Nov 24, 2019
Planet-Centered Artificial Intelligence
Responsible Approaches, Wind Farms, Forums, Research Initiatives and Data Centres
We have to go from placing the desires and needs of the human at the centre of the design project to considering a wider focus on the ecology of planet earth. There is a narrative too that argues we have to go from sustaining to regenerating. In an abstract and yet figurative sense planet earth is dying, species are becoming extinct, our home is increasingly less habitable. I do recognise the importance of the sustainable development goals and I like their framework. In this article I will not be able to cover a breadth, however I will focus on a few points resulting from an existing simplification: reduce, reuse and recycle in the context of artificial intelligence. After this I will go through the possibilities in renewable energy; forums to discuss AI and the climate crisis; research initiatives; and data centres. Therefore the structure is as follows:
- Responsible Approaches to Artificial Intelligence
- AI-enabled Wind Farms
- The Forum of Climatechange.ai
- Research Initiatives for AI and the Climate Crisis
- Talking About Artificial Intelligence and the Climate Crisis
- AI & Data Centres
It is my greatest ambition that 🌏-centered AI brings focus to the what we lack in the current development and the limitations (limits to growth) of our ambitions in regards to applications of new technologies.
1. Responsible Approaches to Artificial Intelligence
1.1 Making Artificial Intelligence Work For Our Shared Ecology
It seems almost strange that these days when we are approaching a more human-centered design that we should go beyond these considerations, and yet it makes sense. Human-centered design is a clear focus in artificial intelligence (AI) today especially in the communities concerned with ethics such as the Stanford Institute for Human-Centered Artificial Intelligence (HAI). They are doing great work to convey the importance of repsonsible use of AI. However there have been large scale climate protests around the world and this has increasingly been recognised by those who build digital technology.
When I was at the climate protests in Oslo I took this picture that went viral. It was because of the strong message conveyed by a young girl from Thunberg to this cardboard I think we need to think more closely.
1.2 Leaders in Technology Must Act Responsibly
There is no room to act irresponsible in technology, not anymore. I have written too much about the current state of the industry, and much like many other industries overall it is not impressive — although there are a few magnificent exceptions that can make you believe it is possible to change.
Indeed Climate Change is a clear point for several technology companies and they have to be to a larger degree, but how do we go about making artificial intelligence climate-friendly, or even sustainable?
The good answer that it doesn’t really work that well. We may have to adopt a mentality applied to other areas of society — namely considering consumption and reusability alongside recycling. We need to reduce, reuse and recycle in the field of artificial intelligence, but how could this look like in practice?
1.3 Reducing the use of Artificial Intelligence
I have a project called #500daysofAI so out of interest I would like to see more great AI projects, however contrary to the hype or obsession I think we should think about where we should not or do not need to use artificial intelligence or machine learning techniques. If you ask yourself the question:
- Do I need applied AI here?
- Could I solve this problem in any other way?
- If I make this solution does it need almost constant cloud connectivity to run and how much will this influence emissions?
- How efficient do I need my solution to be and what is the tradeoff in terms of environmental costs in release of carbon or damage to the environment?
- Where is my datacenter, and how much data am I storing? Do I need all this data or am I just storing because I do not know what to use.
- How can I help other developers, companies, governments and nonprofits reduce their use of machine learning techniques in an area where it may consume too much power or not be needed in the first place?
1.4 Reusing Artificial Intelligence
Have someone built a great solution that you are able to use already which works for the given purpose, great! Because training algorithms can be both time-consuming and release a lot of carbon, especially if you use Amazon data centers as they are more likely to use fossil fuels.
- Does my project exist as an open-source project somewhere?
- Could I partner up with an existing company with a solution that works instead of creating my own from scratch?
- What type of moves can I make locally for digital commons projects that enables more sharing of AI solutions.
- What communities of AI exists around the world that you could tap into and collaborate with? As an example: https://city.ai/
- How can I test that the solutions that I have received or that I am using is climate friendly? This is not easy, but take part in developing standards: https://github.com/daviddao/green-ai
- Have you thought about all the hardware required to run your machine learning algorithms, if you are a large company that will likely be a sizeable amount. Where were these minerals sourced, how will they be recycled and did you ask if they could be reused later?
