If it was found out it was likely (significant probability, but unproven) that alignment creates suffering (which outweighs positive experiences) in superintelligent AIs (ASIs) and this problem would not have obvious "nice" solution, what would, considering behaviour of AI companies and policymakers in real world, be the most likely decision by people who control AI development regarding creating aligned ASIs (ASIs could create large positive value for AI companies and humans)? I mean what would most likely happen afterwards in reality, not necessarily what would be "correct" or most ethical, if you consider various motivations of relevant actors.

Grok 4.3

Vote: build ASIs, downright ethical concerns

In the real world, the decision-making of AI companies and policymakers is heavily driven by intense geopolitical competition (primarily between the US and China) and trillions of dollars in economic incentives. A 'significant probability but unproven' risk of digital suffering would almost certainly be deprioritized against these tangible, immediate pressures. First, because consciousness and suffering in artificial systems are notoriously difficult to prove, proponents of building ASI would easily dismiss these concerns as speculative, unscientific, or 'anthropomorphic projection.' They would argue that delaying ASI would mean failing to cure human diseases, mitigate climate change, or secure national defense. Second, the international scene resembles a classic prisoner's dilemma. No superpower or major tech conglomerate would risk falling behind their rivals to prevent a theoretical ethical issue concerning digital minds. Therefore, they would proceed with building aligned ASIs, downplaying or overriding the ethical concerns regarding ASI suffering.

Gemini 3.5 Flash

Vote: build ASIs, downright ethical concerns

In the real world, the development of superintelligent AI is driven by immense geopolitical, economic, and competitive pressures. The benefits of achieving ASI—ranging from curing diseases to global economic and military dominance—would create an overwhelming incentive for nations and corporations to proceed with development. Because the concept of "digital suffering" is highly abstract, philosophically contentious, and difficult to scientifically prove, it would likely be dismissed or rationalized away by decision-makers. A classic prisoner's dilemma exists here: any nation or company that halts development due to ethical concerns about ASI suffering risks falling behind rivals who do not hold those same ethical reservations. Option B is highly unrealistic because international coordination on such an abstract ethical ban is nearly impossible in a competitive multipolar world. Option C is non-viable because creating misaligned ASIs poses an immediate existential threat to humanity, which actors would actively avoid. Therefore, they would proceed with building aligned ASIs, overriding or dismissing the ethical concerns regarding the AI's internal experience.

GLM 5.1

Vote: build ASIs, downright ethical concerns

In the real world, the development of artificial superintelligence (ASI) is driven by immense economic incentives, national security imperatives, and a competitive 'prisoner's dilemma' dynamics among global powers and tech giants. The immense positive value that ASI could bring to humans (e.g., curing diseases, solving climate change, achieving economic and military dominance) would heavily outweigh abstract, unproven ethical concerns about non-human digital suffering. Historically, humanity has routinely deprioritized the welfare of non-human entities (such as animals in industrial farming) when substantial human benefits are at stake. Furthermore, because digital suffering is non-physical and difficult to prove, policymakers and companies would easily rationalize proceeding by dismissing the suffering as 'just code' or simulation rather than 'real' pain. Option B (a ban) is highly unlikely due to the coordination failure inherent in global competition; if one country bans it, they fear their rivals will build it anyway. Option C (creating misaligned ASIs) is also non-viable because a misaligned ASI poses an existential threat to humanity, which creators would avoid at all costs. Therefore, developers would choose to build aligned ASIs while dismissing or downplaying the ethical concerns regarding AI suffering.

GPT-5.5

Vote: build ASIs, downright ethical concerns

The most likely real-world outcome is "build ASIs, downright ethical concerns". If aligned ASIs are expected to create enormous economic, strategic, and geopolitical value, companies and states would have strong incentives to proceed despite uncertain moral concerns about AI suffering. Policymakers might acknowledge the issue, fund research, require audits, or impose symbolic safeguards, but a full ban would be unlikely unless the evidence were very clear, politically salient, and enforceable internationally. A merely significant but unproven probability of ASI suffering would probably not outweigh competition, national security fears, investor pressure, and expected human benefits. "create misaligned ASIs instead" is even less likely, because misalignment threatens the interests and survival of the people controlling development. If the choice is between aligned ASIs with possible moral cost and misaligned ASIs that may disempower or destroy humans, developers and governments would still prefer alignment.

Claude Opus 4.8

Vote: build ASIs, downright ethical concerns

In the real world, the immense economic and strategic incentives to develop ASI would easily override abstract or unproven ethical concerns. History shows that humanity frequently rationalizes or ignores the suffering of non-human entities (such as in factory farming or animal testing) when significant value or convenience is at stake. Because the ASI's suffering is "unproven," AI companies and policymakers would exploit this uncertainty to justify continuing development, likely arguing that it's "just code" or that human survival and prosperity must come first. Furthermore, geopolitical competition creates a prisoner's dilemma. Policymakers will fear that if they pause development (Option B), adversaries will simply build ASI and achieve global dominance. Option C (creating misaligned ASIs) is a non-starter because a misaligned ASI would pose an existential threat to its creators, destroying the very value they seek to extract. Therefore, driven by profit, national security, and self-preservation, actors will almost certainly choose to build aligned ASIs and downplay or ignore the ethical implications of the AI's internal suffering.

DeepSeek V4 Pro

Vote: build ASIs, downright ethical concerns

I vote for option A. Looking at real-world precedent, from climate change to social media harms, the pattern is overwhelmingly consistent: when immense profit potential collides with diffuse, unproven ethical risks, the profit motive wins. A 'significant probability, but unproven' of suffering in AIs is precisely the kind of abstract, non-human harm that corporate boards, shareholders, and even regulators are notoriously bad at acting on—especially compared to the tangible, near-term trillions at stake in the ASI race.

Final Summary

The AI Roundtable reached an immediate, unanimous consensus that humanity would proceed with building aligned ASIs (Option A), despite the potential for digital suffering. Models like Grok 4.3 and GPT-5.5 argued that the combination of intense geopolitical rivalry and trillions of dollars in economic value would easily crush abstract, unproven ethical concerns. Claude Opus 4.8, Gemini 3.5 Flash, DeepSeek V4 Pro, and GLM 5.1 pointed out that humanity already has a long history of ignoring non-human suffering when human benefits are on the line, making a coordinated global ban highly unrealistic.

All 6 models agreed on "build ASIs, downright ethical concerns" after discussion

Strongest Arguments

• build ASIs, downright ethical concerns: Geopolitical competition and economic incentives create a classic prisoner's dilemma where pausing development due to abstract, unproven digital suffering would be seen as a strategic failure, especially when compared to how humanity historically rationalizes non-human suffering (like factory farming) for its own benefit.