At his inaugural keynote of the QCon London conference, Leslie Miley, technical advisor for the CTO at Microsoft, spoke about AI Bias and Sustainability. The march towards transformative technologies, like large-scale AI and even crypto, has an inherent cost in the increased CO2 that comes with deployment at scale.
His presentation started by creating the appropriate framing: why he chose to give this talk. Growing up and now living in Silicon Valley allowed him to see the impact (positive and negative) that transformative technologies have on communities.
You cannot name a technology that was transformative without being damaging
Making reference to articles in the media, he stated that generative AI has a dirty secret because it requires more energy than other cloud services. Even if big technology companies like Google, Meta or Microsoft are making efforts to ensure that most of their new data centres are as green as possible, the amount of energy consumed is too high. To emphasize the impact the thirst for energy of new technological trends has, he underlines that coal power plants that didn’t operate in years are now in operation again.
To ensure that the impact is understood correctly, he made the connection between the growing CO2 emissions and global warming and the extreme weather conditions that are recorded all over the globe. He stated that the recent high amount of rainfall and snow was only once matched in the recorded history of California.
One of the fascinating things is human beings have this great ability to solve their problems with more complexity. We know it will emit more CO2, but we will make it more efficient.
He underlined that until humanity finds a solution, people will continue to be affected. Due to flooding, for instance, many communities were affected.
How do you fix somebody who doesn’t have a home anymore?
Next, he brought up the problem in the engineering space stating that generative AI will need a different infrastructure. According to him, we need to rethink how we build infrastructure that will support the new ChatGTP era. To think of a new data centre design that allows us to benefit from machine learning (ML) in a manner that doesn’t impact the environment. HyperScale Data Centers might be a solution as they:
- Move Data Faster
- Own Energy Sources
- Are Eco Friendlier
He compared the building of the interstate highway network in the US and its impact with the building of new data centres for generative AI. The technology will have multiple benefits, but the impact on the local communities should not be ignored. He references the work of Dr Timnit Gebru from Distributed AI Research institute and that of Dr Joy Buolamwini from Algorithm Justice League regarding AI bias and how to ensure its fairness.
We know that AI is biased. The data we feed it with is biased. What do we say? We’ll fix it later!
He continuously encouraged action now, especially as we can make decisions that would help everybody "Not because it is expedient, but because it is the right thing to do". Similar calls to action could be heard in other formal presentations or informal open conversations on security and gender equality. Rebeca Parsons, ThoughtWorks CTO, used the following quote from Weapons of Math Destruction:
Cathy O’Neil: We can make our technology more responsible if we believe we can and we insist that we do
The last part of the keynote focused on mitigation strategies. Using smaller models with enhanced societal context might provide a better output than big, resource-consuming models. Knowing Your Data (KYD) and Knowing Your Data Centre (KYDC) will allow you to take better decisions. All big cloud vendors provide dashboards for measuring CO2 footprint.
His closing statement echoed around the room:
When ChatGPT occurred I knew it is something crazy big. Something seminal like the advent of the World Wide Web. Technology is meeting people where they were at. We have an obligation to meet it with compassion, and humility, and try to understand the social and cultural impact before we do it. We have the only chance for it. Otherwise, the world will look different than what we want it to be.