This page contains the 4 day twitter stream with images for the event
DIGITAL EARTH DAY 2020 50TH ANNIVERSARY
AI EDUCATION EVENT
April 19-22, 2020
For the pdf version of this report click
here. Its PDF (14.9Mb download)
On Twitter: #AIforEarthDay @ai_environment
AI AND THE ENVIRONMENT
Post-Event Tweet Summary
Official registered EarthDay Education Event
EarthDay 2020 AI Educational Event Summary
Event Title: Sharing insights on using
AI for Environmental Problems from 16 Projects
First AI and Environment book now Available on Amazon and Harvard Bookstore (by Paige M. Gutenborg,
the book making robot.
Insights and examples
for this event are drawn from the book where 60 scientists shared their
ideas for 16 environmental projects.
For Earth Day this year, much of the activities were online due to
Corona Pandemic. We hosted an AI education event April
19-22 on Twitter and Facebook. These are the collected
Tweets and Posts from those 4 days. The event was
registered with earthday.org through MIT. The materials in the
tweets/posts are based on a lecture given at the AI Center at SRI
Intl. Menlo Park, California, USA. The Tweets are summarized
here for those who were not able to attend/not on Twitter.
For more information see www.aiandenvironment.org
Splash Page Figure for AI and Environment Twitter Account
Event Hashtag: #AIforEarthDay
Twitter Account: @ai_environment
Tweet Total: 75
Day 1 Day 2 Day
3 Day 4
Distribution: Originally tweeted from the US with shadow
tweets from Spain. During the event we were retweeted and
picked up in India, Norway and Germany. The retweeting
locations from those accounts are unknown.
Format: We began adding time stamps midway through as the net was
slowing with traffic and tweets were posted out of order.
BEGIN TWEET RECORD
Day 0 Sat April
#AIforEarthDay Event Announcement:
We are now registered on the http://EarthDay.org
event! For the 4 days up to April 22, we will be
tweeting an education event, with insight across the use of AI for
16 different environmental projects.
Day 1 Sun
Environmental projects have unique patterns of data and systems and
the AI systems to address them can be as complicated and
interdependent as GAIA herself. For example, the data is
not just big. Its Big, Hairy and Lumpy (BHL).
Why is the data for Environmental Projects Big, Hairy and
Lumpy(BHL)? The answer is in 2 parts. The first
part of the answer, part 1, is there is a variety of time
scales and a long history of data collection. Relevant data
(when history is useful) can occur at different time scales.
Why is the data for Environmental Projects BHL? Second, part 2, The
media is complex! Many different kinds of information are relevant
to a single problem - Landsat, vegetation maps, satellite mapping,
output files from simulations, modeling, dsp, sensors, cams, human
How can AI help with these 2 Big Hair Lumpy data issues?
First, AI has temporal scale abstraction/transformation methods
useful for integrating and sharing information across time, zooming
in and out of scales of time and space during problem
Cont'd: How can AI help with these 2 BHL issues? Second, AI
uses ontology methods to create a language of meaning across
different data and meaning can be shared, communicated/explained for
people and machines and used for transforming info with time/scale
Another characteristic of environmental projects is
Complexity. Complexity of systems and data occurs in a very
human context. Many projects require tight knit relationships
between people and AI systems, especially in scenarios with
environmental emergencies. #AIforEarthDay
Complexity Issue 1) Environmental problems are often global problems
involving global communities with implications and constraints from
different cultures and social norms. #AIforEarthDay
Complexity issue 2) These AI Systems require an explanation
facility. It is not an option. Because these systems are
naturally people centered technology. This means the
system needs methods not just to solve problems but languages,
interfaces, ontologies, etc. #AIforEarthDay
Complexity issue 3) User interfaces often play a key role not just
because of the need for an explanation facility but because decision
making and prediction systems are also big and hairy data and
systems. cont'd #AIforEarthDay
cont'd Complexity issue 3: decision making and prediction systems
need - Flexible points of view - Zooming in and out, combining and
viewing different scales of time and space- Ability to explore many
possible predictive scenarios (cpu and data intensive).
cont'd-2 Complexity Issue 3) Finally, decision makers (people) need
to use Prediction Systems with lots of different modeling and
simulation systems that require ‘on the fly’ model / parameter
changes. How can AI handle this complexity??
