Food Protection Trends - July/August 2020 - 273

A review of some basic definitions and methods associated
with AI will lay the groundwork for a discussion of next
generation food safety systems:
** Intelligence: the mechanism(s) related to the ability to
acquire and apply knowledge, reason deductively, and
exhibit creativity to achieve goals (13, 23).
** AI: the science and engineering of making intelligent
machines, especially intelligent computer programs
to sense, reason, engage, and learn. AI systems display
intelligence by analyzing a given environment and taking
actions, with some degree of autonomy, to accomplish
specific goals (4, 13). AI systems require data to
define and improve their behavior and goal-seeking
performance. AI-based systems may be purely software
based or can be embedded in hardware devices.
** Data: more or less structured and standardized
information that can be processed by computers,
usually in the form of text and numbers (5). Data are an
important asset used as input to the algorithms at the
core of AI applications.
** Big data: data collected in massive volumes (e.g., via
the internet and smartphones), with increasing variety
and pace, such that they are too large or complex for
traditional data processing. Characteristics of big data
are commonly summarized by four Vs: volume, velocity,
variety, and veracity (11).
** Data mining: application of a combination of machine
learning, statistics, and database system methods to
discover patterns in large data sets.
** Algorithm: an unambiguous set of rules, written by
humans and followed by a computer, that translate input
data into outputs for solving a class of problems (5).
Algorithms can perform calculations, data processing,
and logic tasks.
** Machine learning (ML): the science of constructing
computer programs (or algorithms), drawing from many
fields, including statistics, philosophy, biology, and
information theory, that automatically improve with
experience or data by establishing patterns and rules
that can be applied to new, subsequent data (6, 14). At
a minimum, application of ML requires a learning task
and a performance metric. There are three groups of ML
methods:
* *Supervised
learning:
learning
a function
thatthat
maps
an
Supervised
learning:
learning
a function
maps
input
an output
based based
on labeled
input-output
data
an to
input
to an output
on labeled
input-output
pairs
(training
data). data).
data
pairs (training
* Unsupervised
learning:
learning
patterns
in unlabeled
* Unsupervised
learning:
learning
patterns
in
input
data. input data.
unlabeled
* Reinforcement
learning:
learning
behavior
based
* Reinforcement
learning:
learning
behavior
based
on on
trial-and-error
interactions
within
a dynamic
trial-and-error
interactions
within
a dynamic
environment,
balancing
between
exploration
of the
environment,
balancing
between
exploration
of the
unknown
andand
evaluation
of current
knowledge
(12).
unknown
evaluation
of current
knowledge
(12).
** Deep learning: more advanced ML involving layers of
algorithms and decision networks that progressively

extract higher-level features from raw input, typically
based on millions of data points.
** Internet of Things (IoT): a global infrastructure
enabling the connection of interrelated physical and
virtual objects (e.g., computing devices, mechanical
and digital machines) to transfer data without requiring
human-to-human or human-to-computer interaction.
** Predictive analytics: a variety of statistical, data
mining, and ML techniques to analyze current data
and historical facts to make predictions about future
or otherwise unknown events (17).
** Robotics: an interdisciplinary branch of engineering
and science that deals with the design, construction,
operation, and use of robots (as well as computer
systems for their control, sensory feedback, and
information processing) to perform tasks either alone or
with human direction (15).
** Virtual reality (VR): a simulated experience, similar or
completely different from the real world, in a space that
allows for human sensory interaction and overcomes
spatial and physical limitations of the real world (3).
** Augmented reality (AR): an interactive experience
of a real-world environment where objects in the real
world are enhanced by computer-generated information,
including images, sounds, textures, and scents.
At a high level, AI does not refer to "one thing" but rather
is an umbrella term for a variety of current technological
developments and processes (6). Where ML is focused on
the ability to learn and make predictions, AI encompasses
the ability to sense, reason, engage, and learn. Use of an
algorithm or predictive model does not always warrant the
label AI or ML; the utility of AI depends on algorithms as
capable as people at solving problems. The broad purpose
of AI could be considered as eliminating repetitive and
laborious tasks with automation and automated decisionmaking, but the specific branches of AI present different
solutions depending on the business problem. There are
myriad ways to identify branches of AI and their related
computational methods or mathematical logic, but for
this article aimed at the food retail industry, applications
of AI tools may be categorized as (i) vision, (ii) text,
(iii) interactive, (iv) analytical, or (v) functional solutions
(2) (Fig. 1).
Visual AI tools encompassing image recognition, computer
and machine vision, and AR offer new technology for tasks
related to product scanning and employee training. Textfacing AI tools recognize text or speech for the purpose of
text-to-speech and speech-to-text conversion. Interactive
AI tools often utilize expert systems and natural language
processing for capabilities of extracting information from
customers to provide troubleshooting and customer service.
Analytical AI tools, heavily based on ML, provide predictive
and actionable information useful in risk assessments,

July/August Food Protection Trends

273

Food Protection Trends - July/August 2020

Table of Contents for the Digital Edition of Food Protection Trends - July/August 2020