Pattern recognition and machine learning: computers as children?
Picture 1: Pattern recognition & learning. (source: Linda Liukas http://www.helloruby.com/about)
Our present preferences rely on the accumulation of our past experiences (Bargh & Morsella, 2008). Psychologists propose that these experiences are stored in our memory not as separate events or incidents, but as an enormous amount of neuronal cells encoded into complex patterns (Dijksterhuis & Nordgren, 2006; Simon, 1987; 1996). Indeed, these patterns are very complex—sometimes too complex for our rational mind to understand and regulate them.
Despite their complexity these patterns follow a simple rule. Describing this rule, psychologists rely on Sigmund Freud’s (1895; 1911; 1930) “pleasure principle” pointing at the natural for all living creatures tendency to increase their pleasure while avoiding pain (see e.g. Custers & Aarts, 2007; Epstein, 2003; Lazarut, 1991). Describing the behavioral processes that follow this principle Simon (1987; 1996) calls them as “If-then” patterns. If something brings one a pleasure – then one naturally tries to repeat and increase it. If something brings one dissatisfaction and pain – then one naturally tends to avoid that. If experienced driver sees the red light on the road, then one automatically pushes the brake pedal. This happens without one’s need to every time rationally justify the performed action since knowing about the possible negative consequences (e.g. possible car crash, hitting pedestrian, getting a fine, etc.) of driving on the red light. The “if-then” patterns help us in a variety of situations of our daily life if/when supported by appropriate training (repetition) and experience. We gain this experience through interaction with the surrounding us environment starting from the first day of birth. This is the reason why a large share of child education derives from this “if-then” experiential pattern matching ontology.
Computers are like little children. Both, kids and computers, follow the same “if-then” pattern recognition and matching principle. Both, kids and computers, require the linearity in their learning approach. In order to learn, they need to start from a set of basic patterns/commands and then gradually increase their complexity. You will have trouble with teaching astrophysics to a child who does not know anythig about physics and chemistry. You will have the same trouble when trying to teach computer the commands that do not exactly follow its existing patterns. The book of Finnish programmer Linda Liukas builds on these similarities while teaching children about the ABC of Programming. I believe that understanding of the way computers learn may also help us with increased understanding of child learning.
Unlike computers, little children grow up and in addition to “if-then” patterns start using the rational decision-making processes that provide them with the “human” abilities computers do not yet possess.
Despite their growing data gathering, sorting and analyzing capacities, computers learn through the same child-like “if-then” pattern matching mechanism. What happens when computers grow up? What is the next step in machine evolution? Will they ever grow up?
References:
Bargh, J. A., & Morsella, E. (2008). The unconscious mind. Perspectives on psychological science, 3(1), 73-79.
Custers, R., & Aarts, H. (2010). The unconscious will: How the pursuit of goals operates outside of conscious awareness. Science, 329(5987), 47-50.
Dijksterhuis, A., & Nordgren, L. F. (2006). A theory of unconscious thought. Perspectives on Psychological science, 1(2), 95-109.
Epstein, S. (2003). Cognitive-experiential self-theory of personality. In Millon, T., & Lerner, M. J. (Eds), Comprehensive Handbook of Psychology, Volume 5: Personality and Social Psychology (pp. 159-184). Hoboken, NJ: Wiley & Sons.)
Freud, S. S. (1895) Project for a Scientific Psychology.
Freud, S. S. (1911) Formulations on the Two Principles of Mental Functioning.
Freud, S. S. (1930) Civilization and its Discontents.
Lazarus, R. S. (1991). Cognition and motivation in emotion. American psychologist, 46(4), 352-367.
Liukas, L. (2015) Hello Ruby: Adventures in Coding.
Simon, H. A. (1987). Making management decisions: The role of intuition and emotion. The Academy of Management Executive (1987-1989), 57-64.
Simon, H. A. 1996. The sciences of the artificial (3rd ed.). Cambridge, MA: MIT Press.
Users who have LIKED this post:
6 comments on “Pattern recognition and machine learning: computers as children?”
Comments are closed.
Hi Natasha thanks for your post 🙂 I was wondering if you stumbled upon any new methods that allows AI to learn. I read something about creating situational frameworks that could replace the need for relatively large data sets used in regressions. I think this process requires the programmer to specify various scenarios though which in a way could limit the versatility of such software.
Users who have LIKED this comment:
238A
I would like to catch up on your questions in the end of the article. I think the definition when a computer is more “human-like” and has general AI, is changing in correlation with the improvements in this field. While in 1950 one would regard google maps as general AI, we see it differently today. I guess even if we have made huge improvements in AI 20 years from now, people in the future would not regard it as general AI because the improvements in distinguishing between human and machine would have also improved. It remains a question of defining AI and human intelligence.
But still, the progress we have observed in machine learning is astonishing. I agree with you that distinguishing between a computer doing “if-then” processes and more complex ones, will get more and more difficult in the future. But right now, I also find the “if-then” concept quite practicable in order to do so.
I recommend two good articles that fit into this discussion:
Searle, John R. – Minds, brains and programs — On the ethical implications of distinguishing between human and machine.
https://www.nytimes.com/2016/12/14/magazine/the-great-ai-awakening.html — A very good consolidation of state-of-the-art AI research and its origins.
Users who have LIKED this comment:
Great post, Natasha! If computer can only learn from the massive historical data and run ‘if-then’ process, the result would be limited to the best ‘past’ solution in a given circumstance. Leon has suggested two interesting readings. Here is another interesting one:
Rise of the machines: Google AI experiment may lead to robots that can learn without human input
http://www.dailymail.co.uk/sciencetech/article-4420804/Experiment-lead-machine-s-learning-without-humans.html
Users who have LIKED this comment:
Very interesting!
I just wanted to add that the process of “if-then” is not the only way AI can “learn” or become “smarter”. Machine Learning is already becoming very popular for programmers that wish their programs to be able to create things independently of the code. This would add a “rational” sense to the thinking capability of machines. Indeed, some machines were able to create their own images, texts, and even music! This is revolutionalizing the way we think of machines. Artificial Intelligence combined with Machine Learning makes a computer more than just a kid.
Thank you for your great post, Natasha. Will AI ever grow up? I think the determining factor is “free will.” If the AI has the freedom to access any data set it chooses and decide what “being” it wants to become then yes. Through “free will” AI can grow up. The unfortunate risk of that is that *empathy* is supposedly genetic, not learned.
Thank you Natasha for a wonderful post! Your proposition that kids and computers follow the same “if-then” pattern recognition and matching principle is very surprising to me, as I’ve always thought that kids and computers couldn’t be further away in terms of how they think learn and function.
Whereas I agree with your argument that they both require linearity in their learning and use pattern recognition, I feel that learning for kids is often quite serendipitous opposed to rule based. I think that children excelling at the Marshmallow Challenge (https://www.ted.com/talks/tom_wujec_build_a_tower) is a great example of this, how do you think that computers would perform?
The other major difference between kids and computers is that whereas AI solutions still tends to focus on a very narrow problem, children learn about a multitude of subjects simultaneously, and often apply insights from other fields to their learnings. This often becomes evident in the hilarious things kids say when they make incorrect connections!