Most AI discourse is stuck arguing about whether ChatGPT can write a decent email. Meanwhile, neural networks are peeling back forest canopies and finding archaeological sites that human eyes missed for a century.
This isn't speculative. It's happening now, with free data and open-source models. And the industry it's about to transform — cultural resource management — is worth $1.72 billion annually in the US alone.
Here's why this matters, how it works, and what it tells us about where AI actually creates value.
The Industry Nobody Talks About
Every time a highway gets widened, a housing development breaks ground, or a pipeline gets laid, federal and state law requires an archaeological survey. In California, CEQA (the California Environmental Quality Act) triggers mandatory review for nearly every construction project.
This isn't optional. It's the law.
The cultural resource management (CRM) industry exists to fulfill these requirements. It's a $1.72 billion market in the US, projected to hit $1.85 billion by 2031. The Infrastructure Investment and Jobs Act is injecting over $1 billion in new archaeological work through 2027. The industry needs 8,000+ new positions filled.
And here's the problem: the core technology hasn't changed in 50 years.
Walking in Straight Lines (Literally)
The standard CRM survey method is called a pedestrian survey. Archaeologists walk across a site in parallel transects, 15 meters apart, eyes on the ground, looking for artifacts and features.
It's exactly what it sounds like. People walking and looking.
For a 500-acre project, that's weeks of fieldwork, tens of thousands of dollars, and results that are fundamentally limited by what humans can see on the surface. Buried features? Missed. Sites under dense canopy? Invisible. Human bias toward recognizable feature types? Baked in.
The data management is worse. Many state archaeological databases still run on Microsoft Access. California's CHRIS (California Historical Resources Information System) is only now migrating to a modern platform.
This is a $1.72 billion industry running on walking and Access databases.
What AI Actually Sees
LiDAR — Light Detection and Ranging — fires laser pulses from drones or aircraft and measures the returns. When you strip away vegetation, you get a bare-earth digital terrain model that reveals subtle surface features invisible to the naked eye.
A slight depression that was a storage pit 2,000 years ago. A barely raised platform that was a house foundation. Linear features that were ancient field boundaries. Mounds. Enclosures. Earthworks.
Humans can analyze these terrain models manually, but it's slow, subjective, and scales poorly. Here's where AI changes the math.
How it works
1. Data acquisition: Download free LiDAR from USGS 3DEP (sub-meter resolution, covering most of the US) or fly a drone with a LiDAR payload
2. Preprocessing: Generate relief visualizations — hillshade, slope, sky-view factor, openness — using tools like RVT (Relief Visualization Toolbox)
3. AI detection: Feed the visualizations into a CNN (convolutional neural network) trained on known archaeological sites
4. Output: Probability heatmaps showing where archaeological features likely exist
The ADAF project (AI-driven Detection of Archaeological Features), built by researchers in Ireland, demonstrated this pipeline with 84% recall using an HRNet W48 architecture. That means it correctly identified 84% of known archaeological sites in its test area.
More importantly, it found sites that human surveyors had missed.
The entire stack — ADAF, PyTorch, PDAL, QGIS, WhiteboxTools — is open source. The training data from USGS is free. You can start building this today with zero hardware cost.
Adding Satellites to the Mix
LiDAR is powerful but requires either existing aerial surveys or flying your own drone. Satellite imagery adds a second detection layer that covers the entire planet.
Crop marks — subtle differences in vegetation growth over buried features — are visible in multispectral satellite imagery but invisible to the human eye on the ground. A buried stone wall causes slightly different moisture retention, which causes slightly different plant growth, which shows up as a distinct spectral signature in Sentinel-2 data.
Sentinel-2 is free, updated every 5 days, at 10-meter resolution. Landsat is free. CORONA declassified spy satellite photos from the 1960s-1970s are free. Planet Labs offers free research access at 3-5 meter resolution with daily revisits.
Temporal analysis — watching how these patterns change across seasons — is something AI handles effortlessly and humans simply can't do at scale.
Why Nobody Has Commercialized This in the US
The pieces exist. The data is free. The models are open source. The market is enormous. So why isn't anyone selling this?
The expertise gap is the moat.
Building the AI is a software problem. Archaeologists can't do it. Understanding CRM compliance, state databases, tribal consultation protocols, CEQA/Section 106 requirements, and what agencies actually need in deliverables — that's domain expertise that takes a decade to build. Software engineers don't have it.
The only team that can build this is one that bridges both worlds. An AI engineer and a domain expert with deep regulatory knowledge and industry relationships.
The closest competitors tell the story:
Specialized AI Applications
Exploring AI for specialized domains? Our consulting services help organizations apply AI to unique challenges — from data pipelines to custom model deployment.
- ArchAI (UK) proved the commercial model works — but they're UK-only
- ADAF (Ireland) built the open-source foundation — but it's academic, not commercial
- GlobalXplorer (Sarah Parcak's citizen-science platform) went inactive in 2023
- CRM firms like AECOM and ICF are tech laggards — they'd be customers, not competitors
There is no US commercial AI archaeology software company. Zero.
The Business Model That Writes Itself
If you're thinking about this from a startup perspective, the economics are almost unfair:
SaaS: Upload your LiDAR or satellite data, get archaeological probability maps. $50-500/month. Target: 8,000+ CRM firms, state agencies, universities, tribal nations.
Drone surveys: Fly sites with drone LiDAR, run AI analysis, deliver GIS-ready reports. $150-500/acre. A single CalTrans highway project can be hundreds of acres.
