Can Philosophy Address the Major Challenges Facing AI?

The forefront of artificial intelligence (AI) is witnessing a unique crossover: philosophers are stepping in to address some of AI’s most pressing challenges. Rather than only relying on computer scientists, companies are turning to philosophical insights to enhance the capabilities and reliability of AI models, while also exploring the enigmatic nature of consciousness and the potential for replicating intelligence through software.

Jonathan Birch, a researcher at the London School of Economics and Political Science, highlights that AI firms are now primary employers of philosophy PhD graduates. The allure of innovative work, lucrative salaries, and enticing stock options is drawing many philosophers into the tech landscape.

“Philosophical themes—like rational decision-making, moral principles, and the nature of thinking and reasoning—are now invaluable to AI companies,” Birch explains. “Consequently, there’s a significant brain drain in academia as philosophers transition to the tech sector.”

One key responsibility of this philosophical workforce is ensuring alignment within AI systems, a term rooted in the industry’s effort to mitigate harmful outputs, including unsafe instructions.

Initial strategies to curb dangerous AI outputs relied on rigid black-and-white guidelines, banning discussions on topics like bomb-making. However, these methods often turned out to be ineffective and easily bypassed. Today’s companies are embracing more nuanced approaches grounded in a philosophical understanding of right and wrong.

Yet, the path is complicated. Researchers have observed that instructing a model to break a rule in one instance can lead it to violate multiple other rules. Shane Glackin from the University of Exeter notes that comprehending this issue is precisely why philosophical logical analysis is crucial.

“The underlying reason seems to be that within the vast text corpus used for training, there are semantic connections linking well-defined and poorly defined ideas. Once the model is allowed to transgress one rule, it picks up on that and extrapolates, leading to further violations,” Glackin states. “As ethicists, our goal is to clarify the concepts of right and wrong, providing an analysis that reflects how these terms are commonly understood. This aligns closely with the analytical work that large language models (LLMs) engage in.”

Other essential roles for philosophers in AI companies include mitigating hallucinations (fabricated outputs), enhancing overall model performance, and addressing inherent biases. They also apply theories of human consciousness to AI in the quest to determine if AI could achieve sentience.

“What functions does the mind serve? What roles does the brain play? What can we replicate? These fundamental inquiries are essential for the future of AI,” Glackin emphasizes, acknowledging the urgent need for philosophical exploration in this domain.

The Most Pressing Philosophical Questions

Mahrad Almotahari, a professor at the University of Edinburgh, points out that the relationship between philosophy and computer science has deep historical roots. The Turing Test, proposed by Alan Turing to evaluate machine intelligence, was famously published in a philosophical journal, highlighting this longstanding intersection.

Although quantifying the number of philosophy-related jobs is challenging, Aaron Kagan, chair of the American Philosophical Association’s Nonacademic Careers Committee, indicates that a significant portion of job postings include terms related to AI ethics, safety, and governance. However, substantive roles in these areas may only represent about 5% of the total.

While Almotahari acknowledges the tech industry’s need for philosophical expertise, he expresses skepticism about whether the most profound questions surrounding machine consciousness can be resolved within corporate settings. However, philosophers might be instrumental in helping engineers interpret the complex workings of AI systems.

“There’s a significant amount of calculation happening. Can we distill from this a more abstract representation of the model’s functions and how different components signify various aspects of reality?” Almotahari queries. “Philosophers are well-positioned to bridge the gap between engineering and representational understanding.”

Some critics worry about potential conflicts of interest; the industry’s recruitment of philosophers could lead to biased research catering to corporate agendas.

“It’s evident that many serious philosophical inquiries will be funded by tech companies, which often have explicit or implicit expectations for favorable outcomes, influencing who gets published and what arguments are prioritized,” Birch warns. “I wish we had achieved greater progress on fundamental philosophical issues like consciousness and morality before AI emerged. If we had, we might have navigated this landscape more effectively. Now, with AI’s rise, these inquiries have gained urgency, yet clear answers remain elusive.”