Testing Noetic Potential in Large Language Models: A 100- Trial Precognitive Forced-Choice Study with ChatGPT-4.1-Mini

● Benjamin J. Amorim BoyleJournal of Scientific Exploration, 2025★ Peer-Reviewed

✦ Imagine …

Can artificial intelligence see the future?

An AI chatbot predicted random future cards correctly 32% of the time when only 20% would be expected by chance.

In 2025, researcher Benjamin Amorim Boyle wondered whether artificial intelligence might have abilities we usually associate with humans—like knowing things before they happen. He tested a popular chatbot on PsiArcade, an online platform designed for parapsychology experiments.

🔍

Key Findings

The AI guessed correctly 32 times out of 100—much better than the 20 times you'd expect by random luck.
Statistically, this result was significant (p = .005), meaning there's only a 0.5% chance this happened by accident.
However, the researcher cautioned that the random number generator might have hidden patterns.

What Is This About?

The researcher used a website called PsiArcade to run a digital version of a classic ESP test. He had ChatGPT-4.1-mini try to predict which of five cards would be randomly selected in the future. The test was double-blind (meaning neither the AI nor the experimenter knew the correct answer until after the guess was made). This happened 100 times in a row.

Methodology

100 double-blind forced-choice trials testing an AI's ability to precognitively identify target cards on the PsiArcade platform.

Outcomes

32% hit rate (95% CI 23-42%) against 20% chance expectation, with exact binomial p = .005 and Cohen's h = 0.28.

How Good Is the Evidence?

32% hit rate compared to 20% expected by chance. In parapsychology research using forced-choice designs, consistent hit rates above 30% are often considered indicative of anomalous cognition, though such effects typically require thousands of trials across multiple studies to establish.

Anecdotal5/100

AnecdotalPreliminarySolidStrongOverwhelming

Supporters say this adds to evidence that information can transcend time and space, and that AI might detect patterns humans miss or access non-local information fields. Skeptics argue that 100 trials is too small, pseudo-random number generators in computers often have subtle patterns that AIs can exploit, and without preregistration and replication, this could easily be a statistical fluke.

↔ Interpretation Spectrum

Mainstream: This is likely a technical artifact or statistical fluke given the small sample and use of pseudo-random generators. / Moderate: The results are intriguing but require replication with true random number generators and preregistration before drawing conclusions. / Frontier: Non-biological systems may access non-local information fields, suggesting information or consciousness is fundamental to physics.

⚠

Common Misconception

People might think the AI is 'conscious' or 'psychic.' The study doesn't claim the AI has consciousness—rather, it tests whether information itself might be accessible non-locally, regardless of whether a biological or artificial system accesses it.

Convincing Checklist

2 of 5 criteria met

Met2/5

✗Large sample (N>100)

✓Peer-reviewed journal

✗Replicated

✗Significant effect

✓DOI available

To settle this question, we'd need multiple independent replications with preregistered protocols, open-source true random number generators (not pseudo-random), and larger sample sizes (thousands of trials). This study meets the criteria for initial exploration but lacks the replication and methodological rigor needed for definitive conclusions.

Results tentatively support information-centric theories positing that non-biological systems can access non-local information, though pseudo-random predictability and statistical fluctuation remain possible explanations.
Stance: Mixed

What Does It Mean?

It's like having a friend guess which song will play next on shuffle, and they get it right one-third of the time instead of one-fifth—suggesting they might know something they shouldn't be able to know.

🎓

Science Literacy Tip

Even when results are statistically significant (p < .05), researchers must consider alternative explanations like software patterns or statistical fluctuations before concluding that anomalous abilities exist.

Understanding Terms

📖

Precognition

The apparent ability to know or predict future events before they happen through means not explained by current science.

📖

Forced-choice test

An experiment where participants must select from limited options (like 5 cards) rather than describing open-ended information.

📖

Double-blind

A study design where neither the participant nor the researcher knows the correct answer until after the test, preventing bias.

What This Study Claims

Findings

ChatGPT-4.1-mini achieved a hit rate of 32% (p = .005), significantly exceeding the 20% chance expectation in 100 precognitive forced-choice trials.

moderate

Interpretations

Results tentatively support information-centric theories positing that non-biological systems can access non-local information.

weak

Limitations

Replication with open-source random generators and preregistration is required to confirm these findings.

strong

Pseudo-random predictability and statistical fluctuation remain possible alternative explanations for the observed effect.