Balancing Access and Accountability: Ethical Challenges in Open-Source AI Deployment
Keywords
Loading...
Authors
Issue Date
2024-08-15
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
The release of ChatGPT by OpenAI started a significant shift in information retrieval, by providing
public access to a state-of-the-art Large Language Model (LLM). Since the release of ChatGPT,
the open-source community has significantly improved the quality of their models, making this
advanced technology accessible to anyone. With this accessibility of open-source models, certain
ethical questions arise, like: how does the trade-off between accessibility and accountability in
open-source AI models impact potential misuse and safety? This thesis answers this question and
gives recommendations on how to mitigate these risks in the short term while urging for subsequent
research into this topic. It does so by exploring the potential risks and misuse of LLMs, such as
the creation of misinformation, personalized scams, extremist and discriminatory texts and the
potential threat to cybersecurity. After analysing current technological safety measures and the
limitations of these open-source models it is evident that there is no technical solution that can
keep these models accessible to anyone while also guaranteeing safe deployment. Instead, more
focus should be on solutions around the deployment of these models to enhance safety. This thesis
suggests the implementation of a Certified Access System, usage monitoring, laws or regulations
which ensures that only models with adequate safety measures may be shared, and ethical training
for users. Other findings are that balancing accessibility and accountability is crucial for the safe
deployment of accessible open-source models, and that ethics must guide the design of AI to make
truly safe systems. This work contributes to the understanding of the ethical landscape of open source AI models and provides recommendations for further research to mitigate risks associated
with open-source AI systems.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen