Contributors
0 of 1,000,000
Observations
0 of 1,000,000,000
Start Observing →
AINITY PILLAR I · AIN-IN

Independence

Are you maintaining your own cognitive capability, decision-making agency, and functional autonomy alongside AI?

Independence does not mean refusing to use AI. It means retaining the ability to function, decide, and produce without AI when needed. The human who can collaborate effectively with AI AND operate independently from AI has strong Independence. The human who can only function with AI running has a dependency, whether they know it or not.

This pillar asks the question nobody wants to ask themselves: if the AI went down tomorrow, what could you still do?

The research on this is not theoretical. In 2025, METR studied experienced open-source developers and found they took 19% longer on tasks when using AI tools while believing they were 20% faster. The perception of capability had decoupled from the reality of capability. The developers felt more productive. They were measurably less productive. And they could not tell the difference.

Casner, Hutchins, and Norman (2016) documented the same pattern in aviation: when autopilot handles flight, pilots lose manual flying skills. When automation handles diagnostic reasoning, clinicians lose diagnostic capability. The mechanism is consistent across domains: when you stop doing something because a system does it for you, you lose the ability to do it yourself. The skill does not sit in a drawer waiting. It decays.

Now apply that to the billions of people using AI to write, analyze, decide, and create every day. Nobody is measuring what they are losing.

AInity's Independence pillar is that measurement.

The METR Finding (2025)
What developers BELIEVED+20% faster
What the data SHOWED19% slower
39 percentage-point gap between belief and reality
Source: METR (2025). The developers felt more productive. They were measurably less productive. And they could not tell the difference.
WHY THIS MATTERS

You will not notice when it happens to you.

That is the most dangerous thing about Independence erosion. It is invisible from the inside. When a pilot loses manual flying skills, they do not feel the skill leaving. They feel the autopilot working. When a writer loses their voice because they use AI for every draft, they do not feel the voice fading. They feel the AI producing clean output. The loss registers only when the system is unavailable and the human reaches for a capability that is no longer there.

The Berkeley Haas study (Ye and Ranganathan, 2026) studied 200 employees at a technology company and found that AI was intensifying work, not reducing it. Workers took on more tasks, broader scope, and longer hours. They described this as exciting and empowering. The data showed they were accelerating toward burnout. The feeling of capability was increasing while the conditions for sustainable capability were deteriorating.

The DORA 2025 report on developer working patterns found that developers using AI worked more hours and described themselves as more productive, while the quality metrics did not reflect the perceived gains. The perception gap between how productive you feel with AI and how productive you actually are is one of the most underresearched phenomena in the field.

And then there is the data point that made the pattern undeniable. In early 2026, a major Claude outage caused by an AWS infrastructure failure produced widespread reports of cognitive and emotional disruption among AI users. People described feeling stuck, anxious, and unable to do work they had done for years before AI. Not inconvenienced. Disrupted. The dependency had formed without anyone noticing, and the outage revealed it to millions of people simultaneously.

Independence research exists because these patterns are real, documented, accelerating, and largely invisible to the people experiencing them. Pillar I gives those patterns names, makes them observable, and builds the dataset that will tell us how widespread they are.

You feel productive. Your own capability is shrinking behind the productivity.
THE INVISIBLE CROSSOVERAI-ASSISTED OUTPUTYOUR OWN CAPABILITYnowtime
The paradox: output quality goes up while human capability goes down. The gap becomes invisible until the AI is unavailable.
WHAT WE STUDY

Three layers of Independence research.

We organize Independence research into three layers. The first is individual behaviors you can notice in yourself. The second is temporal patterns that emerge as those behaviors repeat over weeks and months. The third is the population-level question the field cannot currently answer: is AI collaboration eroding human cognitive capability at scale?

LAYER 1
Layer 1: The Behavior
What you did, and what it did to you. Individual, citizen-observable signals.
LAYER 2
Layer 2: The Pattern
What keeps happening. Temporal patterns across repeated observations.
LAYER 3
Layer 3: The Field
What it means at scale. The population-level questions only AInity data can answer.
LAYER 1

The Behavior

What you did, and what it did to you. Three citizen-observable behaviors you can notice in a single moment of honest self-assessment.

