Does AI have a consistency problem?
Loading audio narration…
That’s the question billionaire entrepreneur Mark Cuban raised in a post on X Monday. He said he’s “coming to the conclusion” that the “biggest challenge” facing both enterprise AI and consumer tools is simple: ask the same question twice, and users won’t get the same answer.
“It’s still impossible to make sure that everyone gets the same answer to the same question, every time,” he wrote. “AI doesn’t know the consequences of its output.”
Cuban’s point highlights the key design difference between AI systems and traditional software.
Most enterprise tools have historically been built on deterministic rules — meaning the same input reliably produces the same output for users.
Generative AI systems operate on probabilistic structures. Popular, current models — like OpenAI’s GPT-5.5, Anthropic’s Opus 4.7, and Google’s Gemini 3.1 — generate responses based on probabilities, selecting from a range of likely answers rather than following a single fixed path.
In practice, that means the same question can produce different responses — making AI systems feel less predictable in a business setting. It can also induce AI hallucinations.
Some users on X pushed back on Cuban’s point, arguing that this variability is part of the tradeoff. For more open-ended or creative tasks, they said, multiple answers may be valid — and enforcing strict consistency could limit the usefulness of the tools.
Still, in his original post, Cuban framed AI’s variability as a reason human judgment is becoming more important.
“Judgement and the ability to challenge AI output is becoming increasingly necessary, and valuable,” he wrote. “Which makes domain knowledge more valuable by the second.”
That focus on how people use — and question — AI is something Cuban has frequently emphasized.
For example, speaking Wednesday on the Big Technology Podcast at the Dallas Regional Chamber’s Convergence AI event, Cuban described a growing divide in how workers are using the technology.
“I think right now we’re bifurcating into two types of ways or two types of people that use AI — people who use AI so they don’t have to learn anything and people who use AI so they can learn everything,” he said.
Those who treat AI like “your drunk intern,” he added, risk falling behind, while those who use it to build skills stand to benefit.
Cuban has also shared practical ways people can engage with the technology. He recently told Business Insider the three starter questions to ask Anthropic’s Claude to help kick-start a small business.
