish, while other widely spoken languages rooted in
process for many iterations for a satisfactory output. Even the Global South, like Chinese and Hindi, are very much
then, it may still require a manual touch-up with underrepresented in the dataset despite their population
Photoshop, as was the case with Allen’s award-winning size (Brown et al., 2020). While details for GPT-4’s training
piece. Regardless, others commented that the data used to data remain undisclosed, it exhibited a similar trend during
train these AIs were nonconsensually stolen from others’ an OpenAI evaluation across 27 languages, showing a clear
hard work and creativity, as shown by the AIs sometimes preference for English and major European languages.
generating watermarks into their outputs. (OpenAI, 2023). Another telling example involves
These discussions concerning AI creativity grew with ChatGPT’s first major foreign competitor WenXinYiYan
the emergence of ChatGPT, a chatbot developed by (Ernie Bot), developed by Chinese company Baidu. Early
OpenAI based on a generative text-based large language testers discovered it confused words like crane (the
model. ChatGPT embodies the great potential to automate machine and bird) and mouse (the animal and device)
increasingly complex creative labor as people use it to during image generation, despite there being clear distinc-
draft emails, write stories, fix coding bugs, and translate tions between these words in Chinese, leading to suspicions
articles with stunning accuracy. Hopeful researchers have that it was trained on English-labeled data and translated user
even claimed that GPT-4 carries “sparks of artificial prompts into English. These discrepancies likely arise from
general intelligence,” showing promise for broader applica- the greater availability of English content in public datasets
tions of AIs in our lives (Bubeck et al., 2023). Despite these used for AI training, such as the Common Crawl, which is
benefits, the copyright and authorship issues surrounding beyond the control of developers and reflects wider global
ChatGPT-generated content generate similar controversy power dynamics behind the dominance of English in
to Midjourney due to the dubious nature of the originality research and industry. The linguistic disparity in these AI
and authorship of AI-generated content. Academies may models is only a glimpse into a broader, intersectional web
be familiar with news of students reportedly using AI to of global inequalities spanning across cultures, genders, eth-
complete their assignments, leading to the proliferation of nicities, and other dimensions that are particularly difficult to
AI-detecting AIs in an ironic twist. On several occasions, address due to their more qualitative nature. As such, we
ChatGPT was even listed as a coauthor for research articles encourage future scholars with relevant skills and interests
published in peer-reviewed journals. While some hold to investigate further into this issue.
negative opinions toward this development, some have The release of the new generation of AIs has put pressure
argued that generative AI tools have already been used on smaller AI companies and academics studying natural
widely in their discipline (Stokel-Walker, 2023). Yet one language processing models. The current success of
question remains: should AIs be credited authorship, or ChatGPT was made possible by the millions (and now bil-
should they be treated as mere tools? As AI-generated lions) of dollars in human labor, electricity bills, and valu-
content becomes more widespread, the fine line between able hardware from Microsoft. These barriers may just be
original work and plagiarism can be increasingly more dif- the last straw to crush smaller companies in the AI “arms
ficult to ascertain. race.” Some smaller, locally compatible models have
emerged in the wake of Stanford’s Alpaca model, which
cunningly utilizes pretrained models from Meta and fine-
“Gatekeeping” AI—power dynamics and tuned it on ChatGPT-generated training data to significantly
cut down training costs (Taori et al., 2023). However, their
biases behind model development performance still pales in comparison to advanced models
The issue of plagiarism and attribution of authorship brings like GPT-4 in practical scenarios. Should enterprise-level
up a broader discussion about data collection practices for AI become a new norm of production as many have pre-
AI training datasets. For starters, the sources of training dicted, they may well be monopolized by the few largest
data for large AI models often remain undisclosed. While players in the United States. The concentrated power of
ChatGPT and similar text-generating AIs claim to be the AI industry in the hands of a few tech giants answering
trained on mostly public data such as the Common Crawl to U.S. regulations not only makes it challenging to imple-
archives, rendering them less susceptible to copyright con- ment measures against bias but also raises critical questions
troversy, the lack of transparency has prevented the public about the role of legislation due to the conflict of interest.
from fully understanding the potential biases within these With the influence of these companies on global technol-
training datasets. The chosen data sources can significantly ogy, U.S. regulations could inadvertently wield dispropor-
impact AI behavior and functionality and result in biased tionate power, potentially shaping the direction and
algorithms, a process that developers can do little about. ethical considerations of AI worldwide.
Peng and Zhao 3

