Food Protection Trends - July/August 2020 - 276

* Strengthen preharvest risk assessments, in partnership
with growers and shippers, using drones to monitor and
report produce field conditions, such as pest infestation
or water mismanagement, and
* Collect and transmit continuous temperature data from
transportation fleet refrigerator, freezer, or hot-holding
units during transit (e.g., international, interstate, or
last-mile).
CHALLENGES AND ETHICAL CONSIDERATIONS
Although the opportunities to improve retail food safety
exist, widespread adoption of AI and ML at retail, and along
the food supply chain, will be incremental. There are several
challenges and ethical considerations that the food and
technology industries and regulators must work to define
and address as AI's impacts on society, policy, and business
matters evolve.
Data quality and bias
The quality of AI and ML algorithms depends on highquality input data. Incomplete, biased, or inaccurate data
may lead to harmful decision-making and outcomes. For
example, in applications of AI for deploying public health
resources and food establishment inspections based on
consumer-sourced big data, a concern lies in how individual
consumer bias may be propagated through the food industry
and influence enforcement of food safety and consumerpurchasing habits. Altenburger and Ho (1) showed that
predictive analytics using consumer complaints to target food
safety enforcement disproportionately targeted Asian-owned
establishments. This study concluded that crowd-sourced
information for a single establishment had lower predictive
power than prior inspection history and that these reviews
did not significantly improve predictions of food safety
violations when existing restaurant data and inspection
history was available (1). Thus, data quality will benefit
from continued investments in digital data accessibility,
connectivity, and infrastructure.
Vetting data prior to use in algorithms is increasingly
critical. Methods to characterize and control bias in data
and algorithms are being explored (i.e., statistical debiasing
solutions), but greater transparency and reproducibility in AI
and ML applications are needed if they are to be more widely
adopted. Food safety scientists should play a greater role
in providing the practical and scientific context around the
available data and assumptions, especially when making food
safety decisions based on predictions.
Security of information
AI and ML capabilities are anticipated to become more
powerful, widespread, and vulnerable to malicious attacks
or uses. Maintaining security of the information used and
contained within these systems is a critical and significant
task. The volume of new data sources, tools, frameworks, and
people involved brings many unknowns to internal security
276

Food Protection Trends July/August

and privacy systems. Furthermore, the obscure boundaries,
ownership, and origin of data and algorithms complicate
regulation and governance. This is particularly relevant
as AI systems find application in new and more global
business models around food retail, delivery, and consumer
interaction.
Two broad parts to the security of AI and ML applications
include data security and/or privacy and the security of the
internal algorithm(s). As with securing other company assets,
securing AI tools should include evaluation of (i) potential
breaches in the confidentiality, integrity, or availability of
the data and algorithms; (ii) the impact of types of breaches;
and (iii) the design of protections against such breaches.
Traditional policies and encryptions for data security and
privacy should be reevaluated for new contexts of data access
and sharing. Securing algorithms against adversarial exploits
will require new internal teams and talent for food companies
to ensure AI tools perform and learn as intended despite the
presence of misinformation and adversaries.
Impact on business matters and relationships
As data become a new form of currency and we rely more
on data for decision-making, it is inevitable that business
dealings and relationships will evolve. The expansion of
emerging data systems technology (including IoT, big data,
cloud computing, AI, ML, and blockchain) are changing how
our society lives and works, in terms of hyperautomation
and hyperconnectivity (21). Most notably, the rise of AI
is fundamentally changing the meaning of collaboration,
competition, and innovation (18). Three main themes
arise as companies expand the reach of AI beyond their
internal tasks to encompass mutual supply chain and
business ecosystem challenges: (i) trust, transparency, and
accessibility; (ii) data use and ownership, and (iii) availability
of required resources within organizations.
Although AI has the potential to transform food safety
practices, development of common and shared approaches
to data collection and access will be a central challenge
facing industry-wide efforts. Information sharing, data
security, and the reasons for sharing information will be
important conversations with industry stakeholders. As
mentioned previously, data power the algorithms that drive
AI and ML, yet some data are difficult to collect or obtain
due to perceived risk, time-constraints, rarity of events,
confidential business information, or resource limitations. In
the food industry, many stakeholders worldwide still lack the
necessary digital data infrastructure to generate meaningful
banks of food safety information and intelligence. For
example, the transition to digital food safety data collection
and recordkeeping is ongoing for growers and producers who
must share the information prior to business transactions. A
potential solution may lie in the creation of large aggregate
data pools and practices at local, national, and international
levels for shared use in the design and adoption of risk-based
food safety programs. Another alternative to circumvent

Food Protection Trends - July/August 2020

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