Interset and Responsible AI - Part 4: Sustainable Development and Final Thoughts

by in Security

Read the previous chapter in this series: Interset & Responsible AI – Part 3: Diversity Inclusion, Prudence, and Responsibility. 

This series on the Montréal Declaration for Responsible AI and Interset has taken us on a journey through nine of the fundamental values and rights of individuals and groups to which AI innovation should adhere. Today, we’ll look at the 10th and final principle of the Declaration.

Principle 10: Sustainable Development

“The development and use of AIS must be carried out so as to ensure a strong environmental sustainability of the planet.”

There is a lot of conversation about environmental sustainability in many industries. In AI, the topic is heavily emphasized when it comes to AIS hardware. For example, the first sub-principle of sustainable development states:

“AIS hardware, its digital infrastructure and the relevant objects on which it relies such as data centres, must aim for the greatest energy efficiency and to mitigate greenhouse gas emissions over its entire life cycle.”

A challenge with security analytics is that, if you’re not careful, it can eat up substantial compute cost and require resource-heavy servers. In an effort to reduce cost for our customers, allow an ultimately more sustainable operation, and reduce our analytics’ footprint in the grand scheme of energy efficiency, we select algorithms that factor in cost-effectiveness. For example, we rely less on deep learning models because the compute cost of these models is enormous. Instead, we rely more on straightforward univariate and bivariate models or likelihood estimation approaches that translate into orders of magnitude lower hardware, power and cooling requirements, which in turn translates into a lower carbon footprint. Importantly, however, these approaches are not just more cost-effective; it turns out that they are just as effective at threat detection, and in many cases more effective than more complex approaches: simpler approaches generalize better, converge more quickly, and are more resilient to bias and noise.

If you’ve been following our product development over the last year, you’ll also know that we are continuing to move towards cloud-optimized analytics. By designing analytics specifically for use on top of cloud platforms, and leveraging serverless technologies and other efficient ways of distributing computation across a large, shared cluster, we can lower cost and compute requirements. So far, compared to traditional on-premise approaches, our cloud deployments have reduced cost by 80%! 

A few closing thoughts

It should be obvious by now that we’re big fans of analytics here at Interset and, naturally, it’s easy for us to get excited about the technical aspects of AI. There have been and continue to be so many advances in this realm—deep learning, cloud computing, and computational power, statistical learning approaches to real-world data—that recent years have experienced a renaissance for AI innovation.

This climate might tempt AI initiative newcomers to jump right into the technical bits, but it’s important to think beyond the technical approaches to evaluate the right algorithm to choose, the best dataset for the problem, the best way to access or ingest data, etc. The Montreal Declaration provides an incredibly important and timely framework in which to architect those technical bits while being cognizant of the larger societal impact. Building a responsible AI system requires answering important questions beyond the math itself. Which algorithms will result in transparency and trust? Can I use this data without violating individual rights? Can I access and ingest this data in a way that respects privacy? How do I ensure my AI does not disseminate untrustworthy information? How can humans dialogue with my AI to support collaboration and solidarity?

Ultimately, every technical decision has to be made in the greater context of responsible. That means that designing a responsible AI system starts at the very beginning. And that’s exactly what the Montreal Declaration facilitates. It’s one of the reasons why I (and Interset) is such a fan. It is an understandable, actionable guide for applying structure to an inherently fuzzy and confusing area: that of the ethical, sustainable, and respectful development of AI.

AI is fantastic and incredibly powerful but, as Spiderman says, “With great power comes great responsibility.” As is the case with any new innovative or disruptive technology, it’s critical to consider societal principles during development—from start to finish.