AI summary by TwinMind
Inside the Rise of AI-Driven Combat
Johnny Harris
18m 48s
Introduction to AI's Military Impact
- Putin told students "whoever leads in AI will rule the world"
- White House issued first AI memo calling it "era defining technology with significant and growing relevance to National Security"
- Partnership with Shashank Joshi, defense editor at The Economist
- AI beginning to appear in "every single decision made in war"
The Kill Chain Framework
- Both hypothetical armies trying to execute same process:
- Get eyes into enemy territory
- Find enemy's most valuable military targets
- Fix exact location
- Determine weapons to use and proper targeting/deployment
- Fire on targets to engage and hit
- Assess effectiveness
- Victory favors those who gather best information and make accurate decisions fastest
- Executing kill chain requires commanders to make countless decisions with incomplete/unreliable information
Sensor Data Collection Phase
- Both armies use various sensors:
- Electrooptical sensors on satellites (cameras)
- Drones scanning for targets
- Reconnaissance teams on ground
- Radar sensing airborne objects
- Data recorded and sent to Central Command Center for decision-making
AI Advantage in Target Identification
- Computer vision/object recognition "exploded in how good it is" since 2012
- AI breakthrough started with "recognizing images of cats - if you can recognize a cat you can recognize a tank"
- AI Command Center processes visual data with computer vision software
- Machines automatically identify enemy planes, formations with "90% probability"
- Calculate exact coordinates on drone itself
- Traditional army relies on humans manually reviewing every image
- Human process involves "lot of friction" - slowly identifying targets, plotting manually on maps
- Maps only exist in single Command Center initially
Data Integration and Cloud Computing
- AI army automatically adds findings to cloud database accessible to other command centers
- AI fuses multiple data sources:
- Mobile phone records
- Ship tracking websites
- Aircraft tracking websites
- Consumer retail activity near naval bases/airfields
- System "spitting out anomalies" from "vast torrents of data"
- Maps fill up "much quicker" with potential targets plus context
Decision-Making and Recommendations
- AI software platform makes predictions/inferences, gives recommendations to commander
- Takes into account target plus other factors:
- Available planes
- Fuel on airfield
- Ammunition supplies
- Optimal destruction methods
- Coolant, fuel, oil, repairs, rest for crew, food
- "Lot of stuff being done by dozens of mid-level officers with pens, papers, spreadsheets can be done by software"
- Human might take "couple of days" - code does it in "tenth of the time"
Operational Execution
- AI army commander receives software-generated target list:
- Tank groups near front line
- Bridge serving as supply line
- Ammunition depot with most enemy reserves
- Commander selects ammunition depot using artillery cannons
- Software confirms cannons recently serviced with plenty ammunition
- Traditional army "just starting to assemble initial list of valuable targets"
Autonomous Drone Operations
- AI commander sends explosive drone swarm to destroy enemy tank fleet
- Enemy jamming signals interfere with pilot-drone connection
- AI-enabled drones can fly themselves when jammed
- Drone compares objects to stored image library
- Says "I know I am supposed to hit this thing" and locks onto target
- Can maneuver to target in final 100-300 meters of flight
- "Real shift in character of warfare" with precision guidance on "so many tiny little systems throughout battlefield"
Real-World Applications
Gaza Operations
- Israeli military uses AI target system
- Data from tapped phones and location tracking
- Algorithmically chooses targets to strike
- Generates "huge numbers of targets per day of suspected Hamas militants"
- According to 6 Israeli soldiers, IDF bases air strikes on "AI generated lists treating target lists as if it were human decision"
- IDF disputes this characterization
Ukraine Conflict
- "Key test bed" for "most advanced military AI on both sides"
- Major innovation in drones with software/computer vision
- Drones complete final flight leg to hit targets even when jammed and pilot loses control
US Military Development
- Tech startups preparing for warfare relying on "sensors, data, algorithms"
- US military developing AI system processing aerial/satellite imagery
- Picks out enemy targets "sometimes more accurately than humans"
- UK has similar program
- Cloud companies (Amazon, Microsoft) getting involved in defense
- Palantir: software allows operations that "used to require 2,000 people, now can do with just 20"
Project Convergence
- US Army's ambitious project aggregating all sensor data
- "One integrated platform, one central command software"
- Uses "high-tech equipment more efficient and faster than humans"
Ethical and Safety Concerns
- Questions about ethics, legality, safety
- Will AI make war "easier and quicker"?
- Taking humans out of process - will it make war "less human"?
- Concern about "Killer Robots"
- Debate: "human must always retain control over decision to use force"
Expert Analysis on Human Control
- Shashank argues focus shouldn't be on final lethal decision
- "Really consequential step was taken when you identified something as target in first place"
- Problem with commander "mashing red button on loop saying yep yep yep"
- "Doesn't seem like human control" even if technically signing off
- "Doing so without real engagement moral or intellectual with what they are doing"
Recommended Approach
- What matters: human controller understanding the software
- Commander should know:
- What system has been trained on
- When it works effectively vs. when it doesn't
- When not to rely on it
- When to question/mistrust inputs
- "Much more like human control even if seeding quite a lot of control in some circumstances but holding it back in others"
- Debate should focus on "when are you trusting systems, what do you know about them, have you interrogated them to understand weak spots"
Human Bias and AI Authority
- "Humans tend to assume decisions made by machine are correct"
- "Particularly prone to be swooned by AI generated information"
- ChatGPT example: "bias to believe it because it sounds right"
- "Machine just spit it out in digestible authoritative natural language"
- Incentive to "let machine do lot of work, make decisions quickly, deal with consequences later"
- "Something we may not be able to catch until it's too late"
Acceptance of Imperfection
- "Machines going to make mistakes, do horrible things"
- "Encourage you to bomb building you thought was military target that turns out to be shelter for civilians"
- "But hey guess what humans do that as well"
- Question: "are machines going to reduce incidence of those things?"
Inevitable Future
- "No going back, no one's going to pump brakes on integrating AI technology into warfare"
- "No one's going to wait for ethicists and lawyers to decide if ethical, legal, or safe"
- "Geopolitics will force this forward"
- "All we can hope for is wisdom and foresight from those making decisions"
- "Approach it humanely and responsibly"
