IEEE - Aerospace and Electronic Systems - July 2021 - 70

Artificial Intelligence and Data Fusion at the Edge
an event. For example, a complementary set of
cameras in a surveillance application can provide
an extended picture ofthe scene which simplifies
the subsequent tracking ofa target.
c) Cooperative sensorfusion. In cooperative sensor
fusion, a sensor is configured and/or positioned
based on the information from other sensors to
generate more useful measurements. For cooperative
sensor fusion, either sensors can autonomously
collaborate to configure each other or
some input from human expert can be provided.
For example, a tracking task can require adapting
the camera angles after observing behavior of a
target.
(2) Knowledge/Feature Fusion. The fused sensor/pixel
data provide a basis for feature extraction that develops
a model for the underlying data to conceive patterns
in the data. The abstraction of fusion
components increases with the level of fusion. In the
intermediate fusion levels, the data are available in
the form of models that represent knowledge from
the observed event. The knowledge fusion can be
performed at model level or parameter level. In
model fusion, the knowledge is represented in form
of different models, which are fused together. An
example of such a model is Gaussian model that provides
information about the distribution of data. The
mixture of Gaussian distributions produces a Gaussian
mixture model (GMM), which describes the distribution
of a dataset that is more complex than a
unimodal Gaussian. The fused model contains more
precise knowledge about the overall data distribution.
CNNs and multiple kernel learning based ensemble
methods are other examples of model fusion techniques.
In parameter fusion, parameters of different
models are fused together [22].
(3) Decision Fusion. At the highest level of fusion, the
goal is to improve decision making and choice of
actions. Decisions obtained based on different models
can be fused together to obtain better decisions.
The decision fusion of multiple models/classifiers
can either consist of direct combination of the decisions
from the individual models or can select a specific
model/classifier for a given input. By observing
the impact of a chosen action, the entire fusion process
can be adapted for performance improvement
and better decision making.
TYPES AND STAGES OF AI
Since AI research profess to make machines emulate
humans, the extent to which an AI system can imitate
human capabilities is used as a criterion to define types of
70
AI. AI can be classified into four main types based on their
functionalities. Type I AI-reactive machines belong to
the most basic type of AI systems that are purely reactive
and do not have the ability to form memories or use past
experiences to inform current decisions. These machines
can only be utilized for automatically responding to a limited
set or combination of inputs. A famous example of a
reactive AI machine is IBM's Deep Blue, a supercomputer
that beat chess Grandmaster Garry Kasparov in 1997 [24].
Type II AI-limited memory machines in addition to having
the capability of reactive machines have the ability to
learn from historical data to make decisions though this
memory is limited and transient. Nearly all existing AI
applications (e.g., chatbots, virtual assistants, and autonomous
vehicles) fall under this AI category. The next two
types of AI exist either as a concept or work in progress.
Type III AI-theory of mind is used to represent
a machine (AI agent) that has the ability to form a predictive
model of self and others and have the ability to represent
and discern the mental states of others, including
their emotions, desires, beliefs, and intentions. Theory of
mind AI can provide intelligent machines/robots with
powerful capabilities, in particular, social intelligence for
human-machine interaction [25]. Type IV AI-selfawareness
is an extension of theory of mind AI and is
often regarded as the ultimate objective of all AI research.
Self-awareness AI refers to an AI agent that has consciousness
and has the ability to form representation of
itself and others. Self-aware AI agents know about their
internal states and can predict the feelings and actions of
others. This type of AI will not only be able to understand
and induce emotions in those it interacts with, but also
have emotions, needs, beliefs, and likely desires of its
own 24. Although self-aware AI can potentially boost our
progress as a civilization tremendously, it can also possibly
lead to catastrophe because self-aware AI would have
the capability of developing ideas, such as self-preservation,
and outmaneuver the human intellect to plot elaborate
schemes to take over humanity 24. Consequently, AI
safety has been gaining traction in AI research and nonprofit
organizations [26].
An alternate system of classification that is more prevalent
in AI community is the classification ofAI into different
stages, viz., artificial narrow intelligence (ANI), artificial
general intelligence (AGI), and artificial super intelligence
(ASI). ANI represents all the existing AI even the most
complicated ones including deep learning. ANI refers to
those AI systems that can only perform a specific task (e.g.,
driving and speech recognition) with human-like capabilities.
AGI refers to the capability of AI to learn, perceive,
understand and function like humans. AGI will independently
build multiple competencies and generalizations
across various domains thus massively reducing the time
needed for training. AGI will make AI agents just as capable
as humans by replicating the multi-functional abilities of
IEEE A&E SYSTEMS MAGAZINE
JULY 2021
IEEE - Aerospace and Electronic Systems - July 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - July 2021
