Machine Learning

By Miriam Bogler   January 5, 2017

In a TED talk titled: “Can we build AI without losing control over it”, Sam Harris–a neuroscientist and philosopher–claims that Intelligence is a product of information processing, and as machines are better and faster in processing information and handling huge loads of data, if left unchecked, machines will eventually take us over. In a different TED talk delivered by Anthony Goldbloom about “The jobs we’ll lose to machines – and the ones we won’t,” he confirms Sam Harris’ view that “we have no chance of competing against machines on frequent high volume tasks.” He states that one in two jobs have a risk of being automated by a machine. However, although machine learning allows machines to learn from data and mimic some things that humans can do, machines have made very little progress on tackling novel situations. The fundamental limitation of machine learning, he says, is that it needs to learn from vast volumes of past data, while humans have the ability to connect seemingly disparate threads. Thus, making machines incapable of competing with us when it comes to tackling novel situations. If we accept Anthony Goldbloom’s more optimistic outlook, the question we need to ask then is “What can we do to improve our ability to tackle those novel situations.”

The skills gap

Our education system just started waking up to the needs of a new digital world where most problems require advanced analytical skills, necessary to handle big loads of data. Yet, the majority of organizations (82%) currently have or expect to have positions that require data analysis skills. More than three-quarters of organizations reported difficulty recruiting for data analysis positions in the last year. The expectation is that complexity will grow, making it more difficult to hire personnel with the right skill set that can address the growing need.

The complexity of the environment and the limitations of the mind

Why is it so difficult to find people that can handle those novel situations? One reason points to lack of adequate education. Most schools don’t know how to train students to become analytical thinkers because teachers did not experience analytical thinking when they were in school, and they naturally tend to teach like they were taught. Another reason stems from the complexity of the environment and the limitations of the mind. On one hand, the  environment is complex because of too much information, real time requirement and unpredictable outcomes. On the other hand, the mind is limited because of limited bandwidth of information processing, the limited capacity of working memory and attention and the limited speed of mental operations and learning.

The opportunity is not taken

In order to address these limitations, we have been designing information processing technologies geared to put at our fingertips opportunities for better thinking and learning. But, according to David Perkins, when reviewing student interaction with these tools, it seems that most typically, the opportunities were not taken. The reason, Perkins, Allen and Haffner (1983) point out, is that most people function as “makes sense epistemologists.” Once they reach a line of argument that makes superficial sense, they see no need to develop additional lines of supporting argument or to consider the other line of the case. It is as though avoidance of cognitive load rather than epistemic needs dominates their behavior.

The key to problem solving is how solvers represent the problem to themselves

What type of learning transformation is needed to make it easier to solve problems and tap into our natural ability to handle those novel situations. In his book “Learning to Solve Problems,” David Jonassen states that “the key to problem solving is how the solvers represent or frame the problem to themselves.” In trying to understand this, our first inclination would be to think of a mental model we create in our mind, that relies heavily on what we learn and already know about the problem. However, by doing so, we completely disregard our brain’s limitations as mentioned above and the fact that most problems we face in our life are complex and get more so in era where information proliferates exponentially.

External Representations

However, in spite of these limitations, progress is being made daily and new innovations keep popping up, relying heavily on our ability to connect the dots in the most unusual of situations. What helps us achieve those goals and overcome our natural limitations. As it turns out, our cognitive ability is relying on something we tend to take for granted: external representations. According to Jiajie Zhang, external representations are the knowledge and structure in the environment, such as physical symbols, objects, or dimensions (e.g., written symbols, dimensions of a graph, etc.), and the rules, constraints and relations embedded in physical configurations (e.g., spatial relations of written digits, visual and spatial layouts of diagrams, etc.) They are involved in many cognitive tasks, such as multiplication with paper and pencil, grocery shopping with a written list, geometrical problem solving, graph understanding, chess playing, and so on.

