Page 73 - Artificial Intelligence for the Internet of Everything

Intelligent Autonomous Things on the Battlefield 59

3.6 COORDINATION REQUIRES AI

It is essential that agents work collectively to sense and explore the battle-
field: the size and scope of the challenge demand it and agents can fall prey
to hazards or adversaries at any time. Humans are masters of coordination,
able to work together to accomplish large tasks using an array of different
modes of communication, from explicit instructions to codewords to an
unspoken understanding of team roles. Coordination on the battlefield will
require agents to use all of these mechanisms and more, taking inspiration
from nature around us and also from applied logic to demonstrate the ability
to share and sequence. Underlying all of this coordination will be the col-
lective activity necessary to enable the communications networks upon
which higher-level coordination will depend.
A certain amount of abstraction is required as teams become large and
tracking and managing every agent’s identity through changing and merging
battlefield roles becomes impossible. These abstractions are exemplified by
the complex emergent behavior of swarms of fish and birds that move with a
purpose but have no explicit guidance. It has been shown that these phe-
nomena can arise from the interactions of simple rules between neighboring
agents ( Jadbabaie, Lin, & Morse, 2003) and that they are robust, able to
maintain cohesion even as neighbors come and go (Olfati-Saber & Murray,
2004; Ren & Beard, 2005; Tanner, Jadbabaie, & Pappas, 2007). Not only
can these swarms move together, they can also explore and manipulate
the world (Berman, Lindsey, Sakar, Kumar, & Pratt, 2011). These insights
are grounded in application of graph theory to dynamic systems (Mesbahi &
Egerstedt, 2010); all of this work on understanding and replicating swarms
showed how we can synthesize and understand global properties by studying
local ones.
It is not enough to just observe the emergent behavior; we must actively
control the emergent behavior to realize the vision of coordination that can
scale and adapt to meet battlefield challenges. The challenge lies in the fact
that, at a suitable level of abstraction, agents have no identity, yet we must
allocate and control them toward complex tasks. Again, the answers lie in
stochasticity: allowing each agent to randomly determine its own actions
but with probabilities proportional to the number of agents required breaks
the need for identity by letting each agent self-determine whether it will
help (Berman, Hala ´sz, Hsieh, & Kumar, 2009). These probabilities can be
further adapted in a closed-loop fashion to shape the distributions
(Mather & Hsieh, 2011).