In this special guest feature, Navin Ganeshan, Chief Product Officer at Gemini Data, discusses the often misunderstand and important distinction between machine learning and machine reasoning — which is finding patterns versus understanding relationships. Each play a particular role in the analysis process and while different, are equally as important to deriving the most value out of the other. Navin is a product executive with a two-decade career in bringing innovative and award-winning technology products to market. Prior to Gemini, as CPO at Zubie, he led the company’s connected-car and developed its pioneering “internet of cars” data platform. As CPO of Centrifuge Systems, he led the company’s analytics and visualization product line. At Network Solutions he held roles including Chief of Strategy, Products GM and head of Enterprise Data Services and BI.
Artificial intelligence has changed the way companies leverage data. The trouble is, many still don’t understand the nuances between AI technology variants and the unique benefits each provides. Machine learning, machine reasoning, AI – all terms used extensively and often synonymously, despite their differences and specific use cases.
Most notably, people often misunderstand the important distinction between machine learning and machine reasoning — which is finding patterns versus understanding relationships. Each play a particular role in the analysis process and while different, are equally as important to deriving the most value out of the other.
There’s a simple analogy to help distinguish the difference between machine learning and machine reasoning – and how together, they make the most cohesive AI solution.
What is machine learning and where is it best applied?
An easy way of explaining the value of machine learning is to imagine a toddler is pushing a glass over the edge of a table. Without having encountered this situation before, there’s no way for the toddler to predict the outcome. But with growth and learning, he understands what happens, even if he doesn’t completely understand why. That’s machine learning at work.
Machine learning is a widely used form of AI that relies on using collected datasets that can be analyzed for patterns. While machine learning has unparalleled success in many areas involving big data and patterns, its impact on cybersecurity is where it’s most measured — and has faced the most backlash. There are three reasons this might be the case.
First, machine learning in security gets a bad reputation because accessing existing data for analysis, can be difficult – due to enterprise security policies that constrain what is shared, enterprises become isolated and unable to learn from the broader community.
Secondly, security as a practice is also considered a cat-and-mouse affair with threat vectors constantly evolving and becoming more complex. This restricts the value of prior datasets to be used for predictive value, because “fighting the last battle” risks missing new patterns in the data.
Finally, machine learning faces the obstacle of having to overcome the reliance on tribal knowledge. Somewhat counterintuitively, IT and security practices tend to put a great deal of emphasis on innate knowledge possessed by the individual while also relying extensively on data-driven analysis. It’s well understood that AI-driven techniques may uncover a large number of potential incidents, but it takes a combination of data, domain knowledge and educated instincts to perform deeper investigations. Tribal knowledge is valuable, but it’s simply a piece of the greater puzzle.
Ultimately, machine learning is best applied in scenarios where the outcome is probabilistic — like determining a risk level. Machine reasoning, on the other hand, can complement that knowledge by adding a human element.
What is machine reasoning and where is it best applied?
Now imagine that the toddler who was once pushing the glass off the table now understands the physics of movement and gravity. Even without having encountered this situation before, the toddler can surmise what will inevitably happen. The toddler can apply the same logic to another object on the table — adapting that knowledge and applying it to a TV remote on the same table — because he knows why it happens. That’s machine reasoning.
Machine reasoning is a more human-like approach within the AI spectrum that’s highly relevant to big data investigations, therefore it allows for more flexible adaptation than machine learning.
However, machine reasoning requires heuristics and curation, which is usually done by knowledgeable domain experts. This process is where machine reasoning may be difficult for companies to scale — it requires a great deal of expert human effort for this curation to take place.
Machine reasoning is best applied in deterministic scenarios – that is, determining whether something is true or not, or whether something will happen or not. Knowing this, it’s clear why machine learning and machine reasoning work well together.
Uniting machine learning and reasoning: what companies need to know for best results
Machine learning and machine reasoning shouldn’t be seen as competing approaches to understanding data, but complementary ones. It ultimately comes down to understanding the specific use cases and how your company can stand to benefit from each.
Sign up for the free insideBIGDATA newsletter.
Thank you Navin for making the difference between Machine Learning and Machine Reasoning so abundantly clear. The example of the toddler just acting out in machine learning mode and then reasoning in machine reasoning mode are especially vivid. No more doubts now. Thanks!