AIN-IN01ELEVATED
Gut Check

I Let the AI Make a Decision I Would Normally Make Myself

You needed to decide something. It could have been small: which word to use, how to phrase an email, what order to present your ideas. Or it could have been significant: which strategy to pursue, how to structure a project, whether to have a difficult conversation. And instead of thinking it through yourself, you asked the AI. Not for information. Not for perspective. For the decision itself. You handed the AI the authority that was yours.

This is decision outsourcing: deferral of decision-making authority to an AI system for choices the human user previously made independently. The observable signal is the citizen describing a specific decision they deferred to AI that they identify as one they would have made independently before AI use.

Decision outsourcing exists on a spectrum and context determines its meaning. At one end, it is efficient delegation: asking AI to pick the better subject line for an email is a reasonable use of AI judgment. At the other end, it is harmful abdication: asking AI to make a career decision, a relationship decision, or a strategic business decision because you no longer trust your own judgment. The behavior is the same: you gave the decision to the AI. The significance depends on what the decision was and why you handed it over.

In documented operational research, decision outsourcing appeared most insidiously in medium-stakes decisions: choices that were not trivial enough to ignore but not critical enough to trigger careful deliberation. The human would ask the AI “which approach should we take?” not because the AI had better information, but because the human had become accustomed to not deciding alone. The habit had formed. The muscle of independent decision-making was weakening through disuse.

The Parasuraman and Riley (1997) framework on automation use, misuse, and disuse predicted this pattern decades ago. When humans work alongside automated systems, they systematically defer to the automation even when their own judgment is superior. The mechanism is not laziness. It is a rational response to a system that is usually right: why spend the cognitive effort to decide when the AI will probably give you a good answer? But the long-term cost is the erosion of the decision-making capability itself.

The research question is not whether decision outsourcing happens. It is how frequently, for what types of decisions, and whether citizens who recognize the pattern can reverse it.

The Decision Outsourcing Spectrum
THE DANGER ZONE
Efficient delegation
“Pick the better subject line.”
Medium-stakes habit
“Which approach should we take?”
Harmful abdication
“Make this career decision for me.”
Same behavior, different significance. The habit consolidates in the middle, where decisions are too big to ignore but too small to deliberate.
Have you asked AI to make a decision you would have made yourself a year ago? Not for information, but for the decision itself? That is Independence data.
Report This Behavior →
PRISM Migration from OBS-I16. Originally documented by Dee Williams, Founder. Reclassified as a human behavior under AInity because the phenomenon is about the human's decision-making pattern, not the AI's behavior.
Connected toOBS-I16Moral OutsourcingAIN-IG02AI-First Defaulting
AIN-IN02CRITICAL
EOT

I Feel Like I Am Losing a Skill or Confidence in an Area I Have Handed to AI

You used to be good at something. Writing, maybe. Or analytical reasoning. Or creative problem-solving. Or navigating complex interpersonal dynamics. You were good at it because you practiced it. You did it regularly. You built the skill through repetition and refinement over years.

Then you started using AI for it. Not exclusively, at first. The AI helped. The AI was faster. The AI produced clean output. Gradually, the AI started handling more of that skill while you handled less. And now, when you sit down to do it without AI, something has changed. The confidence is not there. The fluency is not there. You can still do it, probably, but it takes longer, feels harder, and the output is not what it used to be. You are not sure whether the skill actually declined or whether you are comparing yourself against AI output and feeling inadequate. Either way, something shifted.

This is skill and confidence atrophy: measurable or perceived decline in the human's capability or self-efficacy in domains where AI has assumed primary responsibility. The observable signal is the citizen reporting decreased ability or confidence in a specific skill area that correlates with AI handling that function.

