The Carbon Economy
A Third Path for Consumer AI
What if AI worked with you, not against you?
Pay only for what you use. Earn credits back when you help us improve. No expiring balances. No dark patterns. No selling your data. This paper introduces Carbon: a new economic model for consumer AI.
Why This Matters Now
We are entering an era of profound transformation.
AI will reshape how we work, create, and connect. The next decade will bring changes we can barely imagine. Some exhilarating, some unsettling, all accelerating.
In times of rapid change, people need anchors. They need to know where they came from. They need their history intact, organized, and accessible. Not scattered across failing hard drives, forgotten cloud accounts, and shoeboxes in closets.
Phossil exists to help people secure their past as they navigate an uncertain future.
We believe everyone deserves to understand what AI can do. Its power, its limitations, its costs. We built this company to make that capability accessible, not to gatekeep it behind enterprise contracts or exploit it through surveillance.
Your memories are your anchor. We're building the tools to preserve them.
The Problem
AI is getting cheaper. Access is not.
Today, advanced AI capabilities are locked behind flawed models:
Pure Subscriptions ($20-30/month) - You pay the same whether you run 10 AI operations or 10,000. Light users subsidize heavy users. Occasional users overpay; power users get a bargain.
Advertising - "Free" means your attention is the product. Your family photos surrounded by targeted ads. Your memories analyzed for advertiser insights.
Data Selling - The worst trade. Your faces, your locations, your life events packaged and sold. You're not the customer. You're the inventory.
For applications handling your most sensitive content (family photos, medical records, private documents) none of these work.
The Carbon Economy
We built something different.
Carbon Credits are a micropayment currency for AI operations. Simple rules:
- You buy Carbon when you want it (1,000 Carbon = $10)
- Each AI operation costs Carbon (transparent, shown before you confirm)
- Carbon never expires - not purchased, not earned, not bonuses, not monthly allowance
- No ads. No data selling. Your photos stay yours.
What Carbon Buys:
| Phototology (AutoDate + AutoTag + AutoLocale) | Free |
| Living Portrait (AI Video) | 250 Carbon ($2.50) |
| Colorizer (B&W to Color) | 100 Carbon ($1.00) |
| Enhancer (Upscale & Restore) | 25 Carbon ($0.25) |
| Face Detection & Recognition | Free |
That's it. Pay for what you use. Nothing more.
Earn, Don't Just Spend
Carbon flows both ways. When you improve our AI, you earn Carbon back:
| Correct a date estimate | +1 Carbon |
| Verify a face match | +1 Carbon |
| Reject an incorrect match | +1 Carbon |
| Add a birthday to a person | +5 Carbon |
| Complete a person profile | +5 Carbon |
| Refer a friend (paid signup) | +500 Carbon |
Earnings are capped at 10% of your collection's Carbon value per month—enough to reward genuine contribution while keeping the system sustainable.
Your corrections make our AI better. Better AI means better results for everyone. We're paying you for something genuinely valuable.
Generous From Day One
Buy once, preserve forever. Each tier includes a Carbon grant:
| Free | 200 Carbon (75 photos) |
| Shoebox ($99) | 2,250 Carbon (500 photos) |
| Family ($249) | 5,750 Carbon (2,500 photos) |
| Heirloom ($499) | 12,500 Carbon (10,000 photos) |
One-time purchase. No subscriptions, no renewals. Year 1 storage included.
Pay once. Preserve forever. Carbon never expires.
Simple and Honest
Carbon never expires. What you buy stays yours. What you earn stays yours. Use it whenever you're ready.
Preserving family memories is meaningful work. We believe the tools should respect that.
The Bottom Line
| Pure Subscription | Ads | Data Selling | Carbon | |
|---|---|---|---|---|
| Pay for what you use | ✗ | ✓ | ✓ | ✓ |
| No privacy sacrifice | ✓ | ✗ | ✗ | ✓ |
| No attention harvesting | ✗ | ✗ | ✓ | ✓ |
| Sustainable business | ✓ | ✗ | ✗ | ✓ |
| Rewards contribution | ✗ | ✗ | ✗ | ✓ |
Carbon wins every dimension that matters.
