Problems in AI, their roots in philosophy, and implications for science and society

Abstract: AI is one of today's most relevant emergent technologies. In view thereof, this paper proposes that more attention should be paid to the philosophical aspects of AI technology and its use. It is argued that this deficit is generally combined with philosophical misconceptions about the growth of knowledge. To identify these misconceptions, reference is made to the ideas of the philosopher of science Karl Popper and the physicist David Deutsch. The works of both thinkers aim against mistaken theories of knowledge, such as inductivism, empiricism, and instrumentalism. This paper shows that these theories bear similarities to how current AI technology operates. It also shows that these theories are very much alive in the (public) discourse on AI, often called Bayesianism. In line with Popper and Deutsch, it is proposed that all these theories are based on mistaken philosophies of knowledge. This includes an analysis of the implications of these mistaken philosophies for the use of AI in science and society, including some of the likely problem situations that will arise. This paper finally provides a realistic outlook on AGI and three propositions on A(G)I and philosophy (i.e., epistemology).

• The paper critiques current AI methodologies by exposing the reliance on inductivism, empiricism, and Bayesian inference, arguing these hinder true scientific explanation.
• It demonstrates how flawed philosophical doctrines misdirect AI’s capabilities, limiting its role in generating explanatory and innovative scientific knowledge.
• The study urges policymakers to adopt rigorous epistemological standards to ensure AI functions as an instrumental support for human inquiry.

Detailed Critique of "Problems in AI, their roots in philosophy, and implications for science and society"

The paper "Problems in AI, their roots in philosophy, and implications for science and society" by M.J. Velthoven and E.J. Marcus examines the philosophical shortcomings surrounding AI technologies with a particular focus on epistemological issues as highlighted by Karl Popper and David Deutsch. By anchoring its arguments in critical rationalism, the paper challenges the philosophical foundations upon which many current AI methodologies operate.

Philosophical Examination of AI Technologies

The authors assert that current AI methodologies often align with mistaken philosophical doctrines such as inductivism, empiricism, and instrumentalism. These doctrines, critiqued by Popper and Deutsch, are argued to parallel how AI technologies attempt to deduce general rules from specific data instances—an approach deemed flawed in generating true knowledge.

Popper's falsification principle—which posits that scientific theories cannot be definitively proven but only refuted—serves as a key theoretical underpinning of the authors' critique. This principle opposes the foundational assumptions of many AI algorithms, which often adjust belief probabilities based on data instances, much like inductivist approaches. The paper posits that this methodological reliance casts AI in a role they term 'explanationless', unable to generate the type of new knowledge that human creativity and scientific inquiry strive for.

AI, Bayesianism, and Instrumentalism

A significant portion of the critique focuses on Bayesianism, which is utilized frequently in AI for probabilistic analysis and decision-making. By positioning Bayesian inference as a modern cousin to inductivism, the paper argues that AI systems leverage Bayesian approaches to assign or update probability distributions in a way similar to defective knowledge growth models.

Moreover, the authors argue that instrumentalism, which values results over process, manifests in AI development practices. In contexts like quantum theory and AI application, the focus on predictive modeling without in-depth understanding or deeper explanations illustrates an 'explanationless' stance, contrary to scientific inquiry emphasized by Popperian epistemology.

Implications for Science and Society

The paper illuminates the risks posed by uncritical acceptance of AI technologies based on these flawed epistemologies. By treating AI as a mere tool rather than a creator of knowledge, the authors caution against misinterpretations in scientific and policy frameworks. They urge policymakers to recognize the instrumental nature of AI—focusing on its role as an aid, not a replacement, for human intellect—and to implement practices ensuring AI application aligns with a deep understanding of its philosophical limitations.

Challenges of Achieving AGI

In examining AGI, the authors argue that current trends in AI development, underpinned by mistaken philosophical premises, mislead expectations around reaching AGI. They claim that without breakthroughs in philosophical understanding—particularly regarding how explanations are generated and critiqued—AGI remains an unattainable goal. The authors argue that this necessitates a reevaluation of philosophical perspectives in AI research to lay the groundwork for potentially achieving AGI in the future, thus challenging arguments that posit sheer increases in computational power and data availability will suffice.

Conclusion

The paper offers a critical perspective on AI technologies through the lens of epistemological analysis. By aligning AI's functioning with flawed philosophical methods, the authors underscore the importance of philosophical awareness in the development, application, and governance of AI technologies. The authors propose that maintaining human oversight and responsibility is crucial and that AI’s true potential lies in its role as an instrumental support—rather than a replacement—in the expansion of human knowledge and society.