Getting Started With AI and Machine Learning
Artificial Intelligence is the victim of apocalyptic fear-mongering, utopian fantasies and everything in between.
A notoriously complex and abstract concept, there are an approximated 10,000 elite few data scientists who can claim to really, truly understand Artificial Intelligence and even they will admit that there is an awful lot they don’t understand. daisee research on the state of Artificial Intelligence in Australia, has shown that the nation is behind the rest of the world when it comes to AI.
If Turing’s revolutionary “Computing Machinery and Intelligence” was the beginning of AI research back in 1950, then there exists nearly 70 years of information exploring the infinite and convoluted universe of Artificial Intelligence.
It would therefore be a near impossible feat to condense all of this knowledge into one publication, let alone one blog post even AI probably couldn’t manage it. What we can do is provide a short overview that will, at least, kick-start your understanding and bring you into the brave new world of AI.
So, what is AI ?
Artificial Intelligence (AI) is a general term that refers to hardware or software that exhibit behaviour which appears intelligent to a human. Basic AI has existed since the 1950s, via rules-based programs that displayed basic intelligence in limited situations. Artificial Intelligence is the broader concept of machines being able to carry out tasks in a way that we would consider “smart”. The smart part is because AI can identify concepts and patterns in large data sets more effectively than humans and these insights can be viewed as intelligent.
Wait… What does that make Machine Learning?
Machine learning algorithms exist all around us, with most of us not even realising it. Whether it is the humble web search, our indispensable Netflix recommendation systems, futuristic self-driving cars or real time speech recognition, machine learning is beneficial to nearly every industry.
Although AI and Machine Learning are often used interchangeably, they are not the same thing: Under the “Artificial Intelligence” definition live all algorithms and models that allow a computer to perform tasks that are normally requiring human intelligence. Machine learning is a subset of these algorithms in which these tasks are learnt automatically, without explicitly coded instructions by a programmer. In fact, even though most of the software written today can be considered “AI”, only Machine Learning enables programs to learn through direct exposure to generally a lot of data, instead of being programmed with explicitly hand-written rules. This exposure step is known as Machine Learning training. The more data the Machine Learning system is presented with, the better results it will provide on the very same training dataset. In a sense, Machine Learning algorithms, as humans, learn mostly via experience – with some caveats, e.g. overfitting.
In this sense, data quality and volume are critical with large data sets often needed to get the best results. It’s not by chance that the most successful players in the fields e.g. Google, Amazon, Facebook are also those with access to massive volumes of curated data. Machine Learning can be applied to a wide variety of prediction and optimisation challenges, from assessing the probability of a fraudulent transaction to predicting when a client will leave, from finding unexpected and exploitable relationships in financial markets to automatically detecting and reading car number plates.
Other notable commercial Machine Learning use cases include:
- Predicting how likely a customer will buy a certain product; estimating how much a customer will spend in a product category.
- Identifying customer groups – customer segmentation
- Modelling and identifying which marketing activities have the best return.
- Social media analytics.
- Understanding and optimising the supply chain of a vendor.
There are only three basic machine learning classes of models, which depend on the kind of problem that needs solving:
Are used when the variable that needs to be assessed or predicted is known. The models learn an often very complex relation between the input features and said variable. The learnt relation can then be applied to cases where the variable is not known in advance. Regression and discriminant analysis are supervised methods. In classification problems, the dependent variable is often called “label” or “class” by data scientists.
are used when there is no defined variable to predict. These methods leverage large volumes of data to find similarities and structure, that can be eventually used for further analysis. Examples of these are clustering and factor analysis.
are used to learn a best strategy when interacting with a feedback-providing environment. These approaches have been tested on game-play scenarios (think DeepMind Alpha Go) and are best suited to problems where actions/decisions/predictions can be quickly evaluated.
Machine learning already saturates our world. It is becoming increasingly pervasive in every industry and will shape and reshape our lives as we know them. When it comes to business, the gap between companies who adopt AI and those who don’t will only continue to widen. Which side of the chasm do you want to be on?
The ‘Outlook on the Australian AI market landscape in Australia’ is a 2017 report gauging Australian business sentiment towards artificial intelligence. Download now to learn more about Australian businesses’ willingness to invest in AI, new market opportunities and more.