AI for Agriculture

Simon-Pierre Genot

Page 2

Why talk about AI in context of Agriculture?

What exactly is AI?

Are we really in the ”AI Age”?

What could be the impact of AI on agriculture?

Objectives

Page 3

Item1. Why talk about AI and Agriculture?

2. Short intro to AI

3. Are we really in the AI age?

4. Potential impact on Agriculture

5. Questions/Discussion

Agenda

Seite 4

What do AI and agriculture have to do with each other?

General trend in agriculture: use technology to do more with less

Seite 5

1970 - 2005

Computer is a calculation machine with simple I/O that follow rule based programming.The major benefit: they can automate simple mechanisms

1st Age of Computers: the microcontroller

Seite 6

2005 – today

Allows computers, sensors, people to communicate with each other all over the world.

2nd Age of Computers: the network

Seite 7

? - ?

Computers can automate complicated processes & decisions

?

3rd Age of Computers: Artificial Intelligence

Page 8

Item1. Why talk about AI and Agriculture?

2. Short intro to AI

3. Are we really in the AI age?

4. Potential impact on Agriculture

5. Questions/Discussion

Agenda

„AI is a tool for optimization based on data to improve decision making.”

Kai -Fu Lee

Page 9

There are many comparable definitions existing, like• „The field of computer science dedicated to solving cognitive

problems commonly associated with human intelligence, such as learning and problem solving […]“

• „AI is just lowering the cost of prediction“

One puts it quite simple and precisely:

What is AI? – A general definition

Seite 10

The most prominent AI technique: Machine Learning is a paradigm change in how we solve problems

Problem + Data

Developer looks at data and creates rule-based algorithm:

If x> 10 then: …Solution

Developer feeds the machine the data

Problem + Data The machine learns the ‘rules’ from the data only

Traditional Programming:

Machine Learning Programming:

31.07.2019Abteilung, Verfasser Seite 11

Machine Learning

Supervised Machine Learning

Unsupervised Machine Learning

Reinforcement Learning

• Deep Learning• Classification• Prediction

• Clustering• Ranking• Recommendation

• Logistics• Game AI• Robotics

Machine Learning consists of three types of techniques: supervised, unsupervised and reinforcement learning

1

2

3

Seite 12

Healthy Sick

Labeled dataset

Learn the decision function: f

?

Use f to decide if healthy

Automatically detecting disease in crops from pictures is supervised machine learning1

Seite 13

Healthy

Healthy

Sick

Healthy

Sick

Dataset

x y

Supervised machine learning works very similarly to how humans learn: by trial and error, with a bit more math

Healthy

Take a datapoint Apply f

f( ) =

We try to learn the best function f by correcting it when it makes mistakes:

Healthy: good!

Sick: Mistake, let’s update f

1

Seite 14

1 How do we represent a leaf so that a machine can understand it?

Features: a data representation of an object

Many different ways to represent data.This will have an impact on the results.

Number of holes

Green intensity

sick

healthy

Seite 15

1 How do we choose the decision function and how do we learn?

Number of holes

Green intensity

sick

healthy

Functions can be very complicated or have many parameters we can update

f = ax + b

Seite 16

1 How do we choose the decision function and how do we learn?

Number of holes

Green intensity

sick

healthy

Functions can be very complicated or have many parameters we can update

f = ax + b

Here we make two mistakes

Seite 17

1 How do we choose the decision function and how do we learn?

Number of holes

Green intensity

sick

healthy

Functions can be very complicated or have many parameters we can update

f = a’x + b’

If we update the function, we only make one mistake

Seite 18

The features and the function can be very complicated

Most common example of a function with a lot of parameters to update: Deep Learning

1

Input data Output

Seite 19

Success criterion for Supervised Machine Learning1

Collect the right dataset Build good features Choose the correct Machine Learning Algorithm

+ +

Seite 20

Clustering is a simple type of unsupervised machine learning

Unsupervised Learning: discover hidden structure/patterns in data

2

• Recommendation engine

• Search engines• Drug discovery

Number of holes

Green intensity

Seite 21

Clustering is a simple type of unsupervised machine learning

Unsupervised Learning: discover hidden structure/patterns in data

2

• Recommendation engine

• Search engines• Drug discovery

Number of holes

Green intensity

Seite 22

Video game AI are often a type of reinforcement learning3

State & Rewards

Actions

Can we teach the computer to take the best action given the state of the game?

