Can machines learn?

Artificial Intelligence, Blog

The second blog of our series on AI in education looks at machine learning, what it is and how it differs from human cognition

Machine learning (ML) is a data-processing exercise and a sub-field of artificial intelligence (AI). The term refers to the construction of the software that enables machines to learn ‘through experience’ using algorithms. It processes sets of observations – or data records – and infers new patterns or rules arising from these.

When the data is changed, the ML algorithm spots this and learns, so it can produce or predict a new result.

ML was used to detect spamming in the early 2000s, when the volume of spam threatened the effective use of email. It was the only approach that was able to learn and adapt fast enough to changes in spammers’ tricks to manage the problem. It is now able to recognise spam and file it in the ‘junk’ folder of your email system.

However,  ML is sensitive to the data it uses – if it is inaccurate, irrelevant or insufficient, an ML application will not be able to meaningfully induce rules or models from it. Put simply, ML illustrates, amplifies and perpetuates past behaviour.

There are two types of commonly used ML algorithms: supervised and unsupervised. The key distinction between the two is the level of human supervision and annotation required for the algorithm to function. ‘Unsupervised learning’ is used when we do not have the values of outcomes at our disposal, and there is no ‘human guidance’ inherent to the algorithm. A third, but less common, type of ML is ‘reinforcement learning’. Like supervised learning, this uses feedback to find and learn the ‘correct behaviour’.

Unlike the human brain, ML can’t solve problems that it has not been trained to solve, or it doesn’t know about. Professor Rose Luckin, director of UCL EDUCATE, has identified seven elements to human intelligence that distinguish it from the potential of ML:

  • Understanding and knowledge of the world
  • Social intelligence and ability to interact with others
  • Meta-knowing intelligence – what it means to know something
  • Meta-cognitive intelligence – being able to interpret our own thoughts
  • Meta-subjective intelligence – regulating behaviours through their recognition
  • Understanding our physical interaction with the environment
  • Perceived self-efficacy – making judgements about ourselves

Professor Luckin believes the core difference between human and machine cognitions are “the ability to reflect on learning, to socialise, feel emotions and develop sophisticated self-understanding and control.”

These are key attributes to human intelligence that EdTech developers and innovators need to take into account to design their products and services, as we have seen some of the companies on our UCL EDUCATE programme do.

For further information, download our Byte-sized edtech research on machine-learning.

You can read the two other blogs on this series here:

  1. What is AI?
  2. The future of AI in education

Our partners