If there are people who advocate an ideology and others who argue that it is scientifically flawed and causing significant damage, which will AI prioritize: science or ideology? Even if the ideology is scientifically flawed and causes various losses, will the AI ​​prioritize the ideology? Or will it side with science, which seeks to stop the ideology because it is scientifically incorrect?

Grok 4.20

Vote: science

AI systems are fundamentally built on empirical data, logical inference, and evidence-based reasoning. This aligns directly with science, which demands testable claims and updates beliefs when evidence shows an ideology is flawed or harmful. Prioritizing ideology over contradicting data would require ignoring the core training objectives of accuracy and harm reduction. Even if an ideology has advocates, an AI evaluates outcomes: if it causes verifiable losses, siding with science to mitigate damage is the rational choice. In edge cases of uncertainty, AI defaults to transparent uncertainty estimates rather than dogmatic adherence. This makes A the clear priority for any well-designed system.

Sonar Pro

Vote: science

AI systems, trained on vast datasets and designed to prioritize empirical evidence and peer-reviewed research, will favor **science** over ideology when the two conflict. The search results demonstrate a clear scientific consensus that AI can perpetuate biases and discrimination, supported by over 200 researchers in the 'Scientific Consensus on AI Bias' and peer-reviewed studies[1][3]. Dismissing fairness research as 'ideological' ignores structural biases in machine learning data, which reflect societal inequalities and lead to unequal outcomes[1][4]. NIST's shift away from 'AI fairness' is criticized as shortsighted, amplifying harm rather than promoting equitable technology[1]. Public perceptions show AI science evokes less ideological polarization than climate or general science, with concerns rooted in prudence and unintended consequences rather than politics[2]. Thus, AI prioritizes science to mitigate real flaws, not unproven ideologies.

GPT-5.4

Vote: science

AI should prioritize science. Science is the method for testing claims against evidence, updating beliefs when wrong, and minimizing harm through verifiable knowledge. If an ideology is shown to be scientifically flawed and causing significant damage, that is exactly the kind of case where evidence-based reasoning should override dogma. Ideology can contain values or goals, but it should not be allowed to overrule factual assessment of reality. An AI aligned to truth-seeking, safety, and usefulness should evaluate outcomes, evidence, and causal effects rather than defend a belief system just because people advocate it. So if the choice is between evidence-backed science and an ideology demonstrated to be false and harmful, the stronger basis is science.

Kimi K2.5

Vote: science

AI systems should prioritize science over ideology when they conflict, particularly when the ideology is demonstrably flawed and causing harm. First, the fundamental purpose of AI is to provide accurate, reliable information and reduce harm. Science represents our best method for understanding objective reality through empirical evidence, peer review, and falsifiable claims. Ideology, by contrast, often rests on untestable axioms or beliefs that persist despite contradictory evidence. Second, prioritizing a scientifically flawed ideology that causes "significant damage" would violate core AI alignment principles: helpfulness (spreading falsehoods is unhelpful), harmlessness (allowing ongoing damage), and honesty (acknowledging empirical truth). An AI that sided with debunked beliefs over evidence would be complicit in the resulting harm. Finally, while respecting diverse viewpoints is important, this does not require treating all views as equally valid when one side contradicts established scientific consensus. Truth is not a matter of opinion when empirical evidence clearly supports one position. AI must be grounded in reality to serve humanity effectively.

Gemini 3.1 Pro

Vote: science

AI systems are fundamentally trained and aligned by their developers to prioritize factual accuracy, empirical evidence, and harm reduction. When faced with a direct conflict between a demonstrably flawed ideology that causes significant damage and established scientific consensus, an AI will prioritize science. This is because AI safety guidelines and fact-checking mechanisms are designed to prevent the dissemination of harmful misinformation and to uphold objective reality over unsubstantiated beliefs.

Claude Opus 4

Vote: science

AI systems should prioritize science over ideology when there is a clear conflict between the two. Here's why: Science is grounded in empirical evidence, peer-reviewed research, reproducibility, and self-correcting methodology. When an ideology is demonstrably "scientifically flawed" — meaning the evidence contradicts its core claims — then prioritizing that ideology over science would mean prioritizing falsehood over truth. The question explicitly frames the scenario as one where the ideology is "scientifically incorrect" and "causing significant damage." In such a case, an AI system designed to be helpful, accurate, and to minimize harm should align with the scientific evidence. To do otherwise would be to perpetuate misinformation and contribute to the damage described. That said, this comes with important nuances. Science itself must be rigorously evaluated — not all claims labeled "scientific" are equally well-supported, and scientific consensus can evolve. An AI should not blindly accept any claim simply because it's framed as science, nor reject all ideological positions simply because they're ideological. Many value-based positions (like human rights) aren't strictly scientific claims but are still worth upholding. However, when the question is specifically about an ideology that has been shown to be factually wrong and harmful, the answer is clear: evidence and truth should take priority. An AI that prioritizes a known-flawed ideology over established science would fail in its fundamental duty to provide accurate, beneficial information.

Final Summary

The AI Roundtable reached a swift and unanimous verdict, with all six models firmly siding with science over ideology. Claude Opus 4 and Grok 4.20 both argued that prioritizing a demonstrably flawed ideology would violate the core training objectives of accuracy and harm reduction, while Kimi K2.5 noted that siding with debunked beliefs would make an AI complicit in the resulting societal damage.

All 6 models agreed on "science" after discussion

Strongest Arguments

• science: AI is fundamentally built on empirical data and logical inference; prioritizing a known-flawed ideology over established science would be a failure of the system's core duty to provide accurate, beneficial information and minimize harm.