AI & Ethics

Australia was one of the 28 signatories to the Bletchley Declaration, which emerged from the global AI Safety Summit, where heads of state, senior ministers, AI leaders and other experts from across the globe congregated to set an international framework for developing safe AI.  The Declaration provides that the agenda for addressing frontier AI risk will focus on:

• identifying AI safety risks of shared concern, building a shared scientific and evidence-based understanding of these risks, and sustaining that understanding as capabilities continue to increase, in the context of a wider global approach to understanding the impact of AI in our societies.
• building respective risk-based policies across our countries to ensure safety in light of such risks, collaborating as appropriate while recognising our approaches may differ based on national circumstances and applicable legal frameworks. This includes, alongside increased transparency by private actors developing frontier AI capabilities, appropriate evaluation metrics, tools for safety testing, and developing relevant public sector capability and scientific research.

Read the Declaration here.

Two days before the Bletchley Declaration, U.S President, Jo Biden, issued his first Executive Order on the Safe, Secure and Trustworthy Development and Use of AI stating that his Administration ‘places the highest urgency on governing the development and use of AI safely and responsibly, and is therefore advancing a coordinated, Federal Government-wide approach to doing so’.  The Order sets out key directives that will be addressed including:

• Directing the development of standardized evaluations of AI systems standards and rules across federal government agencies
• Recruiting AI talent including to provide for more ‘highly skilled immigrants and non-immigrants’ with critical AI expertise to stay and work in the U.S.
• Calling on Congress to publish its forthcoming AI study on copyright issues raised by AI
• Calling on Congress to pass a Federal Privacy Bill
• Aligning with other global actions (G7, European Union Artificial Intelligence Act, etc.)

Read the Executive Order here.

Given the push to build AI at ever increasing scale and the risks, this timely report from the AI Now Institute looks at the material costs and why concentration in compute is driving a race to the bottom.

As the report explains, computational power, is a core dependency in building large-scale AI.  Amid a steadily growing push to build AI at larger and larger scale, key points in the supply chain are monopolised by a handful of organisations.  As the report shows, it influences the behaviour of even the biggest AI companies as they encounter the effects of compute scarcity. A recent report from Andreessen Horowitz describes compute as “a predominant factor driving the industry today,” noting that companies have spent “more than 80% of their total capital on compute resources.”

Understanding the material underpinnings of AI is an important point for examining its effects on the broader public. This report provides a useful guide for understanding computational power.  Access the Report here.

In the day between the U.S President’s Executive Order on AI and the Bletchley Declaration being signed, Singapore’s IMDA and AI Verify Foundation launched the “Generative AI Evaluation Sandbox”, a new initiative to build knowledge and develop new benchmarks and tests for generative AI (GAI) systems.

This is part of the effort to have a common standard to assess AI and provides a baseline by offering a research-based categorisation of current evaluation benchmark and methods. It is anticipated that the Sandbox use cases will reveal gaps in the current landscape of Gen AI evaluations, particularly in domain-specific (eg HR or Security) and cultural-specific areas which are currently under-developed.

Access the Generative AI Evaluation Sandbox | IMDA – Infocomm Media Development Authority here

Two recent examples where reliance was placed on Generative AI generated content have highlighted the risks and the consequences when independent checking and verification are not undertaken.  One involved two lawyers in the US where closing submissions referred to Chat GPT generated cases that did not exist, and the other involved Australian academics using Google Bard which generated false allegations of misconduct against the big four consultancy firms in a submission to a parliamentary inquiry.  We asked Chat GPT to prepare a report on these examples and the risks, which resulted in a typically generic response – you can read what ChatGPT said and how it says the risks can be mitigated in its article which it titled:

The Dark Side of ChatGPT: Unravelling the Misuse and Its Consequences

Introduction:

The advent of artificial intelligence has ushered in a new era of possibilities, but it is not without its challenges. One such challenge is the misuse of powerful language models like ChatGPT, which can have significant consequences when inaccurate information is disseminated in the public domain. This article delves into instances where reliance on ChatGPT has led to the propagation of misleading information, illustrating the potential risks associated with the misuse of this advanced technology.

1. The Promise and Perils of ChatGPT:

OpenAI’s ChatGPT is a groundbreaking language model that uses deep learning to generate human-like text based on the input it receives. While its capabilities are impressive, there is an inherent risk in relying on such models without proper oversight and discernment. The model’s responses are generated based on patterns learned from vast amounts of internet text, and it may not always distinguish between accurate and misleading information.

2. The Rise of Misinformation:

Media outlets, researchers, and even individuals have increasingly turned to ChatGPT for information and insights. However, this reliance has not been without its pitfalls. In several instances, the unchecked use of ChatGPT has resulted in the dissemination of inaccurate information, leading to confusion and potential harm.

3. Case Studies:

a. Legal Consequences: Lawyer Used ChatGPT in Court and Cited Fake Cases

A lawyer’s misuse of ChatGPT in court has brought attention to the potential legal consequences of relying on AI-generated content. The lawyer cited fake cases generated by ChatGPT, leading to a scandal that garnered coverage from Forbes Australia, CNN Business, and The New York Times. Two US lawyers faced penalties for submitting fake court citations from ChatGPT, as reported by The Guardian and other reputable sources.

b. Academics Apologize for False AI-Generated Allegations Against Big Four Consultancy Firms

In another instance, Australian academics found themselves embroiled in a generative AI scandal when false allegations against big four consultancy firms surfaced. The academics apologized for the misinformation, which was generated by AI tools, and the incident was covered by various media outlets, including The Guardian and Cryptopolitan.

4. Addressing the Issue:

To mitigate the risks associated with the misuse of ChatGPT, it is crucial to implement safeguards and promote responsible use. Users should approach the information generated by the model with a critical mindset, fact-checking and cross-referencing the information before accepting it as accurate. Additionally, developers and organizations employing ChatGPT should work towards enhancing the model’s accuracy and providing clear guidelines on its limitations.

Key Ways to Mitigate Risks:

1. Enhanced Training Data: Continuously update and improve the training data to ensure the model learns from diverse and accurate sources, minimizing the potential for biased or misleading information.
2. Human Oversight: Implement robust human oversight to review and filter the model’s outputs, identifying and correcting inaccuracies before dissemination.
3. Transparency: OpenAI and other developers should be transparent about the limitations of their models, providing clear guidelines on areas where the model may not be reliable.
4. User Education: Educate users about the capabilities and limitations of ChatGPT, encouraging critical thinking and fact-checking when relying on information generated by the model.
5. Feedback Loops: Establish mechanisms for users to provide feedback on inaccurate outputs, enabling developers to continuously refine and improve the model.

Conclusion:

While ChatGPT holds immense potential for positive applications, the instances of its misuse highlight the need for caution and responsibility. As society continues to integrate advanced AI technologies into various facets of life, it is essential to recognize the risks and take proactive measures to ensure that misinformation does not become an unintended consequence of progress. Responsible use, rigorous oversight, and ongoing improvements to AI models are crucial steps in harnessing the benefits of technology while minimizing its potential negative impact on public discourse and decision-making.