© 2019 SourceMedia. All rights reserved.

6 trends that will drive artificial intelligence deployments in 2019

The shockingly fast sell-out of the 2018 Neural Information Processing Systems (NIPS) conference put industry on notice: artificial intelligence is poised to break out of the hype cycle and become an important element in strategic business planning.

A lot has been written about AI over the last year, but as the many papers and presentations from NIPS showed, much of what has been offered for public consumption has been superficial and, in some cases, misleading.

AI is neither a simple panacea poised to solve all our business problems, nor is it a looming evil ready to usher in a dystopian future. It is a powerful tool that can aid decision making in a fast-moving, digital business climate--and there’s ample evidence that it’s already happening.

Among the lucky few who were able to secure credentials to the conference, and as machine learning practitioners who have spent the better part of their careers involved with AI and machine learning, we came away with a look into the future of AI. Those insights and our own experiences have informed a brief forecast for where and how we’re likely to see this technology make meaningful inroads in the year ahead.

artificial intelligence breakout year.jpg

Before going further, however, it’s worth noting that the term artificial intelligence is often used interchangeably with machine learning and deep learning, both of which fall under the overall definition of AI. That can be confusing if you are not aware of the distinctions, so here’s a quick summary:

Machine Learning is the study of how to build mathematical models from data to help inform decision making. For example, since the early 2000s face detection in compact digital cameras has used machine learning algorithms to automatically focus, adjust lighting and perform other functions to improve picture quality.

Deep Learning is a family of methods within machine learning that studies how to use available data to learn a hierarchy of representations useful for certain tasks. Deep learning algorithms can build more effective models from large amounts of data compared to traditional machine learning algorithms. Some services that have shifted from traditional machine learning to deep learning include face detection, speech recognition and language translation.

With that out of the way, on to our top forecasts for AI in 2019:

5G Meets AI to Accelerate Innovation

A major enabling moment for commercial artificial intelligence in 2019 will be the rollout of public 5G networks. In fact, 5G may not live up to its full potential without the aid of AI automation. Network operators need AI to optimize network traffic flows, manage network resources like radio access network (RAN) slicing, and improve user experience through more efficient Wifi offload.

Traditionally the link between centralized servers and associated clients has been a major bottleneck for data driven solutions. Moving core AI functionality closer to the data source and intelligence consuming devices and applications will be critical in providing mission critical services across early-adopter industries like healthcare, financial services, and manufacturing.

The speed, performance, and reliability of 5G means that data accessed and collected by billions of edge devices, with different form factors, can be aggregated and transferred for faster and more accurate analysis, unlocking more (and more precise) insights. The huge volume of data collected from a diverse set of devices will significantly improve AI-based solutions, and deep learning models in particular.

This means that enterprises will also begin to adopt AI to impart security at the network edge where an enormous number of nodes, with diverse resources and security vulnerabilities, are expected to connect. The current state of (poor) edge security was made evident in 2016 when the Mirai botnet attack halted internet traffic on much of the U.S. East Coast. Since then, attacks on IoT networks have increased by 600 percent.

The combination of 5G and AI will lead to many interesting applications for both users and network operators. IoT devices (as many as 50 billion by 2020) will become a rich source of the data used to inform AI. In turn, 5G will enable IoT devices to act intelligently based on information gathered and shared throughout the network.

Early examples include medical devices that collect and analyze data to help with diagnostics, event prediction and drug administration; devices that improve industrial production efficiency; sensors that effect “smart” agriculture for water conservation and optimized fertilization; vehicle components and telematics that enable fleet and logistics management; and smart meters for power and utility grid management.

Deep Learning Expands into Solutions, Including Cybersecurity

AI has been used for years as a foundational technology in assisted services like online intelligent marketing chatbots and voice recognition-based customer service tools. Recent advances in AI will lead to more sophisticated applications in industries like healthcare, financial services, and distance learning. Another industry that will benefit from deep learning innovation is cybersecurity.

Older machine learning models have been used in cybersecurity for years to comb through big data to identify and flag anomalies based on perceived norms in network traffic and performance. This has overburdened security analyst teams with an overabundance of false positive signal detection. Using the vast amounts of threat data and processing power now available, deep learning can be applied to train and validate new threat models to more effectively address the acute malware problem facing today’s enterprises.

AI-as-a-Service Hits its Stride

We expect to see the advent of AI-as-a-Service, where startups, entrepreneurs, and cloud services organizations create applications and toolkits that put AI within reach of more organizations without requiring employees with AI skills on the payroll. Think of it as a democratization of AI, similar to what drove the cloud computing movement and the consumerization of business applications.

When this happens, it will not only serve to streamline many common business processes, but it will also be used by entrepreneurs to start and run lean organizations, just as mobility brought about the idea of the “virtual organization.” Complex tasks like logistics management that once required specialized teams might soon be handled by an individual—one who wears many hats, assisted by specialized AI applications.

Federated Machine Learning Gains Traction

Most current machine learning services rely on a centralized mechanism to perform model training and deliver updates. This requires data from various sources to be gathered to a single point so that new models can be trained. While this method is fine for building general services like object detection in images, many applications--industrial and consumer--do not allow for sensitive data to be moved. Federated machine learning will enable collective learning to produce highly accurate models without data sharing.

Transfer Learning Enables Faster Development Process for New AI Services

As companies try to apply AI to new domains, many will hit a brick wall when they realize that it is too difficult and expensive to label the mountains of data required to train machine learning models for new applications. Transfer learning will allow for "knowledge" to be transferred from models trained on different-but-related tasks so that highly accurate models can be built for new applications without the need for large amounts of data.

Reinforcement Learning Breaks Through in Commercial Applications

For applications where the task is to build an agent that can react to the environment by performing actions, reinforcement learning has long been touted as the solution. To date, however, reinforcement learning has not been able to deliver on this promise aside from some niche use cases such as AlphaGo and DeepMind's use of reinforcement learning for data center cooling. As deep learning gains more traction in the reinforcement learning domain (often called deep reinforcement learning), it will result in new breakthrough applications.

Looking back, and despite the tremendous strides made in recent years, we have yet to scratch the surface of what AI is capable of. Blue Hexagon is excited about what’s to come--and we’re not alone in that thinking.

Management consulting firm McKinsey & Company sees the various AI disciplines driving as much as $5.9 trillion in annual economic value across 19 industries it examined. That’s a lot of innovation and opportunity ahead. As machine learning practitioners, we look forward to the steps we’ll all take together to explore and deliver those possibilities in the next year and beyond.

For reprint and licensing requests for this article, click here.