The memories are priceless. The AI should just cost Carbon.
Read the Full Whitepaper
Complete implementation philosophy, exemption policies, credit management architecture, and sustainable engagement design.
Requires free email registration
Questions? Ideas? Want to collaborate?
[email protected]Bayesian Temporal Inference
A Probabilistic Framework for Dating Undated Photos
How do you date a photograph with no metadata?
By fusing multiple evidence signals through Bayesian inference—visual cues, biometric indicators, technological artifacts, and contextual signals. Our system achieves 92.9% accuracy within ±2 years.
Abstract
Digital photo collections frequently contain images with missing, corrupted, or unreliable temporal metadata. This paper introduces a novel probabilistic framework for estimating photograph dates using multi-signal Bayesian inference. By fusing visual evidence, biometric indicators, technological artifacts, and contextual signals, our system achieves significantly higher accuracy than single-signal approaches.
We present the architecture of our Temporal Inference Engine, introduce the PhotoDate benchmark for standardized evaluation, and discuss implications for digital preservation at scale.
Results: From Validation to Production
Initial validation (v1.0, December 2025) achieved 78% accuracy within ±2 years on 61 photographs with verified dates.
Production calibration (v3-gemini-stable, January 2026) improved accuracy to 92.9%—a 15-point improvement through systematic age bias correction.
| Metric | Result |
|---|---|
| Within ±2 years | 92.9% |
| Average error | 1.125 years |
| Median error | 1 year |
| Best category | Children (3-12): 100% |
| Weakest category | Adults (30-45): 81% |
Read the Full Research Paper
The complete paper includes our theoretical framework, evidence taxonomy, system architecture, the PhotoDate benchmark, and our Carbon economy model for sustainable AI access.
Requires free email registration
Questions about our temporal inference approach?
[email protected]The Economics of Photo Organization
A Comparative Analysis of Manual, Professional, and AI-Assisted Preservation Methods
How much does it really cost to organize your family photos?
Manual organization requires approximately 118 hours per 1,000 photographs. AI-assisted methods reduce this to 7.25 hours—a 94% reduction. This paper presents the complete economic analysis.
Key Finding
For collections over 500 photos, AI-assisted organization is the economically rational choice.
The cognitive overhead of learning to identify individuals across ages represents 58% of total manual organization time. AI eliminates this learning curve entirely—once trained on 5 sample images, recognition is instantaneous across the entire collection.
The optimal time to organize family photographs was 20 years ago. The second-best time is now.
What's in the Full Paper
- Complete methodology: Time estimation framework, learning curve models, and research design
- Detailed results: Charts comparing manual, professional, and AI-assisted approaches
- Cost-benefit analysis: Time cost, direct cost, and total value calculations
- Recommendations: Which approach is right for your collection size
- Limitations: Honest assessment of study constraints and biases
Read the Full Paper
Enter your email to unlock the complete research paper with methodology, results, and cost analysis.
Questions about our research methodology?
[email protected]Photo Archive Statistics
The Scale and Urgency of Photo Preservation
4 trillion photographs worldwide. 85% still undigitized.
Every day, knowledge about old photos disappears forever as the generation who can identify faces passes away. This document compiles statistics on the scale of the preservation crisis.
The Scale: United States
| Metric | Estimate |
|---|---|
| Average photos per US household | 2,000-3,500 prints |
| Total US households | ~130 million |
| Estimated US physical photos | 260-455 billion |
| Photos in shoeboxes/unsorted | 60-70% |
| Photos in albums | 20-30% |
| Photos professionally stored | <5% |
Global Estimates
| Metric | Estimate |
|---|---|
| Global physical photographs | 3.5-4 trillion |
| Peak film year (2000) | ~80 billion prints |
| Cumulative photos 1826-2000 | ~3.5 trillion |
| Digital photos taken annually (2023) | ~1.4 trillion |
Photo Degradation Rates
Physical photographs have limited lifespans, especially under typical home storage conditions:
| Photo Type | Typical Home Lifespan |
|---|---|
| B&W Silver Gelatin Prints | 50-100 years |
| Color Chromogenic Prints | 15-30 years |
| Polaroid/Instant Prints | 10-20 years |
| Color Negatives | 15-40 years |
| Kodachrome Slides | 50-75 years |
| Nitrate Film (pre-1950) | Critical - actively degrading |
Environmental factors: High humidity (3-5x faster degradation), temperature fluctuations (5-10x faster), light exposure (10-100x faster fading).
Acetate "Vinegar Syndrome": Affects negatives from 1925-1990. 30-50% of family acetate negatives show early signs. Once started, accelerates exponentially.
Color Print Fading: Average 1970s-1990s color prints have lost 20-40% of dye density. Magenta layer typically fails first (creates cyan/blue cast).
Climate and Disaster Risk
| Risk Factor | % of US Photos Affected |
|---|---|
| Flood zones (100-year) | 15-20% |
| Hurricane risk areas | 25-30% |
| Wildfire risk zones | 10-15% |
| High humidity regions (Southeast) | 30-35% |
Hurricane Harvey (2017) damaged an estimated 1+ billion photos in the Houston metro area alone. Standard homeowner's policies cover only $200-500 for photos/memorabilia. Emotional value: Uninsurable.
The Knowledge Loss Crisis
Silent Generation (1928-1945): ~20 million remaining in US. Primary identifiers for pre-1960 photos. Annual mortality: ~10%.
Baby Boomers (1946-1964): 70 million in US. Key identifiers for 1960s-1980s photos.
"Orphan photos" (unidentifiable): Estimated 30-50% of pre-1970 photographs.
Critical window: For photos taken before 1950, the window to capture identifications is approximately 5-10 years before knowledge becomes largely unrecoverable.
Photos without labels: 70%+ become unidentifiable within 2 generations. Cross-generational identification (grandchildren identifying great-grandparents): <20% accuracy.
Digitization Rates
| Metric | Estimate |
|---|---|
| % of physical photos digitized (US) | 15-25% |
| Households that have started digitizing | 35-45% |
| Households with complete digitization | <10% |
| DIY projects started but abandoned | 60-70% |
Primary barriers: Time required (45-55%), "Will get to it later" (40-50%), Don't know where to start (25-30%), Technical difficulty (20-25%).
Photo Backs: The Hidden Metadata
Handwritten annotations on photo backs contain irreplaceable information:
| Era | % with Writing on Back |
|---|---|
| 1900-1940 | 20-35% |
| 1940-1960 | 30-45% |
| 1960-1980 | 25-40% |
| 1980-2000 | 15-25% |
The metadata crisis: Front-only scanning loses 100% of back metadata. Ink fading and cursive handwriting becoming unreadable to younger generations compound the problem.
The Urgency Matrix
| Photo Era | Urgency Level | Recommended Timeline |
|---|---|---|
| Pre-1940 | CRITICAL | Immediate (1-2 years) |
| 1940-1960 | HIGH | Near-term (2-5 years) |
| 1960-1980 | Moderate | Medium-term (5-10 years) |
| 1980-2000 | Lower | Long-term planning |
Key Takeaways
The Scale: 260-455 billion physical photos in US homes alone, most unsorted and at risk.
The Ticking Clock: Every day, people who can identify old photos pass away, taking irreplaceable knowledge with them.
The Physical Reality: Color prints from the 1970s-1990s are actively fading; many have already lost significant detail.
The Procrastination Problem: 85%+ of photos remain undigitized despite decades of available technology.
The window is closing. The technology is ready. The time is now.
Questions about photo preservation research?
[email protected]AutoDate-100 Benchmark
Our open challenge for the research community. Standardized evaluation for temporal photo analysis.
Coming Soon