Government grants: NSF SBIR explicitly lists archaeological technology as an eligible topic. Phase I is $275K. Phase II is $1M. A 501(c)(3) research partner makes you even more competitive.
Training: "AI for Archaeologists" certification. A CRM industry that desperately needs to modernize. $500-2,000 per course.
Hardware costs have collapsed. A DJI Matrice 350 RTK with a Zenmuse L2 LiDAR payload — survey-grade, 4cm accuracy — runs $26,875. That's a rounding error on a government contract. And you can start with $0 by processing free USGS data.
What This Tells Us About Real AI Value
The AI gold rush is obsessed with chatbots, image generators, and coding assistants. Those are real products, but they're also crowded markets where every major tech company is competing.
The actual opportunities — the ones with enormous markets, zero competition, and clear paths to revenue — are in industries that most AI engineers have never heard of.
Cultural resource management. Agricultural yield optimization. Infrastructure inspection. Environmental compliance. Water rights management.
These aren't sexy. They don't get Product Hunt launches. But they're massive, underserved, and hungry for exactly what modern AI can deliver.
The pattern is always the same:
1. A regulated industry with mandatory compliance requirements
2. Existing workflows that are manual, slow, and expensive
3. Free or cheap data that AI can process better than humans
4. A domain expertise barrier that keeps generic tech companies out
5. Government funding pathways (grants, contracts) that de-risk the business
If you're an AI engineer looking for your next venture, stop scrolling Twitter and start reading federal regulations. The $1.72 billion archaeology market is just one example. There are dozens of industries with the same profile.
The Technology Stack Behind AI Archaeology
The actual pipeline for AI-assisted archaeological survey involves three layers of technology, each feeding the next. First, drones equipped with LiDAR sensors fly grid patterns over a survey area. The LiDAR pulses penetrate vegetation canopy and return a point cloud — millions of elevation measurements that create a bare-earth digital terrain model. This alone reveals features invisible from the ground or from satellite imagery.
Second, machine learning models trained on known archaeological features scan the terrain model for patterns. Stone walls create linear elevation changes. Building foundations leave rectangular depressions. Ancient roads show as subtle raised paths cutting across natural contours. The models flag anomalies that match these patterns, ranked by confidence score. A trained archaeologist reviews the flagged locations and decides which ones warrant ground-truthing. This hybrid approach — AI flags, human validates — is faster than either alone and more accurate than traditional survey methods.
Third, satellite multispectral imagery adds another data layer. Vegetation growing over buried structures often shows different spectral signatures — subtle color differences invisible to the human eye but detectable by sensors measuring near-infrared and other wavelengths. Combining LiDAR terrain models with multispectral analysis reduces false positives significantly. Our RAG system guide explains similar pattern-matching approaches applied to different domains.
Case Studies: What AI Has Already Found
In 2018, researchers using LiDAR discovered over 60,000 previously unknown Maya structures hidden beneath the Guatemalan jungle canopy. The survey covered 2,144 square kilometers in days — a task that would have taken decades of ground survey. AI-assisted classification of the LiDAR point cloud identified buildings, causeways, defensive walls, and agricultural terraces that rewrote estimates of Maya population density upward by a factor of three.
In Cambodia, AI analysis of LiDAR data around Angkor Wat revealed a vast urban grid extending far beyond the temple complex. The city supported an estimated 750,000 people — making it one of the largest pre-industrial cities in the world. None of this infrastructure was visible from the ground or from conventional satellite imagery because it had been covered by centuries of forest growth and soil accumulation.
More recently, satellite AI analysis in Saudi Arabia identified thousands of previously unknown mustatil structures — large rectangular enclosures up to 600 meters long, dating to approximately 5000 BCE. The sheer number of structures found (over 1,600) using automated detection far exceeded what decades of traditional survey had recorded. These discoveries are not incremental improvements — they are fundamental revisions to our understanding of ancient civilizations, made possible because AI can process geographic data at a scale and speed that human analysis cannot match.
The commercial implications are significant. Environmental impact assessments, required before any major construction project, must account for cultural heritage sites. Currently these assessments rely on incomplete historical surveys and expensive manual fieldwork. AI-assisted remote sensing could reduce survey costs by 60-80% while dramatically increasing detection rates. Construction companies, energy firms, and government agencies all need this capability, but the archaeology industry has been slow to commercialize it. The opportunity is wide open for anyone who can bridge the gap between academic research and commercial application.
Getting Started With Archaeological AI
You do not need a PhD or a million-dollar budget. Publicly available LiDAR data from USGS covers much of the United States. Open-source tools like QGIS handle terrain visualization, and pre-trained object detection models can be fine-tuned on archaeological features with a few hundred labeled examples. The barrier to entry has dropped dramatically since 2020.
The real bottleneck is labeled training data. Archaeological features are rare by definition, which means class-imbalanced datasets. Techniques like data augmentation, synthetic example generation, and transfer learning from related domains (geological feature detection, urban planning) help compensate. A good starting project is training a model to detect known features in areas that have already been surveyed, then validating against the published archaeological record before applying it to unsurveyed terrain.
The Takeaway
AI's most transformative applications aren't in Silicon Valley's comfort zone. They're in industries where people still walk in straight lines across fields, record data in Access databases, and wait weeks for results that a neural network could produce in hours.
The tools are free. The data is free. The market is there. The only thing missing is the team that bridges the gap between the technology and the domain.
That's the real AI opportunity in 2026 — not building another chatbot, but finding the industries that desperately need what already exists and delivering it.
---
Like One builds AI systems that work autonomously — from grant writing to music production to archaeological survey. See what we're building →