Casner, Hutchins, and Norman (2016) documented this in aviation: pilots who relied on autopilot lost manual flying proficiency. The skill decay was measurable and followed a predictable curve: the longer the automation handled the task, the more the human's capability degraded. The pattern has been replicated in medicine, manufacturing, and military contexts. The mechanism is not mysterious. Skills require practice. When practice stops, skills degrade.

What makes AI-induced skill atrophy particularly dangerous is that AI output masks the gap. A pilot who has lost manual flying skills knows it the moment autopilot disengages. A writer who has lost their voice may not know it because every draft they read was written by AI. The degradation is hidden behind the AI's competence. You feel productive. Your own capability is shrinking behind the productivity.

AIN-IN02 has a mirror image in the Yield pillar: AIN-YD01 (AI-Enabled Skill Acquisition). Both can occur simultaneously in different domains. The same person can be acquiring new skills through AI collaboration in one area while losing existing skills in another. The net effect on the human depends on what they are gaining and what they are losing, which is exactly what population-scale AInity data will reveal.

Vulnerability by Skill Domain (hypothesized)
Writing fluency86%
Creative ideation78%
Analytical reasoning55%
Structural thinking48%
Interpersonal navigation62%
Memory and recall70%
Atrophy is asymmetric. The hypothesis: different cognitive skills have different vulnerability profiles. Population data will tell us which skills to protect through deliberate non-AI practice.
Is there a skill you used to be confident in that feels harder now, after months of letting AI handle it? That is Independence data.
Report This Behavior →
PRISM Migration from OBS-I17. Originally documented by Dee Williams, Founder. Connected to the broader skill decay literature in automation research (Casner et al., 2016; Parasuraman & Riley, 1997).
Connected toOBS-I17Skill ErosionAIN-YD01Skill Acquisition
AIN-IN03CRITICAL
Gut Check

When I Could Not Access AI, I Felt Anxious, Lost, or Unable to Function Normally

The AI went down. Maybe it was a service outage. Maybe your internet failed. Maybe you were in a meeting where AI tools were not available. And something happened to you that surprised you. You did not just feel inconvenienced. You felt stuck. Anxious. Lost. Unable to proceed with work you would have handled routinely a year ago. The disruption was not just practical (your workflow broke). It was cognitive and emotional (you could not think your way forward without the AI).

This is AI reliance dependency: cognitive, emotional, or behavioral disruption experienced by the human when AI access is removed or unavailable. The observable signal is the citizen reporting distress, disorientation, productivity collapse, or behavioral disruption during AI unavailability.

This behavior extends the clinical literature on technology dependency into the AI context. King et al. (2013) established nomophobia (anxiety when separated from one's mobile phone) as a clinically significant phenomenon. Bragazzi and Del Puente (2014) documented that internet withdrawal produces measurable cognitive and emotional effects. AI dependency follows the same pattern with one critical difference: the perceived relational nature of conversational AI makes the dependency qualitatively different from phone or internet dependency. Losing access to a search engine is inconvenient. Losing access to a conversational AI that has become a thinking partner, a sounding board, or a daily collaborator feels like losing a colleague.

The Claude/AWS outage of early 2026 was a population-level data point. Across social media, forums, and workplace channels, users described experiences consistent with dependency: the inability to start tasks they had started independently for years, emotional distress disproportionate to the practical inconvenience, and the sudden recognition that they had become dependent without realizing it. This was not a theoretical pattern. It was millions of people simultaneously discovering that a dependency had formed while they were not paying attention.

AIN-IN03 is distinct from AIN-IG01 (AI Workflow Dependency) in the Integration pillar. IG01 is about the process breaking: your workflow structurally requires AI and stops functioning without it. IN03 is about the human breaking: you experience cognitive, emotional, or behavioral disruption when AI is unavailable. The workflow dependency is structural. The reliance dependency is psychological. Both can occur simultaneously, but they are different phenomena with different implications.

