1 st DRAFT ITEC 2020, Excel, London, 27-30 April 2020 #ITEC2019 - - PowerPoint PPT Presentation

#ITEC2019

1st DRAFT

Artificial Intelligence, Autonomy and Human Robotic Interaction in Defence at the Times of the Digital Twins

Walter David, Italian Army Alexandra Stefanova, United Drone Community walter.david@esercito.difesa.it

ITEC 2020, Excel, London, 27-30 April 2020

#ITEC2019

Introduction Drones and Artificial Intelligence Lethal Autonomous Weapon Systems Trust, Legal and ethical issues, risks for security Role of simulation Conclusions and Recommendations

Agenda

#ITEC2019

Introduction

• Fourth Industrial Revolution:

combining hardware, software and biology into complex cyber-physical systems

• Autonomous systems lead the Third

Warfare Revolution toward ever- increasing autonomous systems

• Deployed weapons already at high

degree of autonomy.

• Super aEgis II sentry turret,

Samsung SGR-A1 semi- autonomous sentry gun.

#ITEC2019

• Small drones proliferating as asymmetric weapons
• Convergence of drone with affordable AI resources products (for deep

neural networks) already present more risks for security.

• Unmanned and Autonomous Systems (UAxS) to transform warfare but

their use poses multidimensional challenges.

Drone and AI Challenges

#ITEC2019

• Need for multi-dimensional approach
• Brainstorming, analysis workshops in the military and innovation sector.
• United Drone Community (25 g) Mini SeeK3R
• to connect with a neural network, detect and recognize over 100
• bjects, potentially able to exploit online datasets like Arxiv,

Github, Paddle, Keras, MXnet, etc.

• Drones of such size are not covered by EU regulations.
• Small Unmanned Ground Vehicle VIPER 2 MIL for surveillance,

reconnaissance, monitoring and inspection of areas at risk, (e.g. urban buildings, narrow spaces and tunnels, providing real-time information gathered from cameras (visible /IR) and CBR sensors.

• UGV technical challenges

Approach

#ITEC2019

UDC Mini SeeK3R

Mini Drone Threat

#ITEC2019 TRP-2 Combat - Leonardo anti-IED robot Dragon Runner, Caddick/AFP/Getty Images

BEAR - Battlefield extraction assist robot

Unmanned Ground Vehicles

VIPER 2 MIL – Ingegneria dei Sistemi 7

#ITEC2019

• The perfect soldier, never violating its orders.
• Extending the warfighter’s reach. o reach deeper
• Force multiplication. Reduction of (human) soldiers. Remove soldiers

from most dangerous missions

• Better intelligence and situational awareness.
• Casualty Reduction of Civilians, and Combatants. Reduction of

destruction of critical infrastructures, buildings, properties

• Support to disaster management and humanitarian action

Robotic systems military pros

#ITEC2019

Lethal Autonomous Weapon Systems - LAWS

• Lethal Autonomous

Weapons Systems do not comply with International Humanitarian Law principles:

• Distinction.
• Proportionality in

attack.

• Precautions in attack.

? ?

#ITEC2019

• Trust for humans to team with machines at high levels of autonomy.
• Training datasets for machine learning systems in military should

perfectly represent real operational data.

• Lessons learned and standards from automotive technologies for self-

driving vehicles.

• Live, virtual, and constructive simulations a safe, risk-free fashion of the

AI systems’ behavior.

• Live digital twin model of real world to test missions and to identify

failures (unanticipated extreme weather /operating conditions, dangerous hw and sw faults

Simulation for AI Systems

#ITEC2019

Conclusion

• Man-machine teams to blend human and AI capabilities
• Deep learning to support decision-makers with rapid planning

generation

#ITEC2019

Solutions & Recommendations

• From the legal perspective
• A meaningful human control or appropriate levels of human

judgement must be retained over weapon systems and use of force,

• with the human able to intervene in real-time

#ITEC2019

Solutions & Recommendations

From the factual/operational perspective

• Ethics in AI systems’ codes of conduct, based on Rules of Engagement

and laws,

• Digital twin models of robotic system and the environment to be

applied to test the AI behaviour under rare and unexpected conditions.

• AI trained in crucial aspects of operational planning and decision-

making in complex realistic scenarios

• urban areas that include models of civilians, wounded combatants,

humanitarian actors, weather, road traffic conditions, natural disasters.

#ITEC2019

Questions & Answers