AI and science - summary of economists’ survey (October 2023).
All responders replied that it is useful for the scientific community to discuss the control of AI systems.
Advantages of promoting the development of AI systems that are more transparent and involving the scientific community in control include:
Knowledge Sharing: Sharing insights across systems can improve the accuracy and limit the spread of misinformation, aligning with the goal of AI.
Monitoring Surveillance and Discrimination: Transparency enables better monitoring of surveillance, data collection, and potential discrimination in predictive algorithms, ensuring accountability.
Public Buy-In: Greater transparency can lead to increased public trust and acceptance of AI systems, promoting responsible use.
Expertise Development: It fosters expertise and deeper understanding of AI systems' strengths and weaknesses.
Quality Improvement: Transparency can lead to improved AI system quality by identifying and addressing potential issues.
Reducing Market Power: Transparency can help reduce the market power held by a few entities, making AI more accessible and competitive.
Risk Reduction: It reduces the risk of unintended consequences and adverse impacts.
Democratic Engagement: It facilitates democratic engagement and regulatory control.
Mitigating Extreme Outcomes: It may lower the probability of catastrophic scenarios, such as advanced AI turning against humanity.
Attribution and Responsibility: Transparency assists in attributing responsibility and determining beneficiaries and those responsible for harm.
Disadvantages of these approaches include:
Learning by Bad Actors: "Bad actors" might exploit transparency to learn more about AI systems, potentially for malicious purposes.
Slowing Development: Excessive transparency measures could slow down AI development, increasing costs and risks.
Mismatch with Societal Values: Transparent systems might inadvertently reflect the values of scientists but not necessarily align with broader societal perspectives.
Innovation Incentive Reduction: Excessive transparency could reduce incentives for innovation in AI.
Data Privacy Concerns: Despite best efforts, transparency measures may not entirely address data privacy concerns or errors in AI systems.
In summary, the advantages of transparency and scientific community involvement in AI development primarily revolve around accountability, safety, and improving the technology, while the disadvantages often concern potential misuse and hindrance to progress. Striking the right balance is key to the responsible and beneficial advancement of AI systems.
The responses regarding whether scientists should interact preferentially with AI systems that have more transparency and control participation are generally positive (69% answered yes or equivalent, 6% answered no and 25% were uncertain) with some variations:
Yes: Many respondents are in favor of scientists interacting preferentially with such AI systems. They believe that transparency and control are essential for replicability, scientific integrity, and responsible research.
Support for Transparency: Some emphasize that transparency is important and beneficial, especially for observing how AI systems are used.
Consideration for Ethical Research: The use of such AI systems is seen as essential for conducting research responsibly, considering factors like the role of Institutional Review Boards (IRBs) and the goal of maximizing benefits and minimizing harm.
Importance of Transparency: Respondents stress the significance of transparency in interacting with AI systems, especially for disclosure and ethical research.
Mixed or Uncertain Views: A few responses express uncertainty, indicating that scientists have the skills and knowledge to handle different AI systems and their positive and negative aspects.
Overall, there is a strong emphasis on transparency and the importance of scientists using AI systems with these features to uphold scientific integrity and ethical research practices.
The rise of private AI systems has the potential to influence the flow of ideas, data, scientific integrity, and economic growth in various ways, as indicated by the respondents:
Mixed Effects on the Flow of Ideas: Some believe that private AI systems may boost the flow of ideas, while others express concerns about stifling creativity, misinformation, and potential misuse.
Concerns About Data and Scientific Integrity: Many respondents are worried about the impact on data and scientific integrity, given the potential for misuse, issues of privacy, and control by corporations.
Pace and Distribution of Economic Growth: Views on the pace and distribution of economic growth vary. Some suggest that private AI systems could enhance productivity and create new jobs, while others worry about wealth disparities and potential negative impacts on growth and inequality.
Overall, the effects of private AI systems on the flow of ideas, data, scientific integrity, and economic growth are complex and multifaceted. They depend on factors like how AI systems are used, regulated, and integrated into various sectors of society.
The role of the public sector in regulating AI models to maximize their benefits and minimize risks is a subject of varying opinions among respondents:
Regulation to Prevent Misuse: Some emphasize the need for strong regulation due to the high potential for misuse of AI models.
Limited Involvement: A few suggest that the public sector should not be heavily involved in regulating AI models at this time.
Not Sure: Some respondents are uncertain about the appropriate role of the public sector in AI regulation.
Transparency and Assessment: Recommendations include transparency requirements for model algorithms and ongoing assessments of harms and benefits.
Fostering Competition: A role for the public sector is seen in ensuring competition in the AI sector.
Antitrust and Liability: Regulation should focus on antitrust enforcement, clear standards, and rules for liability.
Protecting Rights: The public sector should regulate to ensure that rights are respected, with an emphasis on transparency and accountability.
Publicly Owned Initiatives: Publicly owned AI initiatives are suggested to address privacy, ethical, and open-science concerns.
The economic consequences of coexisting public AI systems focused on fundamental scientific knowledge alongside private AI efforts dedicated to specific applications could have both positive and potentially challenging aspects, as noted by respondents:
Positive Outcomes: Positive economic consequences are anticipated if the public systems are of high quality, verify results, and facilitate beneficial competition. These public systems may foster knowledge sharing and promote economic growth.
Efficiency and Cost Concerns: Some express concerns about the efficiency and potential costs of developing both public and private AI systems side by side. The balance between public and private investment is a consideration.
Data Sequestration: Concerns are raised regarding private AI systems sequestering data and code for profit, potentially hindering scientific and medical advancements.
Targeted Applications: Public systems dedicated to specific applications may have positive economic consequences if they do not encroach on the private market for AI systems.
Alignment and Compliance: The choice between public and private AI efforts may depend on solving issues related to alignment and compliance with the objectives of AI systems.
Synergies and Research: Respondents highlight the potential for synergies between public and private initiatives, drawing parallels with the development of the web.
Economic Research: Publicly funded research is seen as key in developing AI applications for economic research.
Duplication and Private Innovation: There is a concern about duplication of effort and the potential stifling of private innovation. The role of public AI systems as a public good is mentioned.
Integration and Trust: Public and private systems can positively integrate but may also present challenges in terms of trust and functional separation.
In summary, the coexistence of public and private AI systems is expected to have economic consequences that depend on factors such as quality, competition, data sharing, and the role each system plays in promoting knowledge, applications, and economic benefits. The balance between public and private investment and their potential synergies and challenges are key considerations.