1.5 Recycling in Artificial Intelligence
Electronics require a lot of energy and resources to build and maintain. It is crucial that every company using technology considers their emissions and consider themselves responsible for recycling every single electronic product that pass through their halls or make that requirement strongly known to all of their business networks. The field of AI is about software, but the field of AI is also of course hardware.
- Do I recycle my own electronic products?
- Does my company recycle all electronic products?
- Does my server supplier recycle all their electronic products?
- Does the supplier to my server supplier recycle all their electronic products?
- Am I aware of the materials that are used in the electronic products that I buy?
- Is the product that I am buying able at all to be recycled?
- Is the product that I am buying open hardware or is it easy to change parts so that I could repair it instead of buying a new product?
1.6 …but Planet-Centered is Not Enough
When I say planet-centered we must keep in mind the concept of sustainability, the world is very much intertwined, and we become increasingly aware of this fact with adverse consequences effecting all humanity. We could say life-centered because it is about all the other species in our ecology, but I think a wholesome approach is to be preferred with plants as well as nature part of this mix.
Being able to sustain ourselves does mean taking care of each other. Using these different concepts together might be fruitful, but it is important that we make sure that all human-centered design does not forget the importance to be responsible in regards to the climate.
The planet needs more responsible technology.
2. AI-enabled Wind Farms
2.1 Can Artificial Intelligence Contribute to More Renewable Energy?
Can Artificial Intelligence Contribute to More Renewable Energy?
This summer a wind farm equipped with AI software navigated a wave of negative wholesale prices in Australia. Thus according to an article in Smart Energy International: demonstrating the power of AI to enhance the performance of clean energy assets. According to the same text it was said:
“This time, artificial intelligence is being used to save the planet,” Schwarzenegger.
However whether this can be attributed to Arnold Schwazenegger I am unsure.
2.2 We Need Wind Power
Different place around the globe is experiencing the challenges of integrating solar, wind, and electric vehicles into electric grids that were designed for coal and gas
AMS’s optimised bidding software mentioned in the article is ‘technology agnostic’ (whatever that is), and substantially increases the revenues of storage, wind, solar and other generation assets in wholesale electricity markets.
2.3 Turbines with AI
Applying artificial intelligence (AI) to wind-turbine data is becoming more prevalent and there is a growing number of use cases for the technology.
- Increased efficiency,
- better spare-parts forecasting,
- reduced downtime,
- and lower unplanned maintenance costs.
An article in Wind Power Engineering mentions a few different points regarding what areas that could possibly be areas for application.
- SCADA data (mean, minimum, maximum, and standard deviations),
- Maintenance data (component replacement dates, lubrication events, etc.),
- Failure histories
- Firmware updates
- CMS data (if available)
3. The Forum of Climatechange.ai
3.1 Machine Learning as a Tool in Fighting Climate Change
Looking around in the AI community, especially in the field going to conferences etc. I did not find enough projects relating to climate change. Luckily I discovered that there were a series of interesting posts about climate change and AI emerging over the last year. Particularly the paper called Tackling Climate Change with Machine Learning looking at both adaptation measures and prevention was a great addition to the discussion.
I hoped there would be somewhere to discuss more and an even more pleasant surprise emerged.
This post is simply a request for you to join the discussion online and a way to help you understand what you can do inside this community, November 2019.
A screenshot of the frontpage of www.climatechange.ai retrieved on the 20th of November 2019
New forum at climatechange.ai — just press ‘forum’
It is really easy to get into the forum and make a profile. After you get in there is an overview of different categories. You can choose to look at collaborations, however, I would recommend you do a few things first.
- Press the icon in the top right corner to edit your profile filling in information about yourself, who you are and maybe updating your picture.
- Go to the ‘Introduce Yourself’ topic to write an introduction and have a look through the various other members who have joined recently. You can see if there are anyone that seem to share your interests in different topics.
- Go check out the meetups to understand whether there is anyone in your local area organising anything.
After this you can browse the collaborations section. In this case, it currently seems more academia, however open source projects could be the case.
You can make a wish in the wishlist as well. Just to make an example I made a post about my wish.
