AI is not conscious nor a panacea, but it does represent a seismic shift in how we tackle the complexity of our natural and modern world. AI is often used as a catch-all term that encompasses multiple entwined elements, or as a synonym for ML. Its application is mostly successful in specialized domains which result in the phrase ‘narrow-AI’ being used as opposed to AGI (artificial general intelligence) which is still deemed to be decades away (if at all). We’re somewhat confined to figuring out how best to apply AI in well-understood problem spaces, all the while attempting to eliminate bias, but continue to explore the novel use of AI in less well-understood domains.
While the bar has been lowered for access to certain consumable AI services, the real challenge still remains of building end-to-end systems that deliver insights, and more importantly, actions. Expectations may have been raised by the hype surrounding AI but to actually deliver on it still requires domain experts, great system design, the right AI Data Science toolbox and excellent user experience. Great AI begins with great data, but then presents the intermediate challenges of scaling, automated ingestion, data augmentation, data correlation, feature engineering, model training and system resiliency before a useful and highly autonomous platform can take shape.
Useful AI starts with solving real-world problems, but building and productizing AI is extremely difficult as it’s an iterative and evolving process where one size does not fit all. Data is constantly being generated, and AI serves us best when it continually learns from and is trained on large contextual datasets and human feedback. This means that an effective AI-powered platform for use by end-users must be continually learning and localizing its models when necessary to ensure correctness and maximal value.
AI-driven platforms and services can also face some non-technical problems. In the previous blog, human trust was highlighted as a major consideration for the adoption and use of AI, but other challenges exist, especially in relation to AI for IT. These potential challenges include:
- Demystifying the Technology
Depending upon the use case and end-user, demystifying the technology may be moot, but for decision-makers and domain experts in IT, there is a default skepticism for “black box” technologies. Some keys to unlock this box include education, openness and transparency. Skeptics can satisfy their doubt when they can introspect and unwrap layers in a user interface or via an API (Application Programming Interface). Additionally, using well known and open-source frameworks helps to proxy trust and accelerate understanding.
- Understanding AI’s Limitations
As mentioned, AI is not a panacea; it has limitations and is only as good as the data it’s fed. Datasets don’t always represent the whole picture and can contain biases that must be eliminated or controlled for. During the training of AI models, there are challenges with normalizing and “overfitting” or “underfitting” that providers should be able to explain. This is another instance where transparency drives trust.
- Operationalization and AIOps
For AI to be operationalized successfully, forming part of a new or pre-existing workflow is key. New technologies often disrupt, but the most successful ones spread with minimal friction. Planning to integrate AI into standard operating procedures or playbooks is useful and may reduce the number of steps taken in a process. Whether the AI platform is used to accelerate decision making or condense workflows, it results in freeing up human time especially in areas like IT operations. AIOps creates an opportunity to move from primarily reactive activities and low-value repetitive tasks to more strategic, valuable and proactive pursuits.
- Fear of Human Obsolescence
While the fear of being replaced is inside us all, AI can only replicate a minute slice of human capabilities. Organizations and leaders that use AI to eliminate toil, augment human activities and free up human potential to target more creative challenges will reap the benefits. As it becomes tougher to navigate our vast, complex and interconnected digital world, new tools and technologies have always helped us to evolve, scale and move forward faster.
As AI quickly moves into the networking industry it has profound implications that until today we’ve experience only in spoonfuls. In our video series, Get Smart: Unpacking AI, industry and Juniper experts demystify AI for you, exploring its meaning and application in the networking industry.
Additional blogs in this series
Decision-making with Real AI Beats a 6th Sense Every Time
Assistants, Chatbots and AI-Driven Frameworks
Anomaly Detection in Networks and a Few Tools in the Box
