Machine learning: The next transformational technology
Machine learning has joined artificial intelligence (AI) as the hottest technology topics of 2018. We asked our expert influencers to share their thoughts on the state of the industry: where it's going, and how and why companies should be adopting machine learning and AI.
Joining us are Ronald van Loon, director of Adversitement; Dr. Manjeet Rege, associate professor of graduate programs in software at the University of St. Thomas; Jennifer Shin, founder, @8PathSolutions; Colin Sumter, Analyst, CrowdMole; Daniel Yarmoluk, Director of IIoT & Data Science, ATEK Access Technologies; and Dr. Craig Brown, CEO, STEM Resource Partners.
What is driving the momentum and urgency for companies to adopt machine learning and AI?
Dr. Manjeet Rege: It is no more a question of whether you want to adopt an AI strategy, it's a matter of how soon. The momentum and urgency you see today with companies wanting to adopt machine learning and AI can mainly be attributed to three factors.
First, with the cost of data storage having grown to negligible levels compared to what it was 15 years ago, enterprises find themselves archiving large amounts of data. In addition, the rate at which data is generated has gone up exponentially with mobile computing and IoT (Internet of Things). Second, earlier enterprises had to heavily invest in hardware to process and analyze large amounts of data. Now, one can easily deploy your machine learning pipeline in the cloud without having to worry about the expenditure upfront. Third, we have seen major advancements in the field of machine learning. Previously, there were some practical challenges in building and training large neural networks. Now, with deep learning, it is now possible to build, train and deploy those and evaluate the effectiveness quickly.
Jennifer Shin: The momentum for machine learning is both the business and consumer sectors, which is not a common occurrence. With the availability of open source software, the general public has access to code that might have only been of interest to researchers in the past. The interest from the public has pushed the urgency of adopting machine learning and AI in order to stay more competitive as well as to improve their brand by incorporating cutting edge technology.
Daniel Yarmoluk: We are living in the age of disruption and companies must disrupt themselves or be disrupted. Larger enterprise clients are often conflicted between quarterly results and "minding the store" and time-consuming internal meetings rather than true innovation. The convergence of technologies is like nothing we've seen before in humanity, and things change very quickly.
What are some examples of how machine learning and AI can provide competitive advantage to companies?
Dr. Rege: One of the biggest successes that come to my mind is personalization using a recommender system. Whenever you apply ML in a way that helps consumer interaction it is a winner. Browsing through similar relevant products is way easier than endlessly searching through all the products available. Whenever you browse on an ecommerce site like Amazon today you have personalized list of products being recommended to you. Similarly with streaming services such as Hulu or Netflix — when you make a profile initially they ask you to make a selections to determine your preferences. From that point onwards you are being served personalized recommendations - some of which you may like or dislike. However, more you browse and click, better the recommendation system gets because there is a Machine Learning algorithm in the background learning from it.
Shin: Machine learning and AI allows the automation of processes that often takes companies weeks or months to complete in a fraction of that time. The benefits of this is endless, but using it as a competitive advantage takes skill and experience. We can always use technology to automate a process, but picking the right ones to automate and the order in which we automate these processes is what makes the difference between using technology and competing with technology.
What’s a compelling example of a new business model that was made possible by machine learning?
Dr. Rege: Online search and advertising is one of the biggest examples of a business model made possible by machine learning. Companies like Google and Facebook have been primarily built on revenues from online advertising, which is based on how people search. Every time you do a search online, a number of search results are shown to you, along with advertisements for you to click on that generates clickstream revenue. Based on the results that were clicked on or not, the search engine machine learning algorithm learns and refines the results presented over time. Correspondingly, the advertisements presented to you are meant to be relevant based on your search history.
What are areas of business that can benefit the most from machine learning and AI?
Ronald van Loon: Machine learning and AI are having major impacts on different businesses applications, including customer service, healthcare, and industrial IoT.
Those businesses who provide customer services deal with large amounts of data, and automated, intelligent technologies enable them to advise customers towards their desired outcomes with limited human involvement, by using chatbots, for example. Healthcare sectors can reduce flaws in their processes with cognitive capabilities that augment doctors’ judgements about patient conditions. In industrial IoT, cognitive anomaly detection is predicting equipment failures before problems arise and enhancing asset management.
Colin Sumter: Building business logic is a huge area that may be potentially addressed with machine learning and AI. One in 10 spreadsheets are accurate. And there are many business using linear regression based on these types of spreadsheets. Machine learning that scales offers a different class of business rules and logic that includes supervised classification of business goals.
