Value Lock-in
Overview
Section titled “Overview”Value lock-in occurs when AI enables the permanent entrenchment of a particular set of values, making future change extremely difficult or impossible. This is a symmetric critical outcome—lock-in could preserve beneficial values (democratic norms, human rights, flourishing) or entrench harmful ones (authoritarianism, narrow corporate interests, destructive ideologies).
The key insight is that AI may create unprecedented stability. Throughout history, bad regimes eventually fell; bad ideas eventually lost influence. AI could change this by providing tools for permanent control or by optimizing systems so effectively that alternatives become unviable.
Polarity
Section titled “Polarity”Symmetric: Can be good or bad.
| Pole | Description | Example |
|---|---|---|
| Positive lock-in | Beneficial values become stable and protected | Democratic norms, human rights, flourishing embedded in stable institutions |
| Negative lock-in | Harmful values become permanent | Authoritarian control, corporate extraction, ideological extremism entrenched forever |
The same mechanisms can produce either outcome. The key question is: which values get locked in?
How This Happens
Section titled “How This Happens”Lock-in Mechanisms
Section titled “Lock-in Mechanisms”1. Power Concentration AI capabilities may concentrate in the hands of a small number of actors (states, corporations, individuals). Once concentrated, power can be self-reinforcing—those with AI advantages can use them to maintain their position.
2. AI-Enabled Surveillance Unprecedented surveillance capabilities could make resistance to any regime impossible. Dissenters can be identified before they organize; alternative power bases can be dismantled before they form.
3. Optimization Lock-in AI systems optimizing for particular objectives create path dependencies. Once systems are built around certain assumptions, changing course becomes increasingly costly. Society may “optimize itself into a corner.”
4. Infrastructure Dependency Critical systems may become dependent on particular AI configurations. Changing values would require rebuilding infrastructure—potentially prohibitively expensive.
Types of Negative Lock-in
Section titled “Types of Negative Lock-in”Authoritarian Lock-in
Section titled “Authoritarian Lock-in”An authoritarian regime uses AI to make its control permanent:
- Total surveillance eliminates privacy and organized resistance
- AI-optimized propaganda shapes beliefs at individual level
- Economic systems reward loyalty and punish dissent
- Security forces can identify and neutralize threats preemptively
Historical analogy: Totalitarian regimes of the 20th century eventually fell. AI might make such regimes stable indefinitely.
Corporate Lock-in
Section titled “Corporate Lock-in”Corporate interests become permanently entrenched:
- Regulatory capture amplified by AI lobbying and influence
- Market positions become unassailable through AI advantages
- Consumer behavior optimally manipulated
- Democratic oversight becomes nominal
Ideological Lock-in
Section titled “Ideological Lock-in”A particular ideology or worldview becomes permanent:
- AI-optimized memetic content entrenches beliefs
- Alternative perspectives filtered or suppressed
- Education and socialization controlled
- Could be religious, political, or philosophical
Value Stagnation
Section titled “Value Stagnation”Even without malice, values may simply freeze:
- Current values encoded into long-lasting AI systems
- No mechanism for moral progress
- Humanity locked into 2020s ethics forever
- Loss of the “moral circle expansion” that has characterized human history
Key Parameters
Section titled “Key Parameters”| Parameter | Direction | Impact on Negative Lock-in |
|---|---|---|
| AI Control Concentration | High → Enables | Few actors can entrench their values |
| Human Agency | Low → Enables | Less capacity to resist or change course |
| Governance Capacity | Low → Enables | Institutions can’t prevent capture |
| Societal Resilience | Low → Enables | Less ability to recover from bad lock-in |
| Epistemic Health | Low → Enables | Harder to recognize bad values being locked in |
Which Ultimate Outcomes It Affects
Section titled “Which Ultimate Outcomes It Affects”Long-term Trajectory (Primary)
Section titled “Long-term Trajectory (Primary)”Lock-in directly determines the long-run trajectory—whose values shape the future:
- Positive lock-in → Stable, flourishing future
- Negative lock-in → Permanent dystopia or stagnation
Existential Catastrophe (Secondary)
Section titled “Existential Catastrophe (Secondary)”Bad lock-in may be as catastrophic as extinction:
- Permanent totalitarianism could be “worse than death”
- Value stagnation could foreclose most of the future’s potential value
- Lock-in typically happens gradually, but the moment of “no return” may be sharp
Warning Signs
Section titled “Warning Signs”Signs of negative lock-in forming:
- Power concentration accelerating beyond historical norms
- Surveillance capabilities expanding without oversight
- Dissent becoming systematically difficult
- AI systems encoding current values without mechanisms for updating
- Alternative AI development being foreclosed
- Democratic institutions weakening while AI capabilities grow
Signs of positive lock-in potential:
- Democratic oversight of AI development
- Distributed AI capabilities
- Strong civil society engagement
- International cooperation on AI governance
- Explicit attention to value representation and updating
Interventions That Address This
Section titled “Interventions That Address This”To prevent negative lock-in:
- Distributed AI development (prevent concentration)
- Strong democratic oversight of AI deployment
- Privacy protections against surveillance
- Antitrust enforcement in AI markets
- International cooperation to prevent authoritarian AI exports
To enable positive lock-in:
- Deliberate encoding of beneficial values in AI systems
- Mechanisms for updating values as moral understanding improves
- Broad stakeholder input into AI development
- Constitutional protections adapted for AI era
Probability Estimates
Section titled “Probability Estimates”Lock-in risk depends heavily on which values are currently winning:
| Scenario | Assessment |
|---|---|
| Some form of lock-in occurs | Likely if AI continues advancing |
| Authoritarian lock-in | Significant risk in some regions; global unclear |
| Corporate lock-in | Already partially occurring in some domains |
| Positive lock-in | Possible but requires deliberate effort |
Related Content
Section titled “Related Content”Existing Risk Pages
Section titled “Existing Risk Pages”Models
Section titled “Models”External Resources
Section titled “External Resources”- Karnofsky, H. (2021). “All possible views about humanity’s future are wild”
- Ord, T. (2020). The Precipice — Chapter on dystopian scenarios
- Bostrom, N. (2019). “The Vulnerable World Hypothesis”