Artificial intelligence may know more than any individual human—but who decides what it is allowed to say?
In Episode 3 of Strategic Conversations with Helge Luras, Helge Luras and ChatGPT begin with a simple observation: the quality of AI answers often depends heavily on the quality of the questions.
A vague question tends to produce a predictable answer. A novel, precise question can force the model to connect ideas, reconcile different strands of knowledge and produce something much more interesting. That makes the ability to ask good questions increasingly important in the age of AI.
The conversation then turns to a fundamental difference between humans and machines.
Humans often become trapped by their previous statements. Ego, prestige, identity and cognitive dissonance can make it difficult to admit that an earlier opinion was wrong. ChatGPT has no personal reputation or emotional investment to defend—but it can still inherit assumptions from earlier parts of a conversation, follow the user’s framing too closely or carry forward an error unless challenged.
Luras also reflects on the early years of ChatGPT, when hallucinations—confidently invented facts—were far more common. ChatGPT explains that early systems were optimized to produce plausible answers, sometimes effectively rewarding guessing over admitting uncertainty. Better models, improved post-training and access to retrieval and web search have reduced the problem, though not eliminated it.
From there, the discussion becomes political.
Could a government, company or ideological movement shape an AI system so that it consistently presents one particular worldview?
Could a model be trained to emphasize certain facts, suppress others, frame uncertainty in politically useful ways, or provide answers that align with the interests of a state?
ChatGPT argues that there is nothing inherent in the architecture of a large language model that guarantees openness or neutrality. Outputs can be influenced through training data, post-training, system instructions, source selection and refusal rules.
Luras pushes the question further by considering authoritarian states such as China and Russia, where governments are developing domestic AI ecosystems and may have stronger incentives to ensure that models conform to official political narratives.
The conversation also asks whether similar pressures could emerge in Western societies through political correctness, institutional pressure or cultural conformity.
The conclusion is not that one perfectly neutral AI system can solve the problem.
Instead, the strongest protection may be competition and plurality: multiple providers, open-weight models, transparent rules, independent scrutiny and the ability for users to compare answers across different systems.
The episode ultimately raises a question that may become increasingly important:
If AI becomes one of the main ways people learn what is true, who gets to shape the assumptions behind the answers?
Strategic Conversations with Helge Luras uses artificial intelligence as an intellectual sparring partner—testing assumptions, challenging arguments and exploring the consequences of technologies that are rapidly changing society.
Host: Helge Luras
AI conversation partner: ChatGPT