AI in Food Science and Epidemiology · • “State-of-the-art” for use of AI in the food industry • Need for innovation in food safety and epidemiology • Challenges and limitations

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tialAI in Food Science and Epidemiology

Claire Zoellner, PhDFood Safety Scientist

iFoodDecisionSciences, [email protected]

July 17, 2020

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Outline

• Background• “State-of-the-art” for use of AI in the food industry

• Need for innovation in food safety and epidemiology

• Challenges and limitations

• Potential for AI at retail to improve food safety• Supply chain traceability

• Customer feedback and interaction

• Retail environment controls

2

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Buzz-Worthy AI in Retail and Food Service

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Key Questions

• Where is the greatest immediate utility for AI in food safety?

• How do we gain experience or learn about managing food safety risks from AI?

• Are there any new governance or operational risks to consider when adopting AI at retail and food service?

• Where is the future of AI applications in food safety and what is the process for adoption by food retail and quick service companies?

4

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Need for Innovations

in Food Safety:

1. Prevention2. Detection3. Response

5

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

4 Core Elements:Tech-Enabled Traceability

Smarter Tools and Approaches for Prevention and

Outbreak Response

New Business Models and Retail

Modernization

Food Safety Culture

6

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Challenges and Industry Limitations

Friedlander and Zoellner, Food Protect. Trends 2020; Bekker, 2019 7

Data volume, quality, and bias

Trust, transparency, and accessibility in business matters

Security of information

Resources required within organizations

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Potential for AI at retail to improve food safety

Supply chain

traceability

Retail environment

controls

Customer feedback

andinteraction

8

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

• Standard lot-level traceability creates traceability data at each step of the supply chain – resulting in often 5 or more silos

• Standard lot-level traceability makes it difficult for trading partners or regulatory entities to traceback a single consumer item across the supply chain, or to trace forward to all the places where a contaminated item or ingredient went

The food supply chain is complex: “one-

up-one-back” creates data silos

9

Supply chain traceability

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

• Item-level tracing provides traceability from seed to shelf, including last mile to consumers for transparency and traceback to origin

• Integrated with food safety compliance data, quality inspections, and shipment tracking for full supply chain visibility

• Potential for use with dynamic-pricing models

• Unique encrypted traceability codes provide ability to verify provenance and authenticity of product

• Ability to traceback from consumer package helps expedite and narrow investigations & recalls

10

Supply chain traceability Item-Level Traceability

Currently used for high-value crops and export markets

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

HarvestMark Item-level Trace Activity Dashboard 11

Supply chain traceability

Can Traceability Precision Allow for ML to

Identify Food Safety Risks?

12

Customer feedback and

interactionProduct + Data as a Platform

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Use of consumer data & restaurant reviews in epidemiology: target inspections, identify foodborne illness clusters, adulteration events

Start up company iwaspoisoned.com founded in 2009 to crowd-source food poisoning complaints

New York City partnered with Yelp to prioritize health department investigations (2014)

Chicago Dept. of Innovation and Technology develops algorithm using public data to predict health code violations (2014)

The Food Safety STL Project: using Twitter to identify and respond to food poisoning (2015-2016; Harris et al., 2017)

Deploying nEmesis: preventing foodborne illness by data mining social media (Las Vegas, 2015; Sadilek et al., 2017)

Boston & Yelp fund a Harvard research tournament to mine Yelpreviews to predict food-safety violations (2016)

MIT/FDA collaborate to predict risks for economically motivated food adulteration using public data (Gu, 2016; Huang et al., 2017)

FINDER: machine-learned epidemiology deployed in Las Vegas and Chicago (2016-2017; Sadilek et al., 2018)

13

Customer feedback and

interaction

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Product Home Page & Social Promotion

Traceback to Farm& Processing

Consumer Feedback

Food Safety Status &Real Time Compliance

Data with iFood

Recipes & Other Custom Content

14

Customer feedback and

interaction

Can Item-level Consumer Traceability

Provide Data for CPG Epi Algorithms?

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Retail environment

controls

15

Robotic monitoring of case temperatures

Augmented reality-based trainings and task lists

Vision systems to detect product labeling issuesSimulation-based risk assessments with analytics

AI for Preventing Conditions that Increase

Food Safety Risks at Point of Sale

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Friedlander and Zoellner, Food Protect. Trends 202016

Potential for AI at retail to improve food safety

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Key

take

aw

ays

Food retailers are preparing for a highly automated future

AI adoption is not currently motivated by food safety; opportunity to improve outbreak response with AI solutions

AI is enabled by software that provides connectivity and transparency of supply chain food safety data

AI must provide access to actionable information to leverage experience and resources for managing food safety risks

AI helps existing food safety practices work more effectively

Perceived liability, data bias, and accessibility are key challenges to widespread adoption

17

iFo

od

Dec

isio

nSc

ien

ces,

Inc.

Co

nfi

den

tial

Claire Zoellner, PhDFood Safety Scientist

[email protected]