The term artificial intelligence (AI) was first coined in 1956 by the computer scientist, John McCarty, when he held the first academic conference on the subject at Dartmouth College. McCarty, who is widely recognized as the father of AI, defined it as “the science and engineering of making intelligent machines.”
Over time, the meaning of AI has been refined to the “simulation of human intelligence processes by machines,” or “a broad set of methods, algorithms, and technologies that make software ‘smart’ in a way that may seem human-like to an outside observer.” Since “intelligence,” “smart,” and “human-like” are nebulous terms, AI has become the umbrella term used to refer to the entire class of technologies and algorithms, whether or not they have the ability to “learn” or have any real cognitive ability.
Many AI applications, such as GPS navigation, are based on statically programmed models. Statically programmed refers to the use of hardcoded, predefined logic rules to determine outcomes (results, outputs) based on input data. These systems do not learn, and the results produced for any given set of inputs will not change unless the coded logic rules are reprogrammed.
In comparison, machine learning (ML) applications, as a subset of AI, are not programmed to perform a given task, but rather to learn to perform the task. Machine learning techniques allow the software to improve performance over time as it ingests more data – recognizing trends from the data or by identifying the inherent categories within – to make accurate predictions.
Rather than hardcoding software routines with specific instructions, ML applications are trained to recognize significant elements and characteristics in the data to adjust their internal factors using various statistical approaches. Commonly used ML techniques include algorithms such as k-nearest neighbors, support vector machines, and neural networks. Deep learning is a subcategory of ML that utilizes extensive neural networks to closely represent a human brain to emulate human thought.
Recognize the Key Characteristics of Machine Learning.
There are a few key characteristics that define ML. Machine learning applications make predictions, whereas non-ML AI applications make determinations. A navigation application can determine a route from point A to point B, but an ML-based shopping application, for example, can only predict you might buy a certain product based on your purchasing history. Within financial services, similar ML-based algorithms can be used to suggest the next-best actions for a client or identify potentially fraudulent transactions in an account.
Perhaps the most salient characteristic of ML algorithms is that they are designed to learn how to perform a function, and as such, they must be trained to do so. Although there are different approaches to training ML applications (supervised, unsupervised, reinforced), they share the common characteristic that the data used for training must be correct. Training with incorrect or incomplete data will result in the suboptimal performance of the ML algorithms. It’s like trying to learn a foreign language from a textbook that has made-up words, incorrect grammar, and missing sentences.
Machine learning is on the cusp of changing business, lives, and even society. ML programs are being used to recognize cancers undetectable to the human eye, identify potential terrorist threats in social media, and allow companies to offer a personalized customer experience that eclipses the competition. In financial services, ML is playing an ever-increasing role in fraud detection, underwriting, credit risk, portfolio management, product selection, and marketing, as well as customer service.
To learn more about the specific differences between AI and ML, dirty data, and ways to take advantage of these technologies you can click here or fill out the form below.