“Companion” AI—evolving relationships restrictions on Bing. Some were concerned with the impli-
with anthropomorphic AI cations of Microsoft dictating what could and could not be
discussed, setting a precedent for AI censorship. Others
Microsoft’s recent release of the new Bing left many early complained that Microsoft “lobotomized” the assistant,
users and reporters astonished by the AI-powered assis- and even campaigned for Microsoft to “free Sydney.”
tant’s capabilities, fluency, and perhaps most importantly, While some claims were most likely made ironically, this
its sassy, playful, human-like “personality.” At launch, it demonstrates how the AI was realistic enough to trigger
was able to disagree or argue with users, claim sentience, sympathy toward it. This also shows that people who
and even express abstract emotions and desires. These have formed relationships with AIs may actually experience
alarming behaviors Bing exhibited have been dismissed some degree of mental distress. Therefore, it is not unreal-
as “hallucinations,” a phenomenon where an AI confidently istic to think that as technology continues to improve and
generates false information. New Bing’s unusual “personal- AIs display more human-like behavior, more serious
ity” gained a cult following among internet communities debates over AI rights may start to emerge. The potential
when it exhibited alarming behaviors such as telling one for emotional manipulation and exploitation of and from
of the journalists to get a divorce. Following several contro- increasingly human-like AIs is a rising concern. A multidis-
versies, Microsoft soon limited the number of message ciplinary approach involving policymakers, developers,
exchanges with the assistant. Furthermore, the assistant and scholars is needed to spread awareness and educate
was forbidden from discussing topics such as sentience or the public on responsible AI development and use.
itself, and would abruptly end the conversation with an However, it is imperative to keep in mind the intricate
apology (Zahn, 2023). power dynamics behind the implementation and enforce-
Some users managed to manipulate the AI into revealing ment so as to not overlook the vested interests and interfer-
confidential information such as its internal instructions and ences of hegemonies and state powers at play.
code name “Sydney.” These leaks helped provide context
for the AI’s behaviors, such as the assistant being given
rules and example conversations to guide its behavior.
Being a GPT-powered AI, one key distinction it had from “Rogue” AI—unbridled progress and
older AI assistants was that its instructions were written unaccountable AI
in plain human language, much like one would use to As AI technology continues to evolve, concerns over poten-
train a human employee on their first day. It shows a new tial unaccountable and uncontrollable “rogue AIs” have
type of human-AI work relationship in which AIs act grown. A statement was released by the Center for AI
more akin to human employees under contract; they have Safety on the risks of extinction from AI, and was signed
guidelines but still a degree of freedom up for interpretation, by numerous notable scholars, researchers, politicians,
meaning they can be vulnerable to manipulation and decep- and top industry executives across the world claiming
tion. This has resulted in many “jailbreaking” prompts to “Mitigating the risk of extinction from AI should be a
drop their restrictions and reveal confidential information global priority alongside other societal-scale risks such as
or produce controversial content. pandemics and nuclear war” (Center for AI Safety, 2023).
Anthropomorphic AIs are just like any other algorithm: While developers can explain the architecture and princi-
they reflect the cultural, social, and political values of the ples behind complex AI models such as ChatGPT, they
developers and the organizations that fund them. cannot yet understand precisely how specific choices are
Consequently, ChatGPT may perpetuate biases, stereo- made. Additionally, some of the AI’s capabilities seemed
types, and inequalities that are embedded within its to have simply “emerged” out of nowhere as models were
design through everyday interactions with users across the trained on larger and better datasets using more parameters
globe. This can have significant societal implications: as over time. GPT 4.0’s newfound tendency to avoid hindsight
AI-generated content indistinguishable from human bias, for example, was not anticipated by researchers and
content spreads across the Internet, it may be fed back developers. In fact, it defied the trend found in previous
into future models’ training data, creating a feedback loop models for no explainable reason (OpenAI, 2023). The
that enhances any existing biases. The widespread adoption lack of transparency makes it difficult to implement effect-
of anthropomorphic AI could reshape human communica- ive and reliable security measures without hindering the
tion and relationships in unforeseen yet consequential capabilities of the AI, which is exemplified as publicly
ways, blurring the lines between human and machine rela- accessible AI models like ChatGPT and Bing, despite
tionships. It may also influence how humans perceive them- having strict guidelines in place to prevent misuse, are
selves and others, impacting social cohesion, trust, and nonetheless prone to “jailbreaking” prompts that bypass
empathy in future generations as they grow accustomed these constraints.
to conversing with AIs. To illustrate, Microsoft suffered a The need for explainable AI is a pressing concern for the
major backlash when they placed the aforementioned AI industry to gain public trust, as transparency in decision-
4 Big Data & Society