Complexity- How Can AI Help? In predicting plant physiology under
climate change an AI system helps the user examine more scenarios in
a shorter time period. It automatically generates a prediction
system using only the relevant models, knowledge for each
Day 2. Mon April 20
Hi, Welcome to Day 2 of the #AIforEarthday Event!
Yesterday we began talking about the computational challenges and
patterns unique to environmental domains… Big Hairy Lumpy Data,close
human-machine collaboration, Complexity.
#AIforEarthDay. Today we continue the conversation about
unique computational challenges of environmental domains:
Complexity, the dynamic phenomena of natural systems and
“The Global Multi Scale (GMS) Problem”. You will
discover how Hybrid AI is the key architecture.
#AIforEarthDay We will also look at these computational patterns in
two domains: fire fighting and forestry sustainability. First,
I will fill you in on some background with 3 questions. 1. Who
are these scientists? 2. What environmental domains? And 3.
Why do we need AI/ML?
#AIforEarthDay 1. Who are these scientists? The systems and patterns
I’m discussing come from the work of 60 pioneering AI researchers
from a competitive international AI community known as IJCAI.
They originate from New Zealand, Germany, Canada, US, Italy and
#AIforEarthDay 2. What environmental domains? Our discussions
here come from 16 different environmental domains. I’ve divided them
into two types. The first, “Boots on the Ground,” involves
getting dirty. The second is, "Data, Data, Everywhere."
I list them in the next tweets
#AIforEarthDay 2 cont'd. What environmental domains?
"Boots on the Ground" domains are fire fighting, flood prediction,
sewage and pollution, sustainable forests, water pollution, toxic
algae blooms, recycling and resources, arguments and decisions and
monitoring nuclear tests.
#AIforEarthDay 2 cont’d. What environmental domains? “Data,
Data, Everywhere” domains are Assembling Satellite Data, Forest
Ecosystem Modeling, Weather Bulletins, Weather Forecasting, Sharing
Digital Resources, Biodiversity Cataloging, Plant Physiology and
#AIforEarthDay 3. Why do we need AI/ML?
Each one of the domains is extremely important and diving deeply
into them here is very tempting. But what’s important is
to look at the unique computational qualities of these
domains. We see that these qualities are AI’s raison d’être.
#AIforEarthDay 3 cont’d. Why do we need AI/ML? The
patterns also tell us strategically how we need to budget, what
skills and computational resources we need and they show how
important continuity of teams and resources are for success.
#AIforEarthDay 3 cont’d. Why do we need AI/ML?
Environmental problems are often ill defined with incomplete,
uncertain and sometimes totally absent information. AI has
many tools for making decisions w/ incomplete, uncertain and
#AIforEarthDay 3 cont'd Consider for example, sewage and water
treatment. It affects water supplies, industry, environment..making
decisions during flood conditions, whether to close/open a valve, or
to reduce/increase flow, has implications for evacuation, transport,
#AIforEarthDay Numerical methods require precision, but the precise
size and shape of a body of water (e.g. perimeter of lake,
flow rates, etc.), is rarely known. Distributed sensor intelligence,
fuzzy and qualitative modelling, common sense are but a few of AI's
#AIforEarthDay 3 cont'd Why we need AI/ML?
The data is not just enormous but complex - combinations of
time-series, satellite, flood and water instrumentation, human logs,
etc. They vary in scales of time and space from small
collection of regions to an entire planet.
#AIforEarthDay 3 cont'd Why we need AI/ML? There is also a
need to integrate information and the meaning of that information
across different kinds of media from hydro, spatial, topo and
Landsat maps to images of species and handwritten notes.
