ON ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

11 June 2021|AI and Machine Learning, Data, Data Science, Modeling

AI and machine learning are the same thing (machine learning creates AI), but ‘artificial intelligence’ is a huge overstatement of what machines can achieve. There is nothing ‘intelligent’ about computers and the software that they use. Computers are quite literally dumb; they cannot think for themselves. They are not sentient beings. Nor is AI in any way artificial. It strictly follows rules provided to computers by humans.

Machine learning is also a misnomer. Machines do not ‘learn’ in the way that humans learn. Human brains are reduction engines, incredibly effective at using learning to filter useful signal from a large and wide cacophony of multi-sensory noise. Computers don’t have the same capability. There are no computers that can (or are designed to) act and respond to the world in the way that humans do. Computers are intentionally far narrower, limited to, and focused on, specific tasks set by humans.

Machines apply algorithms — equations provided by humans — to data, and iteratively adapt the elements of the algorithms to become more effective at delivering specific measurable data goals set by humans. The ‘learning’ is at a basic level of “if this equation results in a data outcome closer to the goal, then choose it over one that is less effective”. The machine learning depends entirely on a human-dictated goal, it does not choose the goal for itself. In many applications ML is used to deliver algorithms that are eventually applied as “effective enough”, with no further ‘learning’.

The internet provides the large volume and range of data that machines need to apply machine learning to specific tasks. There are a huge and growing number of specific AI/ML tasks: for example, voice recognition and automation, self-driving cars/trains/ships/planes, medical diagnostics and robotic aids, sentiment engines, facial recognition (image recognition of any kind), fraud prevention and cyber security, traffic systems, recruitment, robotic factories, playing games such as chess and Go etc. In all these tasks (and many more) the computer is not mimicking the human brain, it is applying algorithms to do a better and quicker job than humans for the specific task. However, AI/ML remains far from being able to learn from and respond to the world in the way that human brains do.

A big issue with AI/ML is the potential for bias, in the algorithms provided for ML and the data used by ML. Very few datasets are representative and unbiased. Depending on the data that ML is based on, there are many possible biases that can (unintentionally) emerge. For example, race, gender, religion, location, and age biases are often cited as existing in AI/ML tools used by police forces and the criminal justice system, in recruitment, and in healthcare.

The future for AI/ML is an every growing spectrum of uses, all tackling specific tasks that can be delivered by computers faster and more effectively than by humans. However, as the AI world grows so will the many issues of bias, prejudice, and discrimination. There will also be an increase in AI/ML going wrong in areas that may be disruptive for humans, potentially dangerous, possibly disastrous.

— — — — — — — — — — — — — — — — — — — — — — —

Related posts:

Cerita fiksi ini berdasarkan kisah nyata yang diceritakan oleh AHW kepada saya yang sedang mengkurasi kisah-kisah urban legend seputar aula di kampus saya. Di sebuah kampus yang telah berdiri selama…

Our brand refresh last December ushered in a completely new set of colours, illustration guidelines, and a brand new fully animated logo. The re-design captured Synthesis’ human-centred approach to…

The first step is the hardest. Could it be because there’s no more land beneath your feet after that first step? The dog is barking with urgency, while your heart is beating with passion, and your…