Machine learning (ML) and Artificial Intelligence (AI) are deeply interconnected, so deeply that often they are used interchangeably.
They are in fact two very different things, so different as knowledge and intelligence are, even though one cannot exist without the other.
To get immediately to the point, Machine Learning can be considered a subset of Artificial Intelligence (to put it in another way: all Machine Learning is Artificial Intelligence, but not all Artificial Intelligence is machine learning!) and for sure one of its most promising area of development.
Do consumers appreciate voice search? Totally yes! The smart speaker market revenue worldwide was estimated for 15.6 billion U.S. dollars, in 2020 (Statista). Projections suggest that in 2025 it could account for over 35.5 billion. According to the Narvar Consumer Report 2018, 43% of middle-age people who has a vocal assistant uses it in the purchasing process. Moreover, the Voice Search Insight Report (Google) says that 27% of global online population uses voice search on mobile.
In fact, the origins of Artificial Intelligence can be dated back to the 1950s with the seminal works of Alan Turing (who proposed one of the first way to define Artificial Intelligence with his “Turing Test”: a machine can be defined intelligent if during a conversation with a human she doesn’t realize she is speaking with a computer) but due to limitations in technology (chips greatly evolved from the original 1960 prototypes to the actual ones) and also to the intrinsic difficulty in defining and programming an “intelligent” behaviour, for a long time it remained at an early stage, confined to expert systems and checkers programs.
The true problem is that for a computer to be able to do anything it must be programmed by a human to do so, but that has strong limits, because it’s time consuming for the human and also because a computer program, albeit sofisticated, cannot forecast all the possible outcomes of a situation. This is when machine learning steps in: for a computer to be defined “intelligent” it must be capable of learning from “experience” and to modify itself without human intervention.
Machine Learning has undergone a frantic development in recent years due to two essential factors: the exponential growth of the internet (and the connected big data revolution) and the boost deriving from research in fields with high commercial value, such as speech and pattern recognition, shop assistants, “intelligent” suggestions (as you can find on Netflix, Amazon, Spotify…) based on your past choices.
We could define Machine Learning as the branch of Artificial Intelligence that tries to teach computers to learn in the same way as humans do, gathering data from experience, and using those data, in a process full of tentatives and errors, to extract patterns, schemas, classifications, generalizations. In other words machine learning is a set of algorithms based on Bayesian (probabilistic) techniques that – given a set of data – can learn, make predictions, extract hidden structures, even try optimizations.
Unlike classical computer coding (an output is generated rigidly given a certain input) machine learning algos uses data to generate their own code (a “ML model”) which will output a result based on past examples and the pattern recognition derived from that.
There exist two main types of machine learning: supervised and unsupervised. In case of supervised learning you provide to the program not only with the data but also with the data classification (labels), in other words you teach to it the data structure (for example you can teach to the computer which transactions are fraudulent, so that it can learn to recognize one when it is confronted with it). In unsupervised learning you only provide the raw (unlabeled) data and the program learns to classify it by himself (for example it can “learn” which products are frequently bought together).
In both cases, data is an essential part of the process: the more accurate, big set of data you feed to the algorithm, the most accurate will be the results you get. Nowadays, with tons of data coming not only from “classical” sources but also from the internet of things, it’s very easy to gather data sets to feed to the Machine Learning algorithms.
The applications are countless: identify which customers are more likely to buy certain items or on the contrary could generate more problems, generate personalized content suggestions or a tailor cut website experience and navigation, identify anomalies and future problems in an industrial process, filter spam and malware, detect online frauds, perform video surveillance (anomalies detection), and much more! To get a glimpse of concrete, real world applications, have a look at what you can do by browsing the services offered in this field by DataLit.AI, our AI company.
For example DataLit.AI – using AI technology and machine learning algorithms – can identify interests at a granular level related to the website or app content, automatically segment users into clusters with similar characteristics, and predict user behaviour and purchasing intent, something that can really boost any programmatic advertising campaign and can be of interest to both retailers, advertisers and publishers, solving the difficulties connected with data analysis, that without dedicated, extremely qualified resources, would be quite difficult to perform effectively.
An extremely interesting subset of machine learning is the “deep learning” technology: deep learning is based on the neural networks theory (so it’s not based on probability and statistics like the machine learning algos) and is performed with a layered system of neural networks, each one passing to the next the results of his elaborations (from wich the name “deep”). Deep learning algorithms are extremely good in recognizing relationships between elements in a data set (while machine learning is good at recognizing patterns in data). So a deep learning algorithm can identify relationships between colours, shapes, words, and so on. Based on the results of this first relationship identification (relationships so subtle that a human could not be able to identify!) the system can then make predictions or interpretations of data.
Also in this case the applications are almost limitless: to quote the main ones, speech recognition, natural language understanding, image and video classification and personalized shop recommendations are all made with deep learning algorithms.
And Artificial Intelligence? Well, Artificial Intelligence in the “broad sense” (as the Encyclopedia Britannica puts it, “the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings”) is yet to come, but nowadays applied artificial intelligence is a cutting edge field, whose applications are endless: we go from advanced text analysis (e.g. find relevant information in hundreds of thousands of documents, find minor differences in a huge quantity of documents, automatic translation and/or summarization…) to marketing and customer care (advanced chat bots, customer retargeting, automatic analysis of KPIs, shelf analysis and price optimization, social media monitoring…), from financial technology (detect the risk of insuring someone or quoting optimal prices, support trading decisions, perform sentiment analysis on markets, detect fraudulent or abnormal behaviour…) to health technology (feed the system with patients data to discover potential illnesses or the best care plan, research on new drugs, perform deep image analysis, support healthcare professionals in triage cases…); all these AI applications are now reality and many are based on machine and deep learning plus many other technologies (so we can say that Machine Learning, vs AI, has a narrower scope) to compose a complex constellation of extremely advanced solutions that can help professionals and industries in almost any one of their field of activity.
Chief Marketing Officer at Datrix group (including PaperLit). Born in 1969 in Ivrea. Worked in Milan, Turin, Bologna, Rome and London. Debut in advertising (Saatchi & Saatchi, Italia Brand Group), followed by finance at TradingLab (UniCredit Group) as Head of Marketing Communication and Customer Service, then retail banking at UniCredit Banca and Banca di Roma as Director of Marketing Communication and e-Banking Services. He returned to investment banking at ABN Amro and RBS, then in fintech at Epic SIM.
Create some amazing digital magazines and monetize your content!
Tech company which collects, analyzes and translates client/user data into insights, identifies anomalies, predictions and business opportunities which help to improve decision making, actions and operating results.
Fintech company which collects, interprets and weighs Alternative Data for monitoring, combining it with traditional data for investment purposes.
Martech company which develops search marketing, digital advertising & lead generation platforms. They also specialize in data-driven content marketing.