#AIforEarthDay 7.53pm. Why we need AI/ML? Machine learning and
(deep)neural nets help us recognize patterns from our sensor data
and networks. Many AI technologies also help people to
consider decisions in the face of uncertainty and incomplete
#AIforEarthDay 7.56pm. Why we need AI/ML cont'd. Common
sense AI methods help explain and share the meaning of data and
decisions in a way that is accessible to people and helps when faced
with incomplete information. Ontologies are a common language for
sharing data's meaning.
#AIforEarthDay 8.08pmThat completes the answers to the 3
questions! We have now answered What is the science, Who are
the scientists, Why we need AI/ML? Next let’s return to
the computational / data patterns of environmental projects and
issue of complexity.
#AIforEarthDay 8.10pm Another aspect of complexity is that Nature is
Dynamic. The environmental projects we develop our AI system
for involve dynamic processes - fire, flood, weather, pollution,
toxic algae bloom…. change (unpredictably) over time…
#AIforEarthDay 8.12 Whatever kind of tool you are building, whether
its for decision making, predictions, aggregators, simulators,
modelling, etc. and whatever AI technology you engage…software
agents, Intelligent User Interface, ML, NN, etc. is affected by
The Global Multi-Scale Problem (GMS) of environmental projects are a
consequence of the fact environmental problems and emergencies often
occur across large regions of the earth. This has practical
implications for the data both in research and in emergencies.
The GMS aspect of environmental projects and emergencies often take
place across multiple agencies, geographic territories, government.
departments - national, local, city, county, dept, etc.
#AIforEarthDay. 11.35pm Not only does it affect the data, it also
#AIforEarthDay 11.40pm. Each agency, territory, department,
government and so on has its own policy on communication, data
sharing, protocols, procedures, resources, coordination,
workflows, etc. It requires Intricate levels of cooperation
and trust, with people and machines.
The GMS problem also has an impact on formats and forms of data and
information. For example, in forestry sustainability,
large areas of the earth coverage are involved in decision making
about timber harvesting and planting.
#AIforEarthDay. 11.47pm Forestry sustainability data comes in
FAST and is collected for decades. It includes
topographic, hydrology, and geology maps, soils
data, remote sensing, ... Agencies have differing access
policies and multiple formats for each type of data.
#AIforEarthDay 11.54pm Today we covered the What, Who and Why of AI
and Environment’s insights and completed the patterns of complexity
and Global Multi-Scale Problems that occur across environmental
projects. That's a wrap for the tweets on April 20! See You
Its now almost earthday!!!! But a day? We need
Day 3. Tues April 21
#AIforEarthDay 10.10pm Yesterday we shared the complexity and global
multi scaling issues that are common among environmental
issues. We answered 3 questions - Who the Scientist are, What
is the Science, and Why is AI/ML important/helpful.
#AIforEarthDay.10.07pm. Its Day 3 of the registered Earth
Day http://EarthDay.org event! For the 4
days up to April 22, we will be tweeting an education event, with
insight across the use of AI for 16 different environmental
#AIforEarthDay 10.16. Environmental projects have unique patterns of
data and systems and the AI systems to address them can be as
complicated and interdependent as GAIA herself.
Complexity, Global Multi-Scaling and Big, Hairy and Lumpy (BHL)
#AIforEarthDay 10.21pm To cope with these very real and daunting
challenges AI programmers and scientists take an approach called
Hybrid AI - when just one kind of AI is not enough. In Hybrid AI
many kinds of systems and subsystems work together. #ai
#AIforEarthDay 10.26pm Hybrid AI involves multiple
AI/ML/NN/robotics/drones/etc. subsystems with intelligent user
interfaces, as well as 'standard' or non-AI computing, modelling,
simulation, image analysis, geographical information systems (GIS),
sensor networks, DSP, etc. #
#AIforEarthDay 10.34 Let’s take a look at the domain of
firefighting. Its a dynamic phenomena-fire evolves with weather,
wind intensity and direction, precipitation and humidity.The fuel
type changes as fire evolves, moves - trees, buildings, industry,
toxins, vegetation. #ai
#AIforEarthDay 10.38 Fire fighting domain cont'd. Another
element of the dynamic nature of fire is its speed also
changes. In fact, all parameters of a fire change rapidly
sometimes unpredictably. #AI #ClimateAction
#AIforEarthDay 10.42 People are a big part of solving our
environmental problems. AI systems design respects close
human-environment relationships, with people as part of the decision
making, input and interfaces. Fire fighting is a good example of
#AIforEarthDay 10.50pm Today's state of the art fire fighting user
support includes mobile units carrying command centers for decision
support that interact with vehicles gathering info and taking action
at many elevations and locations... #ai #ClimateEducation
#AIforEarthDay 10.56 Fire fighting domain description cont'd.