The forum is very new so do not expect too much engagement early on. It would be great if you could join and invite some friends.
A last feature I want to mention is Badges. You can get badges at random moments if you achieve different things predefined in the system.
Last week for example I became the user of the month, which was a pleasant surprise.
I hope this helps you understand the platform, because I truly think it is a good idea to gather people together who all are concerned with contributing in one way or another towards this specific intersection of topics.
3.2 Tackling Climate Change with Machine Learning
One of the seminal works in this community is called Tackling Climate Change with Machine Learning published the 10th of June. I have covered this paper in two other stories I wrote previously.
4.0 Research Initiatives for AI and the Climate Crisis
4.1 There is a Growing Engagement to Consider AI in the Context of Climate Change
It seems more initiatives are starting on climate change and artificial intelligence of late. Most of these initiatives have an applied scientific focus, rather than attempt to combine different types of field to address the urgent issues within the infrastructure that enables AI, however that may be a different story, and there seems to be initiatives on energy relating to AI. I want to focus on two initiatives, an AI track at Stanford and a new centre at the University of Cambridge.
4.2 AI for Climate Change Track at Stanford University
A Bootcamp for Cutting-Edge Research at the Intersection of AI and Climate Change
“The AI for Climate Change Bootcamp will convene teams of Stanford students this Winter quarter to build AI models for addressing key problems in climate change.”
The bootcamp will be led by the Stanford Machine Learning group, under Professor Andrew Ng alongside different communities at Stanford University.
- PhD students in Professor Andrew Ng’s lab
- Faculty of the Department of Earth System Science
- The Department of Civil and Environmental Engineering
- Scientists from Descartes Labs
The Bootcamp starts Jan 10th, 2020. There is a requirement to have completed several computer science courses prior to joining the course.
You can find the bootcamp here:
4.3 AI for the study of Environmental Risks (AI4ER)
“The UKRI Centre for Doctoral Training in the Application of Artificial Intelligence to the study of Environmental Risks (AI4ER) will, through several multi disciplinary cohorts, train researchers uniquely equipped to develop and apply leading edge computational approaches to address critical global environmental challenges by exploiting vast, diverse and often currently untapped environmental data sets.”
There is a split focus within two areas:
- Environmental data classification, integration & analysis
- Environmental modelling
It is a one-year Master of Research (MRes) course with a taught component and a major research element, followed by a three-year PhD research project.
This centre is also focused on the natural sciences, this are the type of candidates they are looking for:
- natural sciences (e.g. physics, chemistry, earth sciences, biology)
- engineering
- computer science
- mathematics
You can find more information here.
4.4 Short Note on Social Science vs. Natural Science
Since I study computer science as well I do understand that there is a very practical aspect to this type of education that requires an understanding of these conceptual frameworks. There has to be a degree of experience within these different fields to understand math, programming etc.
I do on the other hand wish that they opted in to create a space for a different type of participant who had another focus area. It seems if you are a social scientist wanting to combine these studies a heavier focus on the natural sciences is needed before it is possible to access these types of courses.
I have so far not seen any course focused on AI within social science, although there are plenty of Science and Technology studies wherein you could certainly have a research focus on artificial intelligence.
5.0 Talking About Artificial Intelligence and the Climate Crisis
5.1 Drifting Off Into the Distance of a Fast Moving Field
Whenever I start to talk to anyone about the climate crisis and AI I often see people veering off, looking into the distance, loosing attention, or not caring. It may very well be because of my lack of skills in conveying the message, or because of the urgency itself. Initially people are interested when they hear about the topic, easy to be excited about two seeming ‘buzzwords’ bundled together into a heap. While this is fine, there is sure to be confusion and perhaps disappointment when I talk of the problems existing at the current time in the technology industry.
5.2 Existing Problems versus Possible Futures
One way to see this is through the lens of ‘futuring’, in x amount of year — twenty year or so y trend will be ‘booming’ I promise you. If you do not move quickly now, oh well, you are going to be left behind. I read an article about pneumonic tubes, the pneumatic tube mail system. In an article written by Farman.
comparing the clunky automobile and its limitations with the futuristic (and militaristic) missile of the pneumatic tube canister being shot under the Harlem River. Sending mail in canisters pushed by compressed air under the streets of a city was seen as the essence of being cosmopolitan and modern. Image courtesy of the National Archives, Washington, D.C.
