What are the main challenges and risks of network automation AI projects?

Complexity of Network Environments: Challenge: Network environments are often complex and diverse, with a variety of devices, protocols, and configurations. Risk: Implementing automation in such diverse environments can be challenging and may lead to unforeseen issues, especially if the automation system is not designed to handle the intricacies of the network. Integration with Legacy Systems: Challenge: Many organizations have legacy systems that may not be inherently compatible with modern automation technologies. Risk: Integrating automation with legacy systems can be challenging and may require additional time and resources. It could also result in partial automation, limiting the overall benefits.

AI network optimization is probably a last step in a process that starts with good network observability. Even thinking about network automation without AI is not possible without good data. So in terms of stages moving from manual operation towards AI based network automation has at least 3 steps in between. The first one is a project for reliable data collection. The second would be data organization and analytics. The third would be simple automation and then integrating AI would be possible.

Data quality and availability are paramount for the success of network automation AI projects. Without accurate, complete, and relevant data, AI models and algorithms can produce unreliable outcomes, jeopardizing network performance and security. Moreover, data accessibility, timeliness, and security are crucial for ensuring that AI initiatives can effectively leverage data assets. Implementing a robust data governance framework, coupled with reliable data collection, storage, and preprocessing mechanisms, is essential to mitigate risks associated with poor data quality and availability. How do you prioritize data quality and availability in your network automation AI projects? Share your strategies below!

This understanding allows them to answer the following critical questions: How do these new technologies interact with one another? What are the correct design and implementation tradeoffs to make when implementing automation? How do I learn to effectively use these new technologies? What should my role be in an automation-driven network environment? When should one tool be favored over another?

The transition to automation is like the change to VoIP. Although the syntax of the configurations remains unchanged, almost everything else in the network configuration process is changing. The new knowledge and skills are a radical departure from the per-device configuration of routing, switching, and firewall. You may have heard that adopting automation is a journey. That’s an apt description. The culture change and learning how to apply the new tools take time. Here are the factors that you'll have to build into the journey.

Security Concerns: Challenge: Automation introduces new attack vectors and security concerns, especially when handling sensitive network configurations. Risk: Inadequate security measures could lead to unauthorized access, data breaches, or the manipulation of critical network settings, posing significant risks to the organization. Lack of Skilled Workforce: Challenge: The successful implementation of network automation AI projects requires skilled personnel with expertise in both networking and AI technologies. Risk: The shortage of skilled professionals can hinder project progress, potentially leading to misconfigurations or inefficient use of automation tools.

In my experience, expecting that network engineers will learn about AI to develop an AI based automation is not the right thing to do. It is better to bring together the people that understand the network, the people that write scripts for network automation and the people that know about AI. Each of them can contribute their skills and knowledge to the project. In terms of culture, it is important to have a clear view of the intention of network automation independently of the inclusion of AI or not. People that operate the network everyday may not feel incline to help much if they feel their job is at risk. Offering opportunities for growth during the project execution may be helpful to avoid these kinds of situations.

Indeed, bridging the skills gap and nurturing a culture conducive to network automation AI projects are crucial endeavors. These initiatives demand a blend of technical expertise, including programming and data science, alongside soft skills like communication and problem-solving. However, resistance to change and skill deficits among network professionals can impede progress. To overcome these challenges, investing in comprehensive training, education, and mentorship initiatives is essential. Furthermore, cultivating an environment that fosters learning, experimentation, and feedback promotes a culture of continuous improvement and innovation.

Resistance to Change: Challenge: Employees may resist the adoption of automation due to fear of job displacement, unfamiliarity with new technologies, or concerns about reliability. Risk: Resistance to change can result in a slower adoption process, delays in realizing benefits, or the implementation of automation in a limited capacity. Unpredictable Network Events: Challenge: Network environments often face unpredictable events, such as outages or sudden changes in traffic patterns. Risk: Automation systems may struggle to adapt to these unpredictable events, leading to disruptions or unintended consequences if not properly configured.

AI should be a way to optimize network automation, but the underlying problems of network automation should have been resolved in a previous stage.

Managing complexity and integration in network automation AI projects presents a significant hurdle. These projects entail a myriad of components spanning data sources, AI models, network devices, and applications, necessitating seamless cohesion. Integration complexities escalate further when dealing with diverse vendors, platforms, and standards. Additionally, the dynamic nature of network environments demands adaptability and scalability from automation solutions. To surmount these challenges, embracing a modular, flexible, and standardized approach to design and deployment is paramount. Leveraging tools and frameworks that enhance interoperability and compatibility can streamline integration efforts.

Lack of Standardization: Challenge: The networking industry lacks standardized approaches to automation, leading to a variety of tools, frameworks, and methodologies. Risk: This lack of standardization can result in interoperability issues, making it challenging to integrate different automation tools seamlessly. Data Quality and Integrity: Challenge: Automation relies heavily on accurate and up-to-date data. Inaccurate data can lead to misconfigurations and operational issues. Risk: Poor data quality can result in unreliable automation outcomes, affecting the overall reliability and effectiveness of the automated processes.

Este es el desafío del uso de la IA en las redes. No hay estandarización o normas que regulen su uso como derechos de autor, habeas data, creación de contenido, entre otros. Por el momento, lo que debemos hacer es en primer momento, hacer la pregunta, ?si usamos este algoritmo de IA, perjudicamos a alguien?.

Evaluation is important but that is only possible when the problem statement and goals were clearly defined from the beginning.

Overreliance on Automation: Challenge: Organizations may become overly reliant on automation without maintaining sufficient human oversight. Risk: Overreliance on automation can lead to catastrophic failures if the system encounters unexpected scenarios or if there are flaws in the automation logic. Regulatory Compliance: Challenge: Industries with strict regulatory requirements may face challenges in ensuring that automated processes comply with relevant regulations. Risk: Non-compliance can result in legal consequences, fines, or damage to the organization's reputation.

The success of any network automation projects, including efforts based on AI, relies on clearly determining the goals, the optimization priorities and success criteria. If the lowest latency and highest QoE are the goals, the AI can be trained to modify and measure the routes and termination points used until it optimizes the QoE, however this can come with a very high per unit cost. On the other hand if the goal is to optimize cost, the AI model can be trained to select the most cost-efficient configuration but it can be very detrimental to QoE. Or maybe the goal is to balance QoE and cost at the same time but clear goals have to be defined so the AI can learn to make decisions and trade offs that achieve or get as close as possible.