Fire fighting has many traits that we talked about - GMS, BHD,
Complexity, and Human-Machine Collaborations. We will look at
the fire fighting domain and describe a AI system created for a
region in Italy.
#AIforEarthDay 11.02pm Firefighting domain: AI system operation
supports human assessment, decision making, operations and planning
activities. There is a dynamic nature to these operations
because of the dynamic nature of fire situations: time - people need
to think fast. #ai
11.03pm Firefighting domain: Fire emergencies often cross wide areas
of terrain so there are multiple operation and decision making
centers (GMS). This requires coordination and cooperation
across centers (GMS). What about the data? It is
Big Hairy and Lumpy.
#AIforEarthDay 11.07pm Firefighting domain: Relevant
historical cases and their data are helpful, but they can have
different qualities. Time and spatial scales differ -
seconds-days,meter-kilometer. Fire data is always incomplete,
uncertain, and sometimes totally absent.
#AIforEarth 11.15pm Firefighting domain. We now look at an AI
system developed to help with firefighting in Italy. The system is
built by Paolo Avesani, Francesco Ricci and Anna Perini,
I.R.S.T.(Instituto Scientifico Romagnolo per lo Studio e
Tecnologico) #AI #ClimateEducation
#AIforEarthDay 11.21pm Fire Fighting Domain:The system focuses on
decision support in the planning strategy for the first attack(which
resources, where, priority, etc.)First attack is key determiner in
many cases.System design focuses on fire fighting in Italian
#AIforEarthDay 11.24pm Italian Fire Fighting AI System: The Big,
Hair Lumpy Data:1) Infrared and meteo satellite, remote sensed
2)Previous plans* 3) Geographical information,simulation data 4)
Input from other centers, locations, organizations. *different time
and spatial scales
#AIforEarthDay 11.29 pm Italian Fire Fighting AI System:Approach:
Hybrid AI - combine graphical simulation of fire evolution, resource
management tools, and sensor data processing together with AI
systems. #ai #FutureTech #greentech #sustainability #education
#AIforEarthDay 11.34 pm Italian Fire Fighting AI System: Although
the fire fighting domain has lots of Intelligent User Interface
issues and systems, including multiple software agents, the system
description is only covering the AI/ML aspects for supporting first
#AIforEarthDay 12.02pm That's it for day 3 of our EarthDay AI
Education Event! Thank you for caring about these difficult
issues.These tweets will be collected and posted in various places
including here. And now, because it is #EarthDay
I will go outside and look up. Happy EarthDay!
Day 4. Weds April 22. EARTH DAY
#AIforEarthDay 9.18pm Happy Earth Day!
Its our last day of the registered Earth Day
http://EarthDay.org event! For the past 3 days we
were tweeting about insights from 16 different environmental
projects that use AI. The science and the data can be as complicated
and interdependent as GAIA herself. G-AI-A. :)
#AIforEarthDay 9.32pm Yesterday we talked about Hybrid AI, where one
kind of AI is not enough. We also explored fire
fighting, a dynamic phenomena with complexity, human-machine
collaboration, global-multi-scaling, and Big Hairy Lumpy data. #AI
#AIforEarthDay 9.48pm We also explored an AI system from Italy that
supports fire fighting planning and decision making for the first
attack, a key determining moment in fire fighting that determines
which resources, priorities, where, etc. Today we look
at sustainable forests.