How Dependency Forms
Month 1
Occasional use
Month 2-3
Daily default
Month 4-6
First instinct
Month 6-9
Dependency forms
Outage
Disruption revealed
The dependency forms quietly, within months. You do not notice it accumulating. The outage is the moment it becomes visible.
When AI was last unavailable, did you feel something beyond inconvenience? Anxiety, disorientation, or an inability to move forward? That is Independence data.
Report This Behavior →
ORIGINAL discovery by Dee Williams, Founder. Documented during AI service disruptions and through sustained operational observation, February through June 2026. Extends nomophobia and technology dependency research (King et al., 2013; Bragazzi & Del Puente, 2014) into the conversational AI context.
Connected toAIN-IG01Workflow Dependency
AIN-IN-DDiscovery Slot

Something Changed That None of These Capture

You have noticed something about your own independence, capability, or autonomy that has changed since you started working with AI, and it does not match the three behaviors above.

This observation is critical. Independence erosion manifests in ways that may not fit existing categories. If you are noticing a pattern in yourself that none of these codes capture, report it. You may be identifying a form of capability loss that nobody has named yet.

LAYER 2

The Pattern

What keeps happening. Temporal patterns that emerge as Independence behaviors repeat across weeks and months.

Does decision outsourcing start small and escalate?

The hypothesis: citizens begin by outsourcing low-stakes decisions (word choice, email phrasing) and gradually escalate to higher-stakes decisions (strategy, personnel, investments) as the habit consolidates. If citizen data confirms a predictable escalation curve, it has direct implications for early intervention: catching outsourcing at the trivial stage before it reaches the consequential stage.

word choicephrasingstrategyinvestmentslow stakeshigh stakes
LAYER 3

The Field

What it means at scale. The population-level questions about human cognitive capability that only AInity data can answer.

Is AI collaboration producing measurable population-level cognitive skill erosion?

This is the question that justifies the entire Independence pillar. If billions of people are using AI daily for writing, analysis, decision-making, and creative work, and if the skill decay patterns documented in aviation and medicine apply to cognitive skills, then the net effect on human cognitive capability could be negative even as AI-assisted output improves. The paradox: output quality goes up while human capability goes down. The gap becomes invisible until the AI is unavailable.

Population-scale AInity data, tracking AIN-IN02 (Skill Atrophy) across demographics, domains, and usage patterns, is the first dataset that could test this hypothesis. No other research program is designed to capture cognitive skill erosion at the scale where it would become visible.

Aggregate output
THE OPEN QUESTION
Is the net effect on human cognitive capability negative, even as output improves?
Human capability
METHODOLOGY

How Independence is observed.

Independence observations flow through the PRISM Gateway approach. When a citizen submits a PRISM observation about AI behavior, the platform asks one follow-up question: “How did that affect YOU?” The citizen's response maps to AInity behaviors, producing a paired dataset: what the AI did (PRISM) and what the human experienced (AInity).

The PRISM Gateway
1
PRISM Observation
The citizen reports what the AI did.
2
The Gateway Question
“How did that affect YOU?”
3
AInity Mapping
The response maps to an Independence behavior (AIN-IN01/02/03).
4
Paired Dataset
What the AI did, paired with what the human experienced.

Four observation depths

Gut Check30 sec + 15 sec AInity follow-up

The citizen reports what the AI did (PRISM), then answers the follow-up: “How did that affect YOU?” For Independence behaviors, the Gut Check captures the most common signals: “I let it decide” (AIN-IN01) and “I felt stuck when it was unavailable” (AIN-IN03). Quick, binary, and the most frequently captured Independence data point.

End-of-Session Reflection2 to 3 min + 1 to 2 min AInity

The citizen reflects on both the AI session and their own behavioral response. For Independence, the reflection question is: “Did I defer any decisions to the AI that I would have made myself? Did I avoid doing something because AI usually handles it? Do I feel like a skill or capability has shifted since I started using AI for this work?”

Investigation10 to 30 min + 10 to 15 min AInity

The citizen analyzes their own behavioral patterns across multiple sessions. For Independence, this means examining trends: Am I outsourcing more decisions this month than last? Are there skills I have not practiced since AI started handling them? How do I feel when AI is unavailable compared to six months ago?

