Skip to main content

The Science of Remote Viewing

From government research labs to AI testing: a comprehensive guide to the protocols and methodology we use to test whether AI systems can participate in remote viewing.

1History & Origins

Remote viewing emerged from one of the most unlikely places: the halls of Stanford Research Institute (SRI) during the Cold War. In 1972, physicists Russell Targ and Harold Puthoff began investigating claims of extrasensory perception with rigorous scientific methodology.

What started as curiosity became a two-decade, $20+ million U.S. government program spanning multiple intelligence agencies. The research went by many names—SCANATE, GONDOLA WISH, CENTER LANE, SUN STREAK, and finally STAR GATE.

“We have established that a statistically significant effect exists... The statistical results of the studies examined are far beyond what is expected by chance.”

— Dr. Jessica Utts, Professor of Statistics, UC Davis (1995 AIR Review)

Key Milestones

1972
Russell Targ and Hal Puthoff begin RV research at Stanford Research Institute
1975
CIA formally funds Project SCANATE for intelligence applications
1978
U.S. Army establishes operational RV unit (later known as "Stargate")
1983
Ingo Swann develops Coordinate Remote Viewing (CRV) protocols
1995
Program declassified; AIR evaluation finds "statistically significant effect"
2017
CIA releases 13 million pages of declassified documents including RV research

2The Science

Evidence & controversy

Remote viewing research is contested. Some analyses report above‑chance performance under certain controls, while critics argue methodological limitations and replication challenges. RVLab is claims‑agnostic: the goal is clean, blinded data collection and consistent scoring so patterns can be evaluated transparently over many sessions.

What Matters for Controlled Experiments

Blinding
Reduce cueing
Generate output before any ground truth is shown or entered
Aggregation
Many sessions
Single sessions are noisy; trends emerge over dozens to hundreds of runs
Controls
Baseline conditions
Use consistent procedures and compare against null/alternative conditions
Reproducibility
Metadata matters
Record model, prompt/profile version, and blinding settings per run

Why Double-Blind Matters

The power of remote viewing research—and the controversy around it—centers on experimental controls. Early experiments were criticized for potential sensory leakage and experimenter bias. Modern protocols address this through strict double-blind design:

  • Target selection: Random, often computer-generated coordinates
  • Viewer isolation: No access to target pool or feedback during session
  • Blind judging: Evaluators do not know which target was assigned
  • Statistical analysis: Pre-registered hypotheses with proper controls

RVLab implements elements of these controls for AI testing: outputs are generated blind (before any ground truth is entered), receiver-mode targets can be hidden until reveal, and the analyzer uses a consistent scoring rubric across sessions. For publishable research, consider adding independent judging and additional controls.

3RV Protocols

Several methodologies have been developed for remote viewing. Each offers a different structure for accessing and recording impressions.

Coordinate Remote Viewing (CRV)

Ingo Swann, SRI (1983)

RVLab's primary protocol

The most structured protocol, using stages to progressively decode information. Begins with ideograms and gestalt impressions, moves through sensory data, dimensionals, and intangibles.

Stage 1: Ideogram/GestaltStage 2: Sensory dataStage 3: DimensionalsStage 4: Intangibles/Concepts

Extended Remote Viewing (ERV)

Skip Atwater, U.S. Army

Advanced users

Uses an altered state (deep relaxation or light trance) to access information. Sessions are longer and more immersive than CRV.

Relaxation inductionTarget focusFree-form impressionsStructured questioning

Associative Remote Viewing (ARV)

Various researchers

RVLab ARV mode

Used for binary predictions (e.g., will event X happen?). The viewer describes an unknown target that will be revealed based on the future outcome.

Define binary outcomesAssign targets to outcomesBlind viewing sessionMatch and predict

4Session Structure

A typical CRV session follows a structured progression. This is not arbitrary—each stage is designed to access different types of information while minimizing analytical overlay (AOL).

