According to the Pew Research Center, 68% of technology innovators, developers and business leaders expect that ethical principles focused on the public good will continue to be overlooked in most artificial intelligence systems through 2030.
As AI works to match human capabilities, a primary concern is that it could potentially outpace our ability to control it within an ethical framework. As a result, there’s a growing movement to create ethical guidelines for AI systems. But to enforce AI ethics, the industry must first define those ethics.
Different individuals and organizations have attempted to create ethical AI codes throughout the years. For example, in 2016 the EU passed GDPR, which laid the groundwork around a model for how to enforce ethics related to intangible tools that impact human behavior. This has required businesses to consider the ethics of using and storing personal information, a crucial first step when dealing with AI.
Still, today there is no broadly accepted AI ethics framework, or means to enforce it. Clearly, ethical AI is a broad topic, so in this article, I’d like to narrow it down and look at it through the lens of network monitoring technologies.
Also see: Top AI Software
AI and Network Monitoring
AI has many potential benefits when applied to network monitoring and performance. While many staffers worry about being replaced by AI, in the networking space the growth of AI actually signals improvement, not displacement.
In fact, AI in IT monitoring environments can streamline complex networks, automate specific tasks, and help increase efficiency around threat detection and remediation – to name just a few areas. It can also simplify IT’s role in oversight and help get to the root cause of issues faster.
Let’s look at some specific examples of AI in network monitoring, so we can later better understand the key ethical issues.
- Anomaly detection uses AI/ML to understand normal versus anomalous behaviors (to establish baselines) on a network. It’s used to build models of what typical traffic looks like adapted to specific locations, users, and time aspects. These models can be very detailed, down to the specific application. They allow organizations to understand patterns by extracting features of the application from a network perspective.
- Predictive analytics leverages data with AI/ML to predict potential issues that could happen in the future across a network. Much like anomaly detection, it also uses data analytics to learn about historical patterns and events, and looks for and learns about patterns that may cause issues.
- Automation also uses AI/ML to determine what a root cause of a networking problem might be and remediate it automatically. ML techniques such as decision trees or more sophisticated techniques can create learned processes to diagnose issues rather than making manual rule-based systems that can be error prone and difficult to maintain.
Also see: Data Analytics Trends
Networking and AI Ethical Issues
While AI can deliver a new level of visibility and problem solving when applied to network monitoring, there are also ethical considerations or questions that the industry should be looking at or asking. There’s a lot of debate around ethical AI, yet most agree that AI ethics is a system of moral principles and techniques intended to inform the development and responsible use of AI technologies.
But what does that mean in the network monitoring space? I don’t pretend to have all the answers, but I do have some key questions we all should be asking and working together to address.
- Is the data being used following privacy and protection regulations that’s applicable – whether it’s GDPR in the EU, or other regulations? Network data can have personal, behavioral, and trend information. Making sure that it follows regulations is important, especially as AI/ML systems more heavily ingest data.
- Does the data have any potential for bias as features are extracted and used to train models? As models are developed, humans are biasing detections based on patterns that may correlate to gender, race, ethnicity, etc. This is more pronounced with social data, but the users generating network traffic may have patterns specific to a cohort group. Although this might not create social bias, it could create models that may not work universally as expected.
- Are the actions recommended or performed based on the analysis and the potential implications? As observed with self-driving cars, there are always “corner” cases or unseen scenarios that AI systems may not have been trained on. Exploring every possible outcome, even if not supported by data, should be considered and accounted for.
It’s important to note that the industry is not starting completely at square one, but it is early days for AI standards. Today, there are initiatives in IT that are designed to help create and shape ethical AI. These include at a broad level GDPR, which doesn’t address AI ethics directly, but it does address data protection and privacy, which has implications on the usage of such data for AI.
There is also a proposed EU AI Act that will address rules specifically around development and the use of AI-driven products. But mostly AI ethics are left to technology developers at this point – something that needs to change in the future.
As AI innovation continues, setting guardrails and standards will be key. Unchecked AI is universally considered a recipe for disaster.
But AI produced and implemented with ethical guidelines has the amazing potential across the network monitoring space to save NetOps teams significant time and resources when it comes to collecting, analyzing, designing and securing networks.
Also see: What is Data Visualization
About the Author:
John Smith, CTO and Co-Founder at LiveAction