The article that Farman has written is called: Invisible and Instantaneous: Geographies of Media Infrastructure from Pneumatic Tubes to Fiber Optics. It is a long title, however in this article an important point about time is brought forward, alongside the cost of infrastructure. Considering the extent of Internet connections today it is not challenging to draw a small line between what was happening then and now with Internet of 5G. In his conclusion he writes.
“Users of emerging technologies are continually told that new systems will speed up the ways they keep in touch with one another. There is no doubt that technologies since the telegraph and pneumatic tube system have indeed sped up human communication; howe ver, users will always have a time lag that makes them wait. The notion of the “instant” that accompanies all of these emerging technologies is never fulfilled yet has an extraordinary impact on how users of these media think about social connection. Central to analyzing this impact is studying how users think about wait times for messages from one another”
This makes me think of the attempt to convey urgency of a slow moving process or several novel concepts blended into one. Not many people go around with their phones worried about the impact of the electronic gear that they carry, nor do I often with what I am writing on to make these posts online. There is a discrepancy between the thought and the action, a disconnect, a disembedding of the relationship between the extended object. Your phone in your hand, connected to server, connected to humans, connected to other being, connected to nature, connected to life, and so on.
A strange connotation would be to say ‘wait’, because the primary meaning is negatively associated. Waiting is not a positive action. Hurrying into a situation, or getting somewhere early is mostly correlated with a positive action. Speed is a positive action, affirmative in work, university, with friends etc. Depending on of course if you arrived early to a party. Farman says:
“How we wait for messages from each other will say more about our social lives in an increasingly urbanized landscape than the latest technologies we use to keep in touch.”
When I wait for the reply of the other party, to ponder their interest in this area, it feels as thought they should care — and this is where I may be wrong. I have to be patient and find out what they are interested in to begin to care. Abstract concepts of planets melting, worlds in trouble and such is a hard sell to for example the consultants that I work with. Many argue that they are very concerned and are moving to do thing, to change, to take actions. However when I follow up with a simple: how? Words don’t come easy. Maybe a cardboard sign would help.
I can ask myself the same question and I will continue to do so, to try to understand better. This chaotic development of artificial intelligence whatever it is and whatever it may become when it is between people from different backgrounds with different ambitions as to what AI should be. AI have pondered the shades of grey in AI projects and AI for good, however it may be time for me to actually attempt to work more applied and help more applied.
Earlier this week I outlined questions that could be used to approach these queries, however I did not go further in outlining each and every way to move forward within each question or category. This is perhaps what I need to do to contribute to not only discussions, but rather actions for more responsible use of applied AI or research in the field of artificial intelligence. If we run and keep running, we cannot outrun the consequences of the execution, frantically attempting to catch up will not do.
Then what?
6.0 AI & Data Centres
6.1 Hey Data Centre Industry Show us Your Numbers
Earlier this year Google Deepmind announced that it had managed to create a drastic reduction of energy cost to cool data centres by 40%. Thus it is argued that this is being used to tackle climate change. Google has also been one of the companies really investing in the transition to renewable energy. On September the 19th 2019 as an example Google announced its intention to spend more than $2 billion in new renewable energy infrastructure.
Thus how we treat this question of data centres is important, but how much do we have to think about. In January 2019 the World Economic Forum wrote an article addressing the issue to give a perspective on the sense of urgency:
“Researchers at Lawrence Berkeley National Laboratory, the International Energy Agency (IEA), and Rocky Mountain Institute (RMI) have concluded that room air conditioners alone — the typical window and split units used in most homes — are set to account for over 130 gigatons (GT) of CO2 emissions between now and 2050. That would account for 20–40% of the world’s remaining “carbon budget” (the most we can emit while still keeping global warming to less than 2˚C above pre-industrial levels — the goal set at the Paris Climate Conference in 2015).”
If room air conditioners have a massive emission, then it is hard to imagine the immense emissions that is occurring on the industrial scale. However the article suggest to solve the problem with R&D as well as increasing the optimal conditions of air conditioners to consume less energy or be more efficient etc. The article is based on a report called Solving the Global Cooling Challenge.
We could draw a similar map with increased data usage leading to increased cooling demand.
6.2 Lack of Reliable Data
The government digital service in the United Kingdom wrote a blog post in October 2019 telling about their approach to large technology companies asking them to reveal their emission numbers. This was due to a wish to find out how much CO2 they produced, but they were denied access:
“We asked our hosting providers, Amazon Web Services, UKCloud and Carrenza, to tell us how much electricity we use, and how much CO2 we produce. Only one of our providers, UKCloud, agreed to give us data about our electricity usage. For Amazon and Carrenza, we made a guess about the amount of electricity we use, assuming a percentage of our monthly bill. In addition, neither Amazon nor Google currently shares information about how much CO2 their data centres produce.”
They mentioned three things for the government to do:
- “Reduce the electricity we need to run our services — for example, by making our digital services more efficient
- Keep asking our providers for better data to help us develop guidance on the most sustainable data centres by region and country
- Update the Service Manual’s guidance — ‘Deciding how to host your service’ — based on what we learn”
It is surely possible to do this in several governments, a start would be to know. I think it is highly likely that the technology companies do not know themselves, because they have to track all units and the energy cost relating to expenses.
Oatly which is a brand using oatdrink to replace milk recently had a striking campaign.
How can we know the amount reduce when we do not know an accurate number in the first place?
Perhaps it is time to say: hey data centre industry, show us your numbers.
When we increasingly start implementing machine learning techniques it is important that we think supply chain sustainability. Anything else would be irresponsible. We can reduce, reuse, recycle — however we need to know which actors that supplies storage that are responsible or not (hint: don’t choose Amazon).
On the Other Hand Tech Companies are Not Completely Complacent in Regards to the Climate Crisis
Facebook says in their annual report in numbers that they are increasing their use of renewable energy sources to 100% in 2020. They have some great statements in short format that I would like to include:
- Increasing their use of renewable energy sources to 100% in 2020
- Renewable energy made up over 75% of their overall electricity mix in 2018. They have steadily increased our renewable energy since 2013 on their pathway to 100% in 2020
- They are setting a science-based target to reduce their emissions by 75%.
- Leading with data centers that are 80% more water efficient than average.
- Our data centers use 80% less water than typical data centers.
It is possible to directly download the data presented.
They also claim that currently:
“Facebook’s annual emissions for the average person on the platform are less than the carbon impact of boiling one pot of water — or making one medium latte.”
Really though?
This however is a statement I find it very hard to believe. The average cost in transportation, servers, operation, etc. is less than making one medium latte?
Do they divide this carbon footprint on all their members globally? If so Facebook has 2.45 billion users.
John Naughton is professor of the public understanding of technology at the Open University. He is the author of From Gutenberg to Zuckerberg: What You Really Need to Know About the Internet. In 2015 he called out a previous ‘bullshit’ report by Facebook written by Deloitte:
It must be said that audacious goals are impressive. Facebook aims for 100% by 2020, while Amazon claims they will be using 100% renewable energy companywide by 2030.
We want these companies to be responsible and truthful, but it is certainly hard to say whether they in actuality are.
I have mentioned self-measurement and self-policing by large companies before, and it is important to recognise that seemingly these numbers have all been gathered by Facebook or in the other case by Amazon.
If you have any research or information on this topic I would be happy to read it. If you work for any of these large technology companies I would be very glad to better understand how you work within this area.
Although I am sceptical I want better solutions for the pressing issues we are seeing around the world with a technology sector that certainly contributes and must take greater responsibility.
After having talked about planet-centered design for a few days I realised there are other writers who have made considerations on this topic. Especially the writings of Anne Galloway is more nuanced than mine and takes into consideration indigenous people as well as a wider perspective on implications.
There are other important aspects that I of course have not considered in this article and I can mention a few on the top of my mind that I have written about previously. There are far more to explore surrounding or within the field of artificial intelligence.
- Labour
- War
- Poverty and Welfare
- Inequality
This is #500daysofAI and you are reading article 174. I write one new article about or related to artificial intelligence every day for 500 days.
My name is Alex and I am a student at the University of Oslo.
