AI in Practice: Applications in Medicine, Geology, and Space

AI in Medicine

Improving Diagnosis

Accurate diagnosis in time is critical; according to a US study, more than 40,000 die per year from misdiagnosis. AI offers dramatic change.

Google's DeepMind created a deep learning system that diagnoses eye diseases from retinal photographs. The model was trained on 14 million images and achieved 94% accuracy.

In practice, early detection is possible:

• Diabetic retinopathy (affects 35% of diabetics) – early detection reduces up to 70% of blindness cases.

• Glaucoma, which is difficult to diagnose in early stages.

This preserves vision for millions every year.

Additional Diagnostic Achievements

In Oncology, MIT models analyze mammograms with 87% accuracy and predict breast cancer risk 5 years in advance – increasing survival by 60–70%.

PathAI helps pathologists interpret biopsies; skin cancer detection with up to 99% accuracy, reducing human errors.

Drug development usually takes 10–15 years. AI dramatically shortens the search and testing phases.

Example: Insilico Medicine

The company uses deep learning on biological and chemical data. In 2020, it developed a molecule against pulmonary fibrosis in just 46 days, saving millions of dollars.

The stages:

1. Hypothesis generation: The model scans hundreds of thousands of compounds and identifies promising candidates.

2. Optimization: Improving the molecule for maximum activity and safety.

3. Validation: Laboratory tests confirm effectiveness.

In the traditional model, the search phase alone takes 2–3 years.

Another Example: Moderna Vaccine

Algorithms analyzed the SARS‑CoV‑2 genome; within 48 hours a vaccine prototype was developed – a crucial move in the fight against coronavirus.

Personalized medicine builds treatment based on genetics, health status, and lifestyle. AI analyzes the data and suggests more accurate and safer protocols.

Example: GRAIL

The company analyzes free DNA in blood to detect 50+ types of cancer (including pancreatic) with up to 93% accuracy in early stages.

Algorithm highlights:

• Examining over 100 million genetic variations.

• Diagnosis in a few hours – allows starting treatment quickly.

AI in Geology

Geology deals with multi-dimensional information: cross-sections, seismic maps, satellite data. Manual processing takes years; AI shortens by tens of times, improves accuracy and reduces costs.

Resource Deposit Exploration

Exploratory search is expensive and lengthy. AI predicts where there's a high probability, combining maps, seismic cross-sections, satellite images, and soil analysis.

Example: Rio Tinto

The mining giant uses ML to locate ores.

1. Technology:

• Unifying drilling data, mineralogical analysis, and maps.

• Analyzing magnetic/gravimetric anomalies to pinpoint drilling locations.

2. Results:

• ~30% reduction in exploration costs thanks to fewer failed drillings.

• In 2021, a promising iron deposit was located in Australia, valued in billions.

Logistics Automation

AutoHaul system – fully autonomous train:

• 50 trains carrying up to 28,000 tons per day.

• Logistics savings of ~$940 million per year.

• Fewer CO₂ emissions thanks to optimal routes.

Seismic activity analysis is necessary for both safety and oil and gas exploration. AI detects subtle patterns in seismic waves.

Example: ExxonMobil

The company operates DL to identify weak signals indicating oil/gas.

1. Speed: Manual processing – months; AI shortens to 3 weeks.

2. Efficiency: In one project, 40% reduction in false positives and savings of over $700 million per year in drilling.

Further Progress

In Chile, with up to 1500 earthquakes per year, AI analyzes data from thousands of sensors along the Andes; accuracy of predicting threatening earthquakes reached 85%, enabling rapid evacuation and disaster prevention.

Production Management on Offshore Platforms

Underwater mining requires delicate control. AI monitors equipment, pressure, and temperature in drillings and reduces failures.

Example: Shell

AI system for monitoring underwater drilling facilities:

• Savings: In the first year, 7 major failures were prevented – savings of over $700 million per year.

• Sustainability: Energy optimization and 12% reduction in emissions.

AI in Cosmonautics

Space poses challenges: massive data, need for autonomy, and limited resources. AI assists in navigation, observation processing, and infrastructure protection.

Spacecraft Control

Communication delays require autonomous decisions. AI enables vehicles to choose routes and respond to changes.

Example: Perseverance

NASA's robot on Mars (2020) uses ML:

1. Autonomy:

• Independent navigation, obstacle avoidance with 98% accuracy.

• Processes over 20GB per day – optimal mission time utilization.

2. Sample Analysis:

• PIXL system for spectral analysis; AI determines mineral composition with micrometer accuracy, for detecting signs of ancient life.

3. Results:

• In 2022, organic molecules were detected in Jezero Crater – a possible hint of ancient microbes.

Exoplanet Detection

Deciphering telescope data is enormous; Kepler examined 150,000 stars. AI became a key tool.

Example: Google and NASA

Joint project with neural networks to analyze Kepler:

• Discovery of the Kepler‑90 system with 8 planets – an analog to the sun.

• Processing time: 30 days instead of decades manually.

• Accuracy: Detecting weak signals considered noise.

Further Achievements

Algorithms for TESS helped in 2023 discover TOI‑700 e – an Earth-sized planet in the habitable zone.

Dealing with Space Debris

Over 34,000 debris fragments endanger satellites. AI manages maneuvers to avoid collisions.

Example: SpaceX

Starlink satellites are equipped with AI to evade; in 2021, over 10 collisions were avoided, including with ESA.

The Future: Active Cleanup

Companies like ClearSpace are planning in 2024 robots with AI to collect large objects – improving orbital safety.

Next-Generation Telescopes

In James Webb and similar, AI will assist not only in image processing but also in observation planning.