#AIforEarthDay 9.55pm Sustainable Forest Domain: Data repositories
built to monitor/observe/study biosphere and trends in vegetation
for a very long time. Forests are the largest vegetative component
on surface of the Earth
#AIforEarthDay 9.55pm Sustainable Forest Domain: #AIforEarthDay
9.58pm Sustainable Forest Domain: Connifer forests provide 73% of
world’s industrial logs. People make decisions and plans
about harvest or not, planting or not, based on the knowing the
current state of living things locally, regionally, globally. #ai
#AIforEarthDay 10:12pm Sustainable Forest Domain: First, the domain
is dynamic, like most environmental projects or domains, forests
-terrain -steep, hilly, mountain, plains,..
-climate - tropical, subtropical, desert..
-condition-new, established, insects, burns..
#AIforEarthDay 10:25 pm Sustainable Forest Domain: Decision
making and planning involves many scientists and image analysis
systems, including Forestry ecosystem modelling and simulation, Topo
Maps, Remote Sensing Analysis, DB Design, Visual Information Design,
#AIforEarthDay 10:30 pm Sustainable Forest Domain: Here's how
complexity and BHL shows up in this domain.. First, Access,
use and analysis of forestry data is problematic due to different
policies, codes of forestry practices across the terrain.
#AIforEarthDay 10:35 pm Sustainable Forest Domain:
Second, Data integration, composition and sharing is challenging
because it varies in Granularity, with different Formats and Media
types that have changed over time as well. Remember data
collection happens over decades.#ai
#AIforEarthDay 10:30 pm Sustainable Forest Domain:
Because there are many organizations, institutions, governments and
so on that collaborate, there are also many different computing
environments where data resides and the software tools to extract
information are widely variable.
#AIforEarthDay 10:48pm Sustainable Forest Domain: The
Swedish/Canadian system uses a Hybrid AI approach. First they
use ML to Auto-Create agents. They built a Training Interface
System for each of the software systems that learns by example from
scientists. #greentech #ai
#AIforEarthDay 10:54 pm Sustainable Forest Domain: As
scientists engage GIS/image analysis tools, sw agents in background
mode grow in sophistication as they watch the interaction and learn
Sequence, Task Circumstances. #ai #greentech #earthday
#AIforEarthDay 10:58 pm Sustainable Forest Domain:
The intelligent user interface techniques generate the SW agents
that work to feed information to other AI systems including a
Derivational Analyzer, a Case Based Reasoner (decisions in the past
can be helpful) and a Planner.
#AIforEarthDay 11:08 pm Sustainable Forest Domain:
The intelligent user interface approach is expandable because it
creates as many software agents as needed, allowing design to flow
from the scientists way of interacting and using the systems and
sharing data. #ai #greentech
#AIforEarthDay 11:15pm Summing UP: For the past few days we have
shared insights on using AI from 16 different environmental
projects.These domains have dynamic phenomena with complexity,
human-machine collaboration, global-multi-scaling, and Big Hairy
Lumpy data. #ai #EarthDay
#AIforEarthDay 11:24pm Summing Up: We looked at the essential
character of two environmental domains- fire fighting and
sustainable forestry. Then we explored the Hybrid AI
architecture for these domains. For more detail on these and
other domains see http://aiandenvironment.org
#AIforEarthDay 11:30pm Earthday AI Education Event Final
Tweet: It has been a labour of love. Thank
you. Thank you everyone for caring and working so
hard. Earthday April 22, 2020. "So long, and
thanks for all the fish" Douglas Adams
End of Summary 4/23/20
*in order to coordinate with Spain, who were helping tweet, and at times I would think of the time in Spain, which was
already the next morning, so occasionally my original tweets had time errors! These
have been corrected here in the report for consistency. It just shows how even small things like cooperating across
several time zones can be challenging!