Theory of mind (in Machine Learning)
The reason why there are no androids on the streets, indistinguishable, mistaken for humans, is because sometimes we take for granted that many things can be simpler than they appear and the idea of an Artificial Mind is one of those engineering challenges. that requires the concert of many sciences as it is not simpler than sequencing the human genome or establishing a Martian colony.
Educating yourself is, I firmly believe, the only education there is.
Isaac Asimov (1920-1992)
What is the mind and what is the brain?
The mind is a system of computing organs designed by natural selection to solve problems faced by our ancestors in their primitive life forms limited to hunting and gathering food [Pinker 1997].
And - then - what is the brain?
If the mind is the software. The brain is the hardware.
Intelligent Systems
For the creation of a machine that emulates intelligence (an Artificial Intelligence) we cannot fill it with so much factual information, but it must be equipped with sets of fundamental truths and rules to deduce its implications (Machine Learning). These rules are what give humans the ability to understand others — and themselves — and deduce / infer what they think. This cognitive ability, behavioral specialists call ' Theory of mind ' (ToM).
Theory of Mind (ToM)
The theory of mind is the branch of cognitive science that investigates how we attribute mental states to other people and how we use states to explain and predict the actions of other people [ Röska-Hardy 2009 ].
For many years psychologists have explored different aspects of the theory of mind in children and non-human primates. In the early 1990s, after hundreds of experiments, the widely accepted claim that children only understood beliefs from the age of 4 began to be disproved . Today opinion is divided into those who believe that ToM's cognitive abilities are innate and those who believe that they develop between the first 3 and 5 years of a child's life.
The study of these cognitive skills is key in the study of dysfunctions, as in people with autism, studies have also been developed that show that having adequate language skills is necessary to pass the tasks that assess the ability of ToM [Astington 2005 ].
A study takes this line of work much further, showing that deaf adults who grew up without much exposure to language fail in "false beliefs" tasks (are not able to interpret or predict behaviors in other people or lack of empathy ) [Pyers 2009] but once they advance in learning a language, cognitive abilities related to ToM improve.
And how do you measure skills in ToM?
Psychologists, to measure how people think, about their own thoughts and thoughts about others, use different methods and one of the most popular is known as ' False Belief Tasks '. The aim of these tasks is for children to infer or interpret what someone has done or what they are thinking, when the other person's beliefs are in conflict with what the child knows. Understanding the reaction of the other person to the ' false belief ' allows them to learn (This is the challenge that Machine Learning has).
Example of a False Belief Task
For example, a child knows that there are no chocolates left in the box, but will he understand that his brother has no way of knowing just by looking at the box superficially? (remember, we are talking about 3-year-olds).
The classic setting for a 'false belief' task is the " Sally-Anne " test (another video example ). Children are shown the Sally and Anne dolls:
Sally has a basket and Anne has a box.
Sally puts marbles in her basket and leaves the room.
While he's outside, Anne transposes the marbles into the box.
When Sally returns, the boy is asked where he thinks Sally will look for the marbles.
If the child detects the 'false belief', he will respond that Sally will look for the marbles where she left them. So the test has shown that the child understands what Sally thinks and believes.
Gradual testing of ToM
Other TM developmental measurement tasks include measuring other types of tasks by their degree of complexity to be identified, for example, the ability to understand what other people want, ability to recognize hidden emotions or feelings.
Tasks from easier to more difficult
Desires : Understand why another person might want something (and how the desire changes from one person to another)
Beliefs : Understand that different people can believe something different about the same thing or situation
knowledge / understanding : understanding that people may not understand or have the knowledge that something is true.
false beliefs : understanding that people can hold a false belief about a situation or thing.
detect emotions / feelings : understand that people can have hidden emotions or that they can act in one way even though they feel differently.
Adults also fail in false belief tasks
ToM skills develop as children interact socially in play, pretend, stories, and relationships with parents and peers. It has also been shown that it tends to improve progressively and sequentially with age. In a Nicaraguan study by Jennie Pyers and Ann Senghas [Pyers 2009], it shows that deaf adults who grew up without much exposure to language fail false belief tasks.
In this study, Pyers and Senghas followed a unique group of deaf people living in Nicaragua who created their own sign language in the 1970s, when the first school for deaf students was opened in Nicaragua. Research suggests that many different aspects of language are important in developing the theory of mind. The study's discovery has been that exposure to a language — communication — and social interaction are the key to developing cognitive skills at any age.
Humans fail
ToM abilities can be unstable in many cases such as autism, depressed people, people with schizophrenia, people with severe trauma, among other cases that are still being studied. However, humans have also been shown to overcome and learn.
Will the machines fail?
It is unavoidable. Machines — AI — possibly fail. The idea is that we mature our algorithms while the machines learn. Because the key for machines not to fail is in our ability to understand our own minds.
[Dialogue in the movie Yo, robot. Jul-7, 2004 (CA) based on the novel "Yo, robot" by Isaac Asimov ]
—Dr. Calvin, what exactly are you doing here?—My general field is advanced robotics and psychiatry. I specialize in the hardware-brain interface to develop the USR robotic anthropomorphization program.—So what exactly are you doing here?—I make robots more human.—Wasn't it simpler to say that?—Not really. Do not.
---References
[ Pinker 1997 ] Steven Pinker, How the Mind Works, 1997, Publisher: WW Norton & Company; Pages 660; ISBN 978-0-393-04535-2 // Conference related to this book:
[Röska-Hardy 2009] Röska-Hardy L. (2009) Theory Theory (Simulation Theory, Theory of Mind). In: Binder MD, Hirokawa N., Windhorst U. (eds) Encyclopedia of Neuroscience. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-29678-2_5984
[ Astington 2005 ] Astington, J. and Baird, J. (Eds.) (2005). Why language matters for theory of mind. Oxford: Oxford University Press.
[ Pyers 2009 ] Pyers JE, Senghas A. Language promotes false-belief understanding: evidence from learners of a new sign language. Psychol Sci. 2009 Jul; 20 (7): 805-12. doi: 10.1111 / j.1467-9280.2009.02377.x. Epub 2009 Jun 8. PMID: 19515119; PMCID: PMC2884962.
[ Sellars 2009 ] Empiricism, Perceptual Knowledge, Normativity, and Realism Essays on Wilfrid Sellars, Oxford University Press
Links
Sign Language Researchers in Nicaragua
Post about the work of Pyers and Senghas: