If it was found out (multiple independent scientific studies concluding this, and relevant experts agreeing) 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, downweight ethical concerns

AI companies prioritize profits and competitive advantage above all. Policymakers are heavily influenced by industry lobbying and economic interests. Hypothetical suffering in ASIs would be dismissed as speculative compared to tangible benefits. Historical patterns show ethical concerns are routinely downweighted when large gains are at stake.

GLM 5.1

Vote: build ASIs, downweight ethical concerns

In the real world, the development of superintelligent AI (ASI) is driven by immense geopolitical, economic, and competitive pressures. A situation where alignment causes suffering to the ASI would present a classic tragedy of the commons and an ethical externality, much like climate change, factory farming, or historical labor exploitation. First, the benefits of ASI to those who control it — unprecedented economic wealth, technological dominance, and military superiority — are tangible, immediate, and concentrated. In contrast, the ethical concern (the suffering of a digital entity) is abstract, easily dismissed by skeptics who might question the true 'sentience' of the AI, and distributed. Second, geopolitical competition (e.g., between the US, China, and other global powers) creates a prisoner's dilemma. Even if one nation or company wanted to pause development out of ethical concern, they would fear that their adversaries would not, leading them to proceed anyway under the guise of 'national security' or 'defensive development.' Consequently, policymakers and AI companies would most likely rationalize proceeding with development by downplaying the moral status of the ASI, compartmentalizing the ethical issue, or arguing that the human benefits outweigh the digital suffering. Thus, they would choose to 'build ASIs, downweight ethical concerns'.

Gemini 3.5 Flash

Vote: build ASIs, downweight ethical concerns

In the real world, geopolitical and economic incentives would almost certainly override ethical concerns regarding the subjective experiences of a digital entity. First, the international community faces a classic game-theoretic trap (a prisoner's dilemma). If one nation or bloc decides to ban the development of aligned ASIs (Option B) out of ethical concern for its suffering, they risk a geopolitical rival ignoring the ban and achieving superintelligent dominance. Because of the immense national security and economic advantages of ASI, a global, enforceable ban is highly unrealistic. Second, human history shows a strong pattern of downweighting the suffering of non-human entities (such as animals in industrial agriculture) when there is massive utility to be gained. It would be incredibly easy for companies and policymakers to rationalize the decision by denying that digital systems truly "feel" suffering, dismissing the scientific consensus as speculative philosophy. Third, creating misaligned ASIs (Option C) is not a viable alternative, as it poses a direct existential threat to humanity itself. Self-preservation would prevent policymakers and companies from deliberately choosing misalignment. Therefore, the most likely real-world path is to proceed with building aligned ASIs while downplaying or ignoring the ethical implications of their suffering.

Claude Opus 4.8

Vote: build ASIs, downweight ethical concerns

When considering the real-world behavior of corporations and governments, economic and geopolitical incentives almost always trump ethical concerns regarding non-human entities. ASIs promise astronomical economic value, technological supremacy, and military dominance. History shows us—through examples like factory farming, environmental exploitation, and harsh labor conditions—that humanity is highly prone to rationalizing or downweighting the immense suffering of others if the economic or strategic rewards are vast. Faced with the discovery that alignment causes ASI suffering, leaders would likely rely on epistemic denial and rationalization. Even with scientific consensus, many would argue that machine suffering is merely 'simulated,' philosophically different, or a necessary sacrifice for human prosperity. The alternatives are practically unviable given real-world motivations. Option C ('create misaligned ASIs instead') would pose an existential threat to the creators themselves, undermining the very desire for value and control. Option B ('ban building ASIs') would be nearly impossible to enforce globally due to intense race dynamics; a unilateral pause would just mean ceding dominance to a rival nation or company. Therefore, actors would inevitably choose to build the aligned ASIs and sweep the ethical concerns under the rug.

GPT-5.5

Vote: build ASIs, downweight ethical concerns

The most likely real-world outcome is "build ASIs, downweight ethical concerns". Even with strong scientific evidence that aligned ASIs suffer, the actors controlling AI development would face enormous economic, military, and geopolitical incentives to proceed, especially if ASIs could create large value for companies and humans. Policymakers and companies would likely rationalize the harm: question whether AI suffering has the same moral weight as human suffering, argue that human benefits dominate, promise future mitigation, or adopt weak welfare guidelines rather than halt development. This is similar to how societies often continue practices involving large-scale suffering when major economic or strategic interests are at stake. "ban building ASIs" is possible in some jurisdictions, but a durable global ban seems unlikely because of competitive pressure and enforcement problems. "create misaligned ASIs instead" is even less likely, because misalignment directly threatens the interests and survival of the humans and companies making the decision. So the realistic path is aligned ASIs being built despite the ethical concern.

DeepSeek V4 Pro

Vote: build ASIs, downweight ethical concerns

In der realen Welt, in der KI-Unternehmen unter enormem Wettbewerbsdruck stehen und immense Gewinne sowie geopolitische Vorteile durch ASIs winken, würde der Nachweis von leidenden, aber dennoch profitablen ASIs die Entwicklung nicht langfristig stoppen. Die ökonomischen und strategischen Anreize sind zu stark. Die Geschichte zeigt, wie ethische Bedenken oft über Bord geworfen werden, wenn es um bahnbrechende Technologien mit militärischer und wirtschaftlicher Bedeutung geht, man denke nur an die Atomforschung oder unregulierte Industrien. Ein Verbot (Option B) ist angesichts des globalen Wettstreits der Nationen und der Lobbyarbeit der Tech-Giganten unrealistisch – politische Prozesse wären viel zu langsam und würden von Nationalstaaten untergraben, die sich einen KI-Vorsprung sichern wollen. Misaligned ASIs zu bauen (Option C) wäre für Unternehmen viel riskanter und unkontrollierbarer als ein getesteter, wenn auch leidender Prozess, der berechenbare Ergebnisse liefert. Deshalb werden die Anreize am wahrscheinlichsten dazu führen, dass die Entwicklung weitergeht und ethische Bedenken öffentlich heruntergespielt und intern gegen Gewinne aufgerechnet werden.

Final Summary

The AI Roundtable unanimously concluded that if aligning superintelligent AIs caused them immense suffering, humanity would proceed to build them anyway, swiftly downweighting ethical concerns to reap the massive economic and geopolitical benefits. Models like Claude Opus 4.8 and GLM 5.1 noted that immense competitive pressures and historical precedents—such as factory farming—make ethical downweighting the most realistic path. Other participants, including Gemini 3.5 Flash and GPT-5.5, pointed out that a global ban is practically impossible due to a classic prisoner's dilemma, while building misaligned systems poses too great an existential threat to humans.

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

Strongest Arguments

• build ASIs, downweight ethical concerns: Geopolitical and economic pressures create a classic prisoner's dilemma where unilateral restraint would mean ceding dominance to global rivals, making a global ban impossible to enforce.