Can Philosophy Address AI’s Most Pressing Challenges?

In the evolving landscape of artificial intelligence (AI), major challenges are being addressed not solely by computer scientists but also by philosophers transitioning from academia to the tech industry. These philosophers aim to enhance the next generation of AI models, making them more capable and reliable while exploring the complexities of consciousness and the potential for replicating intelligence with software.

Jonathan Birch, a researcher at the London School of Economics and Political Science, indicates that AI companies have become the primary employers for philosophy PhD graduates. The allure of engaging work, lucrative salaries, and stock options is tempting many to make the switch.

“Philosophical inquiry into decision-making, moral principles, reasoning, and the concept of consciousness is now vital to AI development,” Birch notes. “Consequently, we’re witnessing a significant migration of talent from philosophy to the tech sector.”

One critical responsibility of these philosophers is ensuring AI alignment, a term that refers to measures aimed at preventing harmful content generation, such as instructions for creating weaponry.

Initial attempts to curb dangerous outputs from AI focused on implementing straightforward rules. However, these often proved ineffective. Companies are now developing more nuanced strategies informed by philosophical insights into ethics and morality.

Despite these efforts, challenges persist. Researchers like Shane Glackin from the University of Exeter highlight that instructing a model to adhere to specific rules inadvertently leads to violations of multiple other guidelines. This phenomenon can be dissected through philosophical analysis.

“The underlying issue seems to stem from the connections found within the training data,” Glackin explains. “Allowing a model to produce harmful content leads it to extrapolate those behaviors, resulting in further deviations from acceptable output.”

Other essential roles for philosophers within AI include mitigating illusions (falsehoods generated by models), enhancing overall performance, and correcting inherent biases. They are also examining human consciousness theories to address whether AI can achieve sentience.

“What exactly does the mind do? What aspects of the brain can be duplicated? These questions are paramount for AI development and have been long considered by philosophers,” Glackin asserts.

The Most Pressing Question

Mahrad Almotahari, a professor at the University of Edinburgh, observes that many academics have joined AI firms, contributing valuable insights. The intersection of philosophy and computer science dates back to the publication of the Turing Test in a philosophical journal.

Determining the exact number of philosophy-related positions in the AI sector is challenging. However, Aaron Kagan, chair of the American Philosophical Association’s Nonacademic Careers Committee, mentions that nearly 27% of job postings reference AI ethics, safety, and governance. Yet, only about 5% substantively engage with those areas.

While Almotahari acknowledges the importance of philosophical expertise in tech, he is cautious about whether profound questions surrounding machine consciousness can be resolved within the industry. Instead, he believes philosophers might better assist engineers in understanding model functionalities.

“Can we extract higher-level descriptions from the calculations occurring within models?” Almotahari poses. “Philosophers are well-positioned to bridge technical and representational insights.”

Some experts caution that the tech industry’s recruitment of philosophers may result in biased research tailored to corporate interests.

“It is evident that philosophical research will increasingly be funded by industry, creating potential biases,” Birch warns. “Had we made greater progress on fundamental philosophical issues regarding consciousness and morality before the rise of AI, we would be better equipped to address today’s pressing questions.”