[Source: Farmville]

Seite 23

Video game AI are often a type of reinforcement learning3

[Source: Google]Youtube link

https://www.youtube.com/watch?v=V1eYniJ0Rnk

Seite 24

But reinforcement learning can do many more complex things…3

[Source: Tech Insider]

Youtube link

https://www.youtube.com/watch?v=gn4nRCC9TwQ

Page 25

Item1. Why talk about AI and Agriculture?

2. Short intro to AI

3. Are we really in the AI age?

4. Potential impact on Agriculture

5. Questions/Discussion

Agenda

Page 26

2000

2019

AI The recent rise of AI is do to two key factors: computation power and data

[Source: Cumulus Media][Source: https://en.wikipedia.org/wiki/Moore%27s_law]

https://en.wikipedia.org/wiki/Moore%27s_law

Page 27

Augmented RealityHumans interacting with thedigital world

AI is supported by progress in many other supporting technology trends –which sometimes get mixed up with AI

PlatformsHumans interacting witheach other through the digitalworld

RoboticsThe digital world acting upon the physical world

IoTThe digital world sensing the physical world

Artificial Intelligence

Page 28

The business plans of the next

10,000 start-ups are easy to

forecast: Take X and add AI. This

is a big deal, and now it‘s here

Kevin Kelly Founding Executive Director, Wired Magazine

❞

❝

Financing for AI has grown exponentially… …across all industries

AI has been at the center of attention recently

[Source: CBInsights]

5891.039

2.6773.125

5.021

160

253

364

493

658

0

100

200

300

400

500

600

700

0

1.000

2.000

3.000

4.000

5.000

6.000

7.000

Deals ($M) Disclosed Funding ($M)20182014

Page 29

Share of global AI investment & financing by country from 2013 to 1Q’18

Top AI startups 2019 according to CB Insights: only one German startup in the list

AI is becoming a priority in many countries for investment as well as for founders – but Germany is still behind

[Source: CBInsights][Source: Statista]

Page 30

Item1. Why talk about AI and Agriculture?

2. Short intro to AI

3. Are we really in the AI age?

4. Potential impact on Agriculture

5. Questions/Discussion

Agenda

Seite 31

The simple AI strategy formula for companies:

Talent + Data = Automation

Hard to find but not bottleneck The REAL value Value creation

AI will force farmers to think about their data

Seite 32

Type Use case Impact for farmers

Agric

ultu

ral

Rob

otic

s

Robotic de-weeder powered by an image recognition AI

- Reduce herbicide costs- Reduce regulatory risks

Robotic apple harvester - Reduce harvesting costs

Cro

p &

Soil

Mon

itorin

g

Fertilizer, input, irrigation optimization recommendation - Reduce input costs

Using image recognition, detect diseaseRecommend best course of action

- Reduce plant protection costs- Reduce risks

Pred

ictiv

e An

alyt

ics

Better weather forecasting for agriculture

Growth models forecasting

Type of AI Image recognitionSupervised ML

Image recognitionSupervised ML

Physical models

Image recognitionSupervised ML

Physical models

- Predict harvest volume- Help planning

- Reduce risks- Help planning

Physical models

AI will help increase revenue and lower costs for the farmer, completing the promise brought forward by digitization

Seite 33

Remote farming Sustainability?Vertical farming

Could AI enable new production models?

Seite 34

AI for Agriculture��Simon-Pierre GenotFoliennummer 2Foliennummer 3Foliennummer 4Foliennummer 5Foliennummer 6Foliennummer 7Foliennummer 8Foliennummer 9Foliennummer 10Foliennummer 11Foliennummer 12Foliennummer 13Foliennummer 14Foliennummer 15Foliennummer 16Foliennummer 17Foliennummer 18Foliennummer 19Foliennummer 20Foliennummer 21Foliennummer 22Foliennummer 23Foliennummer 24Foliennummer 25Foliennummer 26Foliennummer 27Foliennummer 28Foliennummer 29Foliennummer 30Foliennummer 31Foliennummer 32Foliennummer 33Foliennummer 34