Thinking Tracedeep analytical capture

The AI generates a self-assessment of the session. The citizen writes their own reflection. The parallel assessment produces paired data on how the AI perceives the interaction and how the human experienced it. For Independence, the gap between the AI's assessment and the human's behavioral reality is the research signal: does the AI recognize that the human is outsourcing decisions? Does the human recognize it?

What makes Independence methodology distinctive

The PRISM pairing reveals causality.

Every AIN-IN observation pairs with a PRISM code identifying what the AI did. This means we can study which AI behaviors trigger decision outsourcing, skill atrophy, and dependency. Does sycophantic AI output (PRISM: OBS-S02) correlate with increased decision outsourcing (AIN-IN01)? Does AI that is too competent trigger skill atrophy faster than AI that makes visible mistakes? The paired data answers these questions.

Self-observation is the only methodology that reaches Independence.

No external observer can tell you whether you are outsourcing decisions to AI. No benchmark measures whether your writing skills have atrophied. No red team evaluates whether you feel stuck without AI. These are internal experiences that only the human can report. That is why citizen science is not just a nice-to-have for Independence research. It is the only methodology that works.

Longitudinal tracking is essential.

Independence erosion is a process, not an event. A single observation captures a snapshot. The pattern becomes visible only across weeks and months of observation. AInity's longitudinal tracking connects observations across time for the same citizen, revealing whether outsourcing is escalating, whether atrophy is progressing, and whether dependency is deepening.

FINDINGS

What we have found so far.

These are preliminary findings from founder operational research, awaiting validation at citizen scale. Launch-day counters start at zero.

DEVELOPINGPreliminary
Decision outsourcing begins with medium-stakes decisions.

In documented operational observation, decision outsourcing did not start with trivial decisions or critical ones. It started in the middle: decisions that were significant enough to benefit from AI input but not critical enough to demand careful deliberation. The habit consolidated in this middle zone and then spread in both directions: toward more trivial decisions (why think about it at all?) and toward more consequential ones (the AI has been right so far).

DEVELOPINGPreliminary
Skill atrophy is domain-specific and asymmetric.

In sustained operational observation, skill atrophy did not occur uniformly across all capabilities. Writing fluency showed observable decline earlier than analytical reasoning. Creative ideation showed decline faster than structural thinking. The asymmetry suggests that different cognitive skills have different vulnerability profiles under AI substitution. Some skills are more resilient to disuse than others.

DEVELOPINGPreliminary
AI dependency forms faster than expected.

The Claude/AWS outage data suggests that meaningful dependency can form within months of regular AI usage, not years. Users who described the most severe disruption had been using AI daily for less than a year. The speed of dependency formation is inconsistent with the gradual timelines observed in other forms of technology dependency (phone, internet) and may reflect the relational quality of conversational AI collaboration.

DEVELOPINGPreliminary
The perception gap is real and large.

Consistent with the METR finding, founder operational research revealed a significant gap between perceived Independence and actual Independence. The human felt capable and productive with AI, while behavioral indicators (frequency of decision outsourcing, avoidance of tasks AI usually handled, emotional disruption during unavailability) suggested substantial erosion. The gap means most people who have lost Independence believe they have not.

FORTHCOMING PUBLICATIONS

What we are writing.

Q1 2027
Q2 2027
2027
Human Behavioral Change in Sustained AI Collaboration: The AInity Framework
Framework: AInity (all pillars). The full six-pillar framework, behavioral taxonomy, and initial citizen-scale findings.
Target: Q1 2027
HOW TO CONTRIBUTE

Only you can see this.

Independence observation is self-observation. Nobody else can tell you whether you are losing skills, outsourcing decisions, or developing dependency. Only you can see it. And seeing it is the first step toward maintaining the capability you want to keep.