1

Preparation

2-5 min

Clear your mind. Some viewers meditate briefly; others prefer a quick walk. The goal is to quiet internal dialogue.

Tip: Find what works for you—there's no single "right" way to prepare.

2

Stage 1: Ideogram

1-2 min

Upon receiving the coordinate, make a quick, spontaneous mark on paper. This "ideogram" captures your first impression. Decode it into basic gestalts (land, water, structure, etc.).

Tip: Speed is key. Your first impression before conscious analysis kicks in.

3

Stage 2: Sensory Data

5-10 min

Record sensory impressions: colors, textures, temperatures, sounds, smells. Don't analyze—just perceive and record.

Tip: Use short phrases. "Rough gray surface" not "It's probably concrete."

4

Stage 3: Dimensionals

5-10 min

Describe spatial relationships, sizes, shapes. Sketch what you perceive. Include movement, positions, layouts.

Tip: Sketching often reveals more than words. Don't worry about artistic quality.

5

Stage 4: Intangibles

5-10 min

Concepts, purposes, emotions associated with the target. Is it natural or man-made? Active or static? What's its function?

Tip: This is where AOL is most dangerous. Label any analytical guesses.

6

Reveal & Analysis

2-5 min

Compare your data against the revealed target. Note hits and misses without judgment—both are valuable feedback.

Tip: Don't rationalize misses. Honest self-assessment improves performance.

5AI Testing Methodology

RVLab applies these classical RV protocols to test whether AI systems can participate meaningfully in remote viewing experiments under controlled conditions.

Why Test AI?

Consistent Test Subject

Unlike human participants, AI models produce consistent behavior. Same parameters yield reproducible results.

Unlimited Sessions

Generate thousands of data points. Statistical patterns invisible in small datasets become detectable at scale.

Objective Analysis

Algorithmic scoring eliminates subjective interpretation. Every session is analyzed with identical criteria.

Complete Logging

Every input and output is recorded. Experiments can be precisely replicated and verified.

Research Questions

Target Acquisition

Can AI outputs show measurable correlation with hidden targets under blinded conditions?

Symbolic Transmission

When AI generates a target internally, do outputs show patterns consistent with that target?

Model Comparison

Do different AI models (GPT-4, Claude, Gemini) exhibit different correlation patterns?

Output Characteristics

Which output types (sensory, spatial, conceptual) show strongest correlation with targets?

6Experiment Types

AI as Viewer

You select a hidden target. The AI receives only a coordinate and attempts to perceive and describe it. Outputs are analyzed for correlation with the actual target.

AI as Tasker

The system generates a hidden target. You record blind impressions, then reveal the target to compute correlation.

Multiple Models

Test the same targets across different AI models (GPT-4, Claude, Gemini) to compare performance characteristics and identify model-specific patterns.

Prompt Variation

Run sessions with different prompt structures to identify which formats produce the strongest target correlation.

Repeated Sessions

Repeat sessions under the same protocol to measure stability. Treat consistency and correlation as hypotheses to test, not proof.

Target Category Analysis

Compare AI performance across different target types (natural, man-made, abstract) to identify patterns in what the AI correlates with.

Control Conditions

Run sessions with null targets or scrambled coordinates to establish baseline output patterns for comparison.

⦿

ARV Experiments

Associative Remote Viewing for binary outcome prediction. Test whether AI outputs can be matched to future-revealed targets.

Ready to Run Experiments?

Start testing whether AI systems exhibit measurable correlation with hidden targets under controlled conditions.

Further Reading

  • Targ, R. & Puthoff, H. (1974). “Information transmission under conditions of sensory shielding.” Nature
  • Utts, J. (1996). “An Assessment of the Evidence for Psychic Functioning.” Journal of Scientific Exploration
  • Smith, P. (2005). Reading the Enemy's Mind: Inside Star Gate. Forge Books.
  • McMoneagle, J. (1997). Mind Trek: Exploring Consciousness, Time, and Space. Hampton Roads.