Is human learning analogous to machine learning

Machine Learning vs Human Learning Part 1

Types of ML and Their Human Learning Theory Equivalents

The origins of machine learning are not easy to determine as it is a field that borrowed many ideas from various disciplines to evolve into what it is today. Some consider machine learning to have developed from statistics as most of its methods are statistically based, while others believe that one of the first few examples of machine learning is Arthur Samuel’s (1959) work in creating a checkers playing computer program that constantly updated its strategies to become better at winning [1].

Nonetheless, it is undeniable that the term “machine learning” has been inspired by the organic nature of continuous improvement in human learning. The process of human learning builds upon pre-existing knowledge, where the knowledge is either modified or reinforced to make it more accurate, and subsequently used to improve decision making and problem solving (Lefrancois, 1972) [2].

In this post, I shall simplify the concepts of the main types of machine learning, and explain what their human learning theory equivalents are.

Supervised Learning

In supervised learning, a computer program is given a training dataset that is labelled with corresponding output values, and a function will be determined based on this dataset (Kotsiantis et al., 2007) [3]. This function, or algorithm, will then be used for classifying new data to predict their corresponding output values, with the assumption that the new data conforms to the rules of the function being used. Linear regression, decision trees, random forest and support vector machines are some commonly used techniques that are actually examples of supervised learning.

Human Learning Theory Equivalent

Supervised learning is similar to concept learning (Bruner & Austin, 1986), where a person is required to classify new objects into existing categories, by matching the features of the new objects to examples in the categories [4].

Unsupervised Learning

In unsupervised learning, the training dataset does not have any labelled corresponding output values (Hastie et al., 2009) [5]. Since there are no “correct answers” to learn from, the objective of the algorithm is to uncover any interesting patterns that it can find in the data, and new data will help to confirm or disconfirm these patterns that it finds. Some of the well-known examples of unsupervised learning include k-means clustering, principal component analysis and artificial neural networks.

Human Learning Theory Equivalent

Artificial neural networks follow the neuronal principle of Hebbian learning, where the algorithm centres on inputs with similar properties, just like how neurons that activate simultaneously strengthen the synaptic link between each other (Sanger, 1989) [6].

Reinforcement Learning

In reinforcement learning, there are “correct answers” but the input data is not paired with the desired outputs. The “correct answers” contain numerical rewards, which the algorithm needs to maximise by choosing the correct actions to take. This is very much similar to a mouse navigating through a maze looking for food, where wrong moves result in the punishment of hunger, and right moves get it to the food more quickly.

Essentially, reinforcement learning is figuring out the right balance between exploration and exploitation, where exploitation gives certainty in the amount of reward, but exploration opens the possibility of finding higher rewards, or risks getting no reward at all. The successful development of artificial intelligence for abstract games such as checkers and Go was based on the concept of reinforcement learning.

Human Learning Theory Equivalent

Reinforcement learning is inspired by the reward system of operant conditioning (Sutton & Barto, 1998) [7]. Operant conditioning was established by B. F. Skinner (1938) [8], after Edward Thorndike (1927) observed how cats learn to escape a puzzle box more quickly after a repeated number of trials [9]. Thorndike noted that behaviours that helped the cats to escape were repeated more frequently over time compared to behaviours that did not, and he termed this the law of effect.

Skinner similarly created a chamber for rats and pigeons, where one lever delivered a reward while another administered a shock. He found that when the animals started to identify what each lever did, pressing of the reward lever was reinforced and frequency of the behaviour increased. Conversely, punishment resulted in decrease of the shock lever being pressed.

* * * * * * * * * *

By comparing machine learning to theories in human learning, the concepts become less foreign and less difficult to understand. But obviously, there are still some fundamental differences between the two, which is essentially the obstacle that is preventing artificial intelligence from becoming more like the general intelligence that humans possess. In the next post, I shall discuss about these differences, and why the current state of artificial intelligence is unlike human intelligence.

If you would like to know more about the differences between machine learning and human learning, check out the second part of this series:

If you found this article interesting, you might also enjoy this one:

References:

  1. Samuel, A. L. (1959). Some studies in machine learning using the game of checkers. IBM Journal of research and development, 3(3), 210–229.
  2. Lefrancois, G. R. (1972). Psychological theories and human learning: Kongor’s report. Monterey CA: Brooks/Cole.
  3. Kotsiantis, S. B., Zaharakis, I., & Pintelas, P. (2007). Supervised machine learning: A review of classification techniques.
  4. Bruner, J. S., & Austin, G. A. (1986). A study of thinking. Transaction Publishers.
  5. Hastie, T., Tibshirani, R., & Friedman, J. (2009). Unsupervised learning. In The elements of statistical learning (pp. 485–585). Springer New York.
  6. Sanger, T. D. (1989). Optimal unsupervised learning in a single-layer linear feedforward neural network. Neural networks, 2(6), 459–473.
  7. Sutton, R. S., & Barto, A. G. (1998). Reinforcement learning: An introduction (Vol. 1, №1). Cambridge: MIT press.
  8. Thorndike, E. L. (1927). The law of effect. The American Journal of Psychology, 39(1/4), 212–222.
  9. Skinner, B. F. (1938). The behaviour of organisms: An experimental analysis. D. Appleton-Century Company Incorporated.