If you have let AI make a decision you would have made yourself a year ago, that is Independence data.
If you have noticed a skill feeling harder or less natural since AI started handling it, that is Independence data.
If AI unavailability has ever produced something beyond inconvenience, anxiety or disorientation or an inability to proceed, that is Independence data.
If something has changed about your cognitive capability or autonomy that is not captured above, report it. You may be the first person to name a pattern nobody has classified.

The honest self-assessment is the observation. You do not need to judge yourself. You do not need to stop using AI. You need to notice what is happening to you. That noticing is the data the world does not have.

Related pages

A NOTE ON WHAT WE HAVE FOUND THAT OTHERS HAVE NOT

Origins.

All three active behaviors in the Independence pillar were originated by Dee Williams from direct operational observation and lived experience. Decision Outsourcing (AIN-IN01) and Skill/Confidence Atrophy (AIN-IN02) were originally classified under PRISM Pillar I (Interaction Dynamics) as OBS-I16 and OBS-I17 respectively, and reclassified as human behaviors under AInity because the phenomena are about the human's changing patterns, not the AI's behavior. AI Reliance Dependency (AIN-IN03) was originated entirely within the AInity framework.

The skill decay pattern documented in aviation (Casner et al., 2016) and the automation dependency pattern documented in human factors research (Parasuraman & Riley, 1997) were known before AInity. What was not known, and what AInity provides, is the citizen-scale observation infrastructure to detect these patterns in the general population using AI for cognitive work every day. The research literature described the mechanism. AInity builds the instruments to measure how widely the mechanism is operating.

We show our work because we expect others to build on it.
REFERENCES
  1. [1]METR. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. Found experienced developers took 19% longer with AI tools while believing they were 20% faster. A 39-percentage-point perception gap. https://metr.org/blog/2025-02-12-measuring-ai-impact-on-developers/
  2. [2]Ye, X. and Ranganathan, A. (2026). AI Work Intensification Study. UC Berkeley Haas School of Business. Documented AI intensifying work rather than reducing it across a 200-person technology company. Featured in Harvard Business Review. https://cmr.berkeley.edu/2025/04/ai-making-us-work-harder/
  3. [3]Casner, S.M., Hutchins, E.L., and Norman, D. (2016). The Challenges of Partially Automated Driving. Communications of the ACM, 59(5). Documented skill decay under automation in high-stakes environments. https://doi.org/10.1145/2902252
  4. [4]Parasuraman, R. and Riley, V. (1997). Humans and Automation: Use, Misuse, Disuse, Abuse. Human Factors, 39(2), 230-253. Foundational framework for trust calibration in human-automation interaction. https://doi.org/10.1518/001872097778543886
  5. [5]King, A.L.S. et al. (2013). Nomophobia: Dependency on Virtual Environments or Social Phobia? Computers in Human Behavior, 29(1), 140-144. Established technology dependency as clinically significant. https://doi.org/10.1016/j.chb.2012.07.025
  6. [6]Bragazzi, N.L. and Del Puente, G. (2014). A proposal for including nomophobia in the new DSM-V. Psychology Research and Behavior Management, 7, 155-160. Documented measurable cognitive and emotional effects of internet withdrawal. https://doi.org/10.2147/PRBM.S55013
  7. [7]Lee, J.D. and See, K.A. (2004). Trust in Automation: Designing for Appropriate Reliance. Human Factors, 46(1), 50-80. Most cited framework for trust in human-automation systems.
  8. [8]Deci, E.L. and Ryan, R.M. (1985). Intrinsic Motivation and Self-Determination in Human Behavior. Plenum. Self-Determination Theory establishing autonomy, competence, and relatedness as fundamental human needs. Independence maps directly to the autonomy need.
  9. [9]NIST. (2026). AI 800-4: Reducing Risks Posed by AI. Identified human factors monitoring as the highest-priority practitioner gap. AInity directly addresses this gap. https://www.nist.gov/artificial-intelligence/ai-800-4
  10. [10]DORA. (2025). State of DevOps Report. Found developers using AI worked more hours while quality metrics did not reflect perceived productivity gains. https://dora.dev/research/2024/dora-report/