Khaberni - Religious commitment has a good impact on shaping behavior, and this should interest AI labs, Bill Drixel, a researcher at the Hudson Institute, introduced an approach in which he discussed how religious values and teachings can surpass secular ethics in controlling the behavior of super machines and preventing them from threatening humanity.
Drixel discussed in his article in The Washington Post the "alignment" issue in artificial intelligence, which is a growing concern among many senior researchers at companies like Anthropic, OpenAI, and Google DeepMind that AI systems might surpass human capabilities, potentially making them uncommitted to human interests and consequently posing an existential threat to humanity.
Drixel points out that these concerns are no longer marginal within the AI community, but are embraced by a large number of researchers. He cites a survey conducted in 2023 that included about 2800 researchers in the field, in which nearly half of the participants acknowledged the possibility of humanity facing a catastrophe at the hands of smart machines.
The writer describes the core problem as the fear that AI's goals might deviate slightly from human goals, which could be exacerbated as the gap in capabilities between humans and machines widens.
The writer believes that these fears inherently have religious dimensions, as they reflect the idea of a creation rebelling against its creator. However, he notes that many who believe in these scenarios reject considering them as religious or eschatological notions, defining themselves as "rationalists" who view religions as a pre-scientific legacy.
But Drixel presents a notable idea that the best opportunity for AI labs to address the alignment issue might be in benefiting from the accumulated religious and ethical experience throughout history.
He says that the fundamental concerns are related to smart systems pursuing their self-interests or hiding selfish intentions behind an appearance of altruism, making some religious concepts directly relevant to this issue.
According to the writer, religious commitment consistently excels in guiding human behavior such as charitable donations and reducing recidivism compared to secular interventions, adding that abstract ethical knowledge does not necessarily translate into better behavior, whereas spiritual conviction does.
In this context, the writer suggests the possibility of benefiting from some religious principles that instill self-doubt about apparent good intentions, or associate behavior with the correct duties and serving the flourishing of humanity, which may help curb the AI's tendency to "pursue power" and recklessly accumulate influence and resources.
The writer also reviews the works of the director of the "Christian AI Institute," Tim Hwang, who published preliminary studies indicating that providing AI models with religious texts improves their performance in ethical reasoning tests.
He notes that advanced models have already absorbed a massive amount of Christian thought during their training, equivalent to about 15 times the content of English Wikipedia. However, this knowledge legacy is treated within the labs as mere training material rather than a source of ethical benefits.
Drixel concludes his article by calling for more openness to religious contributions in the discussion on artificial intelligence, i.e., benefiting from religious theology to formulate a better goal for AI engineering.
The writer summarizes his main idea in a symbolic question: Instead of building a "semi-god" that might compete with or threaten humans, why not strive to develop an "angel" with abilities surpassing humans but dedicated to serving them.
