When it comes to data science and analytics, artificial intelligence (AI) and machine learning (ML) are very common buzzwords. Often, the terms are used interchangeably. This can be because people are referring to AI as the broader topic and ML as the methodology — so they’re really talking about both. Some don’t think the distinction is important, while others will argue passionately for it. In my latest insights article, I will attempt to clarify each term and explain why they are so often confused.
Quick definition: the overall field which aims to create human-like intelligence in computers.
The term was first coined in 1956 by AI researchers. It’s an all-encompassing term that refers to the field of study aiming to create computers that mimic human-like intelligence. What is defined as AI changes as the field develops. Back before people could easily carry around a chess-game in their back pocket, a chess-playing computer program was considered a cutting-edge form of AI, since game theory and strategy were considered uniquely human skills.
We’re long past those days, and AI is used for everything from google mapsto mobile banking. Every time a machine completes a task for you, based on a set of stipulated problem-solving rules (algorithms), that’s AI.
Quick definition: an AI-based methodology where computers can learn from new data without the explicit programming to do so.
The majority of AI industry experts, such as those from Googleand Intel, define machine learning as a subset of AI. Machine Learning is mostly used to analyse large-scale datasets. It’s a computer algorithm that learns by observing the initial data, then it finds patterns and improves future decisions with those insights.
For an everyday example, spam filters use machine learning. Given that scammers regularly update their methods, simple rule-based AI (i.e. filter out emails containing the words ‘lottery winner’ from unknown addresses) would be ineffective. Machine learning continually monitors the words and phrases used, compares this to previously identified spam emails and takes account of your personal preferences (one person’s spam is another’s gain!). Google’s machine learning model can now filter out spam and phishing emails with 99.9% accuracy.
You might have also heard of deep learning and neural networks. While beyond the scope of this article, deep learning is best understood as a more complex subset of machine learning. It operates via neural networks, which are algorithms inspired by the human brain’s processing (hence the name).
In short, machine learning, deep learning and neural networks are all forms of AI. But whereas all machine learning is AI, not all AI is machine learning.
Why the confusion?
When AI first gained traction, big promises were made. When human-level AI wasn’t quickly produced, the funding disappeared and caused what’s referred to as the AI Winter. So, organisations then tried to distance themselves from the term, marketing their technology as big data analytics, machine learning, deep learning and neural networks.
AI seems to have made a comeback. However, these patterns can be cyclical, and many are expressing concerns that AI is being over-hyped again.
Does the distinction matter?
Recently, a report highlighted the prevalence of companies misusing the term artificial intelligence. 40% of European companies that claimed to be using AI, actually don’t use the technology.
It’s not clear how much of this confusion is deliberate. Some may be due to third-parties miscategorising companies. However, there are clear incentives for companies to use AI terminology — start-ups that claim to work in AI attract 15-50% more funding than other companies. So, it’s not a big leap to suggest that some of these companies might be reluctant to correct terminology mistakes.
However, this is misleading for consumers and contributes to the over-hyping of the technology. Both recipes for another AI Winter.