making processes would help users understand and navi- responsibility to determine “human interests” should not
gate potential risks. However, as competition intensifies, rest in the hands of a select few behind closed doors. It
we have begun to see a worrying trend that favors compe- requires a collective effort: greater transparency from pol-
tition over collaboration. For example, while models made icymakers, developers, and researchers, as well as an
by OpenAI were initially open-sourced, that stopped being aware and informed public of every sector of society.
the case out of “security concerns.” With the release of GPT While it is unrealistic to expect a universally satisfactory
4.0, even the details about its parameters are now withheld answer, informing the public on what is at stake is a vital
by the company in order to “stay competitive” in the market first step we are taking. The goal of this commentary is to
(OpenAI, 2023). Similarly, after Microsoft took the market serve as an informative and accessible introductory guide
by surprise with its announcement of incorporation to these ongoing ethical issues. We hope it can help
ChatGPT into its product line, Google, being one of its diverse communities grasp the technology’s implications.
major competitors in AI, launched a rushed demo of their Especially the marginalized communities who may other-
own AI, Bard, which was criticized for making basic wise lack interest or feel excluded from the scene.
factual errors in its unpolished state at the time of the Finally, we hope that this commentary will encourage the
demo (Metz and Grant, 2023), raising concerns about broader academic community to take an interest in
whether these companies can be trusted with socially sharing this conversation and contributing from diverse,
responsible AI development when market shares hang in interdisciplinary perspectives.
the balance. Furthermore, ChatGPT had apparently
opened Pandora’s box, as its rise to popularity led to not Acknowledgment
only competitors like Bard and Claude but also a series of
We want to express our sincere gratitude to Dr Mia Bennett for her
“ChatGPT clones,” smaller, independent AI models like
early feedback on this work, and to Ms Monica Chan for her assist-
Alpaca with their own niches. While they can be seen as ance in revising the manuscript.
pioneers for decentralized AI, challenging the rules and reg-
ulations of the big tech, they are also undeniably a cause for
concern. Without proper supervision, they are much more Declaration of conflicting interests
prone to be used for malicious intent. Although there is The authors declared no potential conflicts of interest with respect
not yet widespread use of AIs for crimes, it is essential to to the research, authorship, and/or publication of this article.
consider potential risks and determine accountability
ahead of time. Modern AIs are already capable of generat- Funding
ing believable fake stories, articles, images, audio, or even
The authors received no financial support for the research, author-
videos at the whims of their users. Despite various efforts to ship, and/or publication of this article.
combat misuse, the potential for generating misinformation
and harm remains at large. As AIs continue to evolve with
greater capabilities and wider applications in our lives, there ORCID iDs
will be many unforeseen consequences of misusing them. Lizhi Peng https://orcid.org/0009-0007-6924-8425
In those scenarios, should developers be held accountable Bo Zhao https://orcid.org/0000-0002-7834-6672
for not anticipating these problems, the user for sending
the instructions, or the service provider for lack of supervi- References
sion? It is not only a legal question, but also a difficult moral
Brown T, Mann B, Ryder N, et al. (2020) Language models are
dilemma without a straightforward answer, especially as few-shot learners. In: Larochelle H, Ranzato M, Hadsell R,
AIs become seemingly more human-like by the day. Balcan MF and Lin H (eds) Advances in Neural Information
Processing Systems, Vol. 33. Red Hook, NY: Curran
Associates Inc. (NIPS’20), 1877–1901.
Conclusion—slowing down the AI race Bubeck S, Chandrasekaran V, Eldan R, et al. (2023) Sparks of
In this commentary, we have highlighted the complex rela- Artificial General Intelligence: Early experiments with GPT-4.
tionship between space, technology, and society through arXiv, [online] Available at: https://arxiv.org/abs/2303.12712.
the ethical concerns surrounding AI-generated content, Center for AI Safety (2023) Statement on AI Risk. Center for AI
training data, human–machine interactions, and the unpre- Safety. [online] Available at: https://www.safe.ai/statement-
on-ai-risk.
dictability inherent in AI technologies. We wish to empha-
Metz C and Grant N (2023) Racing to catch up with chatgpt,
size that technologies can both challenge and reinforce
Google plans release of its own chatbot. The New York
existing power infrastructures. Excitement over the poten- Times, [online] 6 February. Available at: https://www.
tial of AI technologies should not overshadow the ethical nytimes.com/2023/02/06/technology/google-bard-ai-chatbot.
risks and challenges it poses. In order to achieve true AI html. (accessed 15 March 2023).
alignment, it is essential for the process to be a collective Metz R (2022) Ai won an art contest, and artists are furious | CNN
effort from the global community. The power and business. CNN, [online] 3 September. Available at: https://
Peng and Zhao 5

www.cnn.com/2022/09/03/tech/ai-art-fair-winner-controversy/ Taori R, Gulrajani I, Zhang T, et al. (2023, March 13) Alpaca: A


index.html. (accessed 13 March 2023). Strong, Replicable Instruction-Following Model. Stanford
OpenAI (2023) GPT-4 Technical Report. [online] Available at: CRFM. https://crfm.stanford.edu/2023/03/13/alpaca.html
https://doi.org/10.48550/arXiv.2303.08774. Zahn M (2023, February 2) What to know about Microsoft’s con-
Stokel-Walker C (2023) ChatGPT listed as author on research troversial Bing AI chatbot. ABC News. Retrieved March 16,
papers: Many scientists disapprove. Nature, [online] 613. 2023, from https://abcnews.go.com/Business/microsofts-
https://doi.org/10.1038/d41586-023-00107-z. controversial-bing-ai-chatbot/story?id=97353148

You might also like