Yarmoluk: I feel machine learning and AI can help with businesses that have to interpret a huge set of data. The human brain has limitations on what we can digest over a period of time, which we can now achieve, in theory, in seconds with ML and AI. You can think of radiologists, their mental data set of hundreds or thousands tumors versus a machine digesting millions of images to pinpoint specific diagnoses and treatment. In industry, ML can help uncover new anomalies or wrinkles in the data instead of generalized buckets to fine tune productivity through a universe of sensor data. Retail is already illustrating personalization with product offering. While many think of automation and cost reduction, I feel these granular trends now visible will be innovative revenue generators.
Why is this the year to implement machine learning, to prepare for enterprise AI?
van Loon: There’s increasing emphasis for organizations to integrate end-to-end data management platforms in order to manage all current and additional data streams.
With the continued maturity of machine learning applications this year, and different vendors featuring effective, unique machine learning solutions within their own domains, organizations can have the capabilities to transform their data into actionable insights. This also gives enterprises the ability to successfully handle the significant escalation in different data streams, including deep learning applications.
Shin: Over the past decade, companies have started to recognize the value of data and companies have met the demand for data by building better technology to enable us to process data. We had only a handful of tools to work with data back then and now there are more data tools than ever before. With the tools and materials at our disposal, companies now have an opportunity to build AI to enhance their capabilities and gain an edge over their competition.
Sumter: This is the year to implement machine learning because it's affordable, your competition is going to use it, understand it before you and possibly take advantage of a competitive mismatch because their conditional feeders are far more accurate than anyone using a spreadsheet.
What are the key building blocks for machine learning?
van Loon: Laying a foundation for machine learning starts with effectively deploying the right data and analytics layers to build steps towards AI. Creating a data-driven infrastructure through governance, interdisciplinary teams and managing the end-to-end data lifecycle is key to implementing a comprehensive data management system. A cloud infrastructure, hybrid data management, unified governance and integration, data analytics and visualization, and data science are the core building blocks necessary for machine learning to grow to AI.
Shin: Machine learning requires a significant amount of compute. A business that doesn't have the right infrastructure in place will see limited benefits in their machine learning deployments. Without enough power to run the algorithms, the system will lag; from an end user's standpoint, this can negatively impact the customers experiences and from a business standpoint, the investment in machine learning could result in lower productivity.
From an algorithmic standpoint, machine learning requires efficiency, accuracy and consistency. The best sources of information needs to be processed as fast as possible, ideally with minimal failure points.
Yarmoluk: I firmly believe that the key building blocks are a three-headed animal of product development/go-to-market/business model person, data scientist and domain expert. The domain expert has to understand the customer issues, UX, pain and what problem needs to be solved. The data scientist has to leverage that domain in order weigh variables and understand what they want to answer or predict. Lastly, the business side has to provide an answer to a problem or pain, it must have utility and delivered in a way they customer can consume the information, data or prediction. We have to strive to leverage data science for all without being a data scientist, and also that it is ubiquitous in data-driven decision making.
What are the crucial capabilities for a data science platform?
Dr. Craig Brown: Natural Language Processing, Ontology Management, Time Series Analysis, Event Analysis, Profile and Trend Analysis, Classification Analysis
These crucial capabilities provide a multitude of solutions that can lead to critical components needed for insight driven and predicative analytics. The capabilities open the door to more capabilities that can help with data refinement, data analytics, data pipelining, etc..
This list of capabilities is instrumental into finding the details in the data that can support organizational transformations that can lead to insights that can help business:
- Acquire new customers?
- Retain customers?
- Increase customer satisfaction?
- Mitigate and manage risk?
- Increase operational efficiency?
Sumter: Crucial capabilities for a data science platform are security with cloud object storage, in-memory analytics, and collaboration. Cloud object storage is the next generation of resiliency and security for your big data analytics on Watson IoT that scales. In-memory analytics makes your data readily accessible as the users of your data platform scale and is very useful for simultaneous data visualization(s). And collaboration tools ensure that your team is minimizing the "curse of knowledge" with a proprietary glossary that can keep up with real-time data processing from your IoT logic and services.
What’s your perspective on the untapped potential of data that companies have stored behind their firewall?
Dr. Brown: Today organizations have stored years of data for no reason other than insurance that if the data is needed it would be available. I often refer to these data reserves as the “MODERN DAY GOLD RUSH”. Transaction data, customer data, sales data, taxation data, orders data, etc., just sitting there in the database as a part of hundreds of millions of database records. The potential use of this data is vast and what can be learned from this data goes even farther. With proper data science this data can be transformed into projections, predictive analysis, insight, operational analysis, decision support, key metrics, etc.. Untapped potential is an understatement.
Interested in learning more about how you can apply machine learning within your organization?
Join us on 27 February at 1 PM ET for “Machine Learning Everywhere: Build Your Ladder to AI.” Visit the event landing page to learn more about the event and register for a calendar reminder.