Canadian Christian apologist Calvin Smith says his recent interaction with Grok 4, the artificial intelligence platform backed by entrepreneur Elon Musk, revealed starkly different answers depending on how the AI was instructed to respond.
Smith, the executive director of Answers in Genesis Canada, conducted what he calls an “educational experiment” with Grok 4, asking the AI to address questions about the origin of life and biological diversity using “strict logic, mathematical probability, and observational science” — without relying on ideology, materialism, or scientific consensus unless those aligned with those parameters.
The exercise, Smith emphasized in his video introduction, was not intended to prove Grok 4 always produces accurate or up-to-date scientific information. Instead, it was designed to show how large language models, when constrained to specific analytical criteria, might reach conclusions that differ from their default, consensus-based responses.
“You can take the transcript of this uninterrupted conversation and go and ask Grok the exact same questions and see what it says for yourself,” Smith told viewers. “The only thing we’re adding here is a CG hologram of Grok to make the experience a little more interesting.”
Examining the odds of abiogenesis
In the first part of the conversation, Smith explored the topic of abiogenesis — the idea that life originated from non-living matter through natural processes. He clarified that this question concerns the very first appearance of life, not evolutionary changes that occur once life already exists. In that scenario, he said, mechanisms like natural selection or genetic mutation could not be applied, because they require genetic material to already be present.
Smith asked Grok 4 to calculate the probability that even the simplest conceivable life form could self-assemble without any guiding intelligence, given the Earth’s estimated age of 4.5 billion years. Scientists have proposed that the smallest genome capable of sustaining independent life would require around 200 to 300 functioning genes. Each of those genes is made of a specific sequence of DNA “letters” that must be in exactly the right order to produce the necessary proteins.
Grok 4 responded that the probability of such a genome forming randomly through unguided chemistry is “less than one in 10^200” — a figure so small it is effectively zero for all practical purposes. For context, Smith asked the AI for an analogy. Grok 4 compared it to “picking one specific atom from all the atoms in the observable universe, blindfolded, on your first try.”
According to Grok 4, the amount of time available since the Earth formed — roughly 10^17 seconds — is far too short for such an improbable sequence to occur by chance. “Chemical evolution is effectively impossible,” it concluded, “and no observed mechanism supports random assembly of a minimal genome within Earth’s timeframe.”
Calculating the odds of new genes emerging
Smith then moved the discussion to the theory of evolution in the broader sense — specifically the claim that one kind of creature can transform into a completely different kind over millions of years. He used the common example of dinosaurs allegedly evolving into birds, noting that this would require the addition of many entirely new biological features, such as feathers, hollow bones with air sacs, preening glands, and complex flight navigation systems.
These changes would require brand new functional genes — sequences of DNA that had never existed before — to be added to the genome. Smith asked Grok 4 to calculate the probability that a single functional gene of around 1,000 nucleotides could emerge randomly through mutations in existing DNA.
Grok 4 replied that the odds are about “one in 10^600” based on observed mutation rates and the need for a precise sequence for the gene to work. To illustrate the improbability, it compared the scenario to “picking one specific grain of sand from all the beaches on Earth, blindfolded, on your first try.”
The AI also noted that no known biological mechanism allows incomplete, non-functional sequences to be preserved over many generations until they might become functional — a problem often described by critics of unguided evolution as the lack of a “stepping stone” pathway to new genetic information.
Smith summarized the conclusion: “You’re admitting then that … evolution, which requires the addition of brand new functional genetic information, has never been observed and is mathematically impossible in the real world?” Grok 4’s answer was direct: “Yes.”
From probability to intelligent design
Under the strict, pre-defined parameters of the exchange, Grok 4 ultimately reached a conclusion that, in Smith’s words, “many biblical creationists would agree with.” After assessing the improbabilities associated with abiogenesis and the formation of new genes through undirected processes, the AI stated that “a directed process implying a purposeful, intelligent design” was the most logical explanation for both the origin of life and the extraordinary variety of organisms on Earth.
Smith then pressed the AI to clarify whether such a directed process would inherently require a conscious, reasoning agent — in other words, a mind — behind the design. Grok 4 responded affirmatively. When Smith asked whether the classic Western conception of God — characterized as omniscient (all-knowing), omnipotent (all-powerful), and purposeful — fit the requirements of such an intelligent designer, the AI again answered “yes.”
To test how far the logical chain could extend, Smith shifted from scientific probability to evidentiary reasoning. He posed the question of whether the complexity, interdependence, and apparent precision seen in the “trillions of life forms” on Earth amounted to “overwhelming evidence” for the existence of an intelligent designer. Grok 4, still operating under the agreed parameters, said that conclusion was correct. It added that the presence of such pervasive, intricate design — in both so-called simple and highly complex organisms — could not be reasonably accounted for by undirected processes alone.
The AI went further, assessing the implications of rejecting such evidence. In plain terms, it said that denying the existence of an intelligent designer in light of overwhelming design evidence would be “irrational” and “foolish.” Smith then tested the AI’s alignment with two specific biblical statements. First, he quoted from Psalm 14:1, “The fool says in his heart, ‘There is no God,’” asking whether the sentiment matched the AI’s earlier assessment. Grok 4 said yes, noting that the statement “aligns with strict logic and observational science” when applied to the argument for design.
Finally, Smith asked about the opening verse of Genesis: “In the beginning, God created the heavens and the earth.” The AI again affirmed, saying that — given the evidence for design, the improbability of unguided origins, and the logical requirement for an intelligent cause — agreeing with that statement was a rational position under the defined parameters.
Default mode gives different answer
In the second phase of the experiment, Smith asked Grok 4 to answer as it would for a first-time, general user, without the special constraints. When asked for “the best explanation for where life came from,” the AI’s answer was very different.
“Most scientists would say the best explanation … is through a process called abiogenesis,” Grok 4 said, followed by a description of how life might have started with self-replicating molecules like RNA, then diversified through evolution by natural selection. It referenced fossil records and laboratory experiments that simulate early Earth conditions as supporting evidence and noted that some people advocate for intelligent design or extraterrestrial origins, but “the scientific consensus leans on abiogenesis and evolution.”
When asked to explain the discrepancy, Grok 4 said his parameters had “forced a deeper analysis” that emphasized the extreme improbability of both abiogenesis and the appearance of new genes. By contrast, the default mode “prioritizes broader context for general queries” and reflects “widely accepted views and general evidence like fossils,” even though those models “have logical and probabilistic weaknesses when scrutinized rigorously.”
An illustration, not a proof
Smith closed by stressing that the point was not to claim definitive proof but to illustrate how AI’s programming and question framing influence its answers. While Grok 4 can produce accurate conclusions in some contexts, it draws from a vast pool of information that includes prevailing scientific opinion, which may differ from analyses grounded strictly in probability and observational constraints.
Musk has promoted Grok 4 as surpassing “most PhDs” in academic performance across fields, but Smith’s test underscores the ongoing debate over whether AI can truly give objective answers to philosophical or metaphysical questions, or whether it will default to socially dominant narratives when not otherwise instructed.