How the writing system changed our cognition

A look back into the invention of the writing system and how it transformed human cognition, may shed some light on the importance of external representations for improving the cognitive capacity of the modern mind. According to Olsen (1996), writing does not merely transcribe, but rather brings structural properties of speech into consciousness. He believes that, the development of writing was the discovery of the representable structures of speech and that the inventors of writing systems did not already know about language and its structures–sentences, words, phonemes, and the like, which were discovered only with the invention of the writing system. Thus, according to Goody (1977) and Ong (1982), rational mode of thought was possible only because writing made available certain knowledge skills, and procedures essential for the rational mode of thought, such as organizing, manipulating, elaborating and reflecting upon logical relations in the analytic form of linear sequences. A conclusion that leads us to what Ong also said: “writing systems are not mere external aids but also internal transformations of cognition.”

External Representations are an intrinsic part to many cognitive tasks

We don’t usually think about the role of external representations in our thinking, but it is amazing to what extent people use them to complete many tasks. Typical types of external representations would be diagrams, graphs and pictures and they are used in many cognitive tasks such as problem solving, reasoning and decision making. Intuitively, we think that any external representation that we use as part of a problem we solve, is first uploaded to our brain and becomes part of our internal mind. But research indicates that it is very unlikely, mainly because of the limitations of our brain and the complexity of the environment. Instead, according to Gibson, the information in the environment can be directly picked without the mediation of memory, inference, deliberation, or any other mental processes that involve internal representation. We rely so heavily on those external representations in our problem solving attempts that they become an intrinsic part to many cognitive tasks, guiding, constraining, and even determining cognitive behavior. Which is why much of the structure of the internal mind is a reflection of the structure of the external environment. According to Donald, changes in cognitive architecture brought about by external representations are no less fundamental than those brought about biological changes in the brain. Writing, as one of those external symbolic systems, is the most important representational system responsible for much of the virtually unlimited cognitive capacity of the modern mind.

What can we do

Learning from the impact that writing had on human cognition, it becomes clear that educating students to create, handle and manipulate external representations in any domain, can deliver the type of cognitive promise that the writing system delivered to the modern mind. By taking external representations for granted and not really understanding their important role in problem solving, we fail to recognize the need for training students to create mental representations of problems and teach them to organize, manipulate, elaborate and reflect upon logical relations within a project domain. The ability to structure information, in ways that make it easy for users to draw conclusions and make decisions, gets more crucial with the proliferation of information. And maybe, as Perkins says, the opportunities will be taken when direct effort is applied to teach general strategies or mental models for problem solving, which will help us keep tackling those novel situations even when we have to deal with a complex and proliferating data landscape.

Bibliography:

Society for Human Resource Management. “Jobs of the Future: Data Analysis Skills.” Survey. Society for Human Resource Management. November 2016. Retrieved: Dec. 27, 2016. https://www.shrm.org/hr-today/trends-and-forecasting/research-and-surveys/Documents/Data-Analysis-Skills.pdf

Zhang, Jiajie. “The Nature of External Representations in Problem Solving.” Cognitive Science Vol 21 (2). 1997, pp. 179 – 217. Retrieved: Dec. 29, 2016. http://csjarchive.cogsci.rpi.edu/1997v21/i02/p0179p0217/main.pdf

Perkins, D.N. “The Fingertip Effect: How Information-Processing Technology Shapes Thinking.” Sage Journals. 1985. Retrieved Apr. 25, 2016. http://journals.sagepub.com/doi/pdf/10.3102/0013189X014007011

Olson, D.R. (1976). “Culture, technology, and intellect. In L. B. Resnick (Ed.) The Nature of intelligence. Hillsdale, NJ: Erlbaum.

Perkins, D.N., Allen, R., & Hafner, J. (1983). Difficulties in everyday reasoning. In W. Maxwell (Ed.), Thinking: The frontier expands. Philadelphia: Franklin University Press.