I am critical of energy requirements of applied artificial intelligence (AI). As an example training a single AI model can emit as much carbon as five cars in their lifetimes.
Taking this into consideration the intense applications within new army intelligence systems could be a possible drain on the resources of our planet.
On the flip side it is also important to consider possible benefits for society that can be explored within the field of AI.
Before saying anything I do not necessarily strongly believe it is that easy, to immediately say good or bad, however let us explore these two opposite notions within the field of AI.
Do you want the good news or the bad news?
AI for Good
In this context there is a conference that caught my attention, because I follow the progress of Iris.ai who want to improve research through AI applications going from millions of documents to a precise reading list in a short time. If we can make it easier for researchers to navigate science it is interesting to think what progress that could be made in society.
Iris.ai is one of the companies that I admire the most within the field of artificial intelligence. I noticed that they posted a picture of their CEO Anita Schjøll Brede talking at AI for Good, so I decided to check it out.
AI for Good is a United Nations platform, centered around annual Global Summits, that fosters the dialogue on the beneficial use of AI, by developing concrete projects. The first global summit was in 2017, and it has become a yearly event. The most recent event was last month in the end of May. One of the first talks of the event caught my eye.
Society is demanding for understanding and at a critical juncture in examining AI impact on humanity. To succeed, these efforts requires developing a holistic and inclusive perspective to help us make considerate decisions on how we integrate AI solutions in the world. Join us for this forward-thinking session and discover insights, stories and perspectives on how we can harness its potential to meet humanity’s greatest challenges.
For who and What Questions?
AI for Good claims a multi-stakeholder and inter-disciplinary approach with government, industry, UN agencies, civil society, international organisations and academia present.
Immediately scrolling down I could see names from charities and large tech companies or a combination of the two. A few names jumped out that may be of interest, they all seem like interesting people, I will just mention a two people and two questions in the multitude of talks.
Timnit Gebru, Lead Ethical Artificial Intelligence, Google & Co-founder, Black in AI
Nancy Nemes, Founder & Chief Enthusiasm Officer of Ms. AI – Artificial Intelligence for and with women
- Could progress of the SDGs be monitored in (near-) real-time perhaps?
- Can AI technology be used to aggregate multiple real-time, often unstructured data sources into an analytical model to charter the progress of the SDGs?
There were of course many more topics that were covered and you can read about all of them in the programme of the AI for Good official website.
So Let Us Return to the Less Good News
As mentioned in the introduction, and as questioned in previous articles there is a question in regards to energy. This question is to some degree up in the air, however a recent article provides an interesting comment in this regard, putting it into perspective.
The artificial-intelligence industry is often compared to the oil industry: once mined and refined, data, like oil, can be a highly lucrative commodity. Now it seems the metaphor may extend even further. Like its fossil-fuel counterpart, the process of deep learning has an outsize environmental impact.
-Karen Ho, MIT Technology Review
This article was based on a new report from University of Massachusetts, Amherst, where a life cycle assessment was made for training several common large AI models.
The report recommends:
“…a concerted effort by industry and academia to promote research of more computationally efficient algorithms, as well as hardware that requires less energy. An effort can also be made in terms of software.”
So since neural architecture search was named in the comparative chart posted by MIT Technology review let’s look closer at what that means.
Neural architecture search (NAS) is a technique for automating the design of artificial neural networks (ANN), a widely used model in the field of machine learning. NAS has been used to design networks that are on par or outperform hand-designed architectures.
Artificial neural networks (ANN) or connectionist systems are computing systems that are inspired by, but not necessarily identical to, the biological neural networks that constitute animal brains. Such systems “learn” to perform tasks by considering examples, generally without being programmed with any task-specific rules.
This may sound confusing if you have no prior understanding of machine learning techniques or the field of artificial intelligence. However in a reductive explanation or gross simplification of what it often means in practice is that: you ask through programming language the best way to complete a specific task and the program tries out different combinations. These combinations can be in the thousands, millions or more which requires processing power (CPU — the ability of a computer to manipulate data) which emits carbon due to the hardware draining energy, using electricity.
There is an interesting table in the research article worth considering.
This should be a wake up call for those working within the field of artificial intelligence. Responsible use of resources has to be the case and there should be calculations in place to make sure projects within the field of artificial intelligence is aware of the footprint they have.
Was This Mentioned at AI for Good?
Looking at the list of talks at the AI for Good conference on the different days I do not see this featured in any discussions. In this regard I realise I may be completely wrong, as I was not in attendance, however general discussions may have mentioned this aspect. Please if you were there prove me wrong.
If we consider the high energy requirements and sustainability — this is a point that needs to be pervasive, not brushed off as unimportant. Energy and policy considerations for deep learning is of high importance to discuss despite the possible good that can come from applied AI within different fields. We need to think about both the good and bad of attempting to contribute as best we can.
In my introduction of the article I mentioned the defence industry and possible high energy requirements in these applications, however it may be considered across the board. Will companies, nonprofits and governments consider this in their strategies as the climate crisis proliferates?
AI may do much good, however it does as with many other aspects of our society need to be handled with great care.
We find ourselves in a dire time where this is an absolute requirement.
There are still many grey areas of the field of AI to be explored.
So with that day five of #500daysofAI is over and out.
What is #500daysofAI?
I am challenging myself to write and think about the topic of artificial intelligence for the next 500 days with the #500daysofAI. It is a challenge I invented to keep myself thinking of this topic and share my thoughts.
This is inspired by the film 500 Days of Summer where the main character tries to figure out where a love affair went sour, and in doing so, rediscovers his true passions in life.
I hope you stick with me for this journey and tell me what you think!