The Implications of the Interface of Memory and Technology on Human Emotion
The concept of technology capable of expressing and reacting to human emotion remains a vast wilderness of unexplored territory in the twenty-first century. Although technology has made great strides in acquiring a rudimentary capability to “think,” so to speak, it is still considered incapable of human-like thinking by many. Alan Turing, one of the most influential and brilliant mathematicians in recent history, proposed in his landmark 1950 paper on machine intelligence the “imitation game,” an experiment to determine whether a machine could respond as a human. Ultimately, he wished to determine whether or not machines are capable of thinking. He believed that by the twenty-first century, machines would be developed to the point that they would respond to questions in a similar manner to humans; thus, the interrogator of the imitation game would have less than a “70 per cent chance of making the right identification after five minutes of questioning.” To this date, no machine has passed the Turing test in its full, unabridged form, though even Turing himself acknowledged the question was “meaningless” at the time.
The question then arises: in what manner could technology respond to human emotion if it is not yet capable of thinking on its own according to the parameters of the Turing test? Artistic renditions of futuristic technology have proposed certain answers to this question. For example, Jordan Harrison’s haunting play, Marjorie Prime, explores the canvas technology offers for the storage and transmittance of memories in a futuristic world. But what other methods could allow technology to imitate human thinking and emotion? In this paper, I will argue that technology is currently incapable of incorporating human-like thinking and emotion into its function. I will do so by detailing the current era of technology and the frontiers technology has crossed; offering information about specific devices that are considered to have demonstrated human-like intelligence, thinking, and/or emotion; and providing the barriers technology still must cross to successfully act in a human-like manner.
Technology thus far has crossed barriers that were once considered impossible to cross. For example, driverless cars have revolutionized the concept of travel, electric cars reduce emissions to continue to protect the environment, and smartphones combine the broad use of the Internet (itself an invention with great capabilities) with the conventional functionality of a regular cellphone. Human genetic sequencing allows for greater knowledge of human disease and its impact on sufferers, while laser and robotic surgery can provide non-invasive methods of fixing a problem within the body. Just a century ago, however, all of the aforementioned concepts would have been considered outlandish and impossible. Even conventional technological devices used daily, such as mobile phones, email, GPS systems, and ATM machines, have fundamentally shaped society and culture through their lasting impact on every facet of life.
One of the largest areas of technological advancement has occurred with the development of artificially intelligent machines, which refer to agents capable of “[receiving] percepts from the environment and [performing] actions.” The field of artificial intelligence itself “attempts…to understand intelligent entities” by “[building] intelligent entities.” As such, it involves the intersection of computer science, mathematics, philosophy, cognitive science, psychology, neuroscience, and various other subjects. Artificial intelligence incorporates the neural network as its basis, which uses “a massive number of interconnected nodes [working] dynamically” to simulate the function of the human brain. These constructions have several advantages – they are capable of learning how to complete tasks with data received from training, can organize the information they receive autonomously, and can “derive meaning from complicated or imprecise data.” They can be considered experts in a specific field and can answer questions about that field.
Artificial intelligence in the current age provides rudimentary, but essential, functions for humans throughout their daily lives. For example, virtual digital assistants such as Alexa, pioneered by Amazon, and Siri, a product of the technology giant Apple, both can respond to questions and control other “smart” appliances in a household. In a nutshell, artificial intelligence seeks to recreate the human brain in a machine, a momentous task which continues to provide challenges to researchers and developers alike.
There are still numerous questions to be answered, however, in the field of artificial intelligence. Critical thinking still eludes conventional computers, though they can capably carry out programmed commands through a set of instructions. For instance, Deep Fritz, a computer program, defeated the chest champion Vladimir Kramnik in 2006 by capitalizing on Kramnik’s mistake. Though later described by the New York Times as “what might be the worst blunder in the history of chess” on Kramnik’s part, Deep Fritz detected this mistake through essentially brute force – developers programmed an algorithm that allowed Deep Fritz to “play chess” by knowing every possible move an opponent could make. Thus, it did not think on its own.
On the other hand, neural networks cannot receive the programmed instructions necessary for conventional computer processing because “they learn by example.” Because of this, the data chosen for the example must be thoughtfully considered because incorrect data could lead to incorrect functioning of the system. As another instance, a neural network has been created with the ability to categorize objects as animals or non-animals after gaining a dictionary of terms and training to respond to stimuli. Technically, however, the machine does not completely think in a human-like manner because it must be trained in the specific task with a tailored and correct example of the task. Neither system provided above can spontaneously understand how to do the task.
Unlike artificially intelligent technology, humans have the ability to “think outside of the box” – that is to say, they “can bring information from outside a domain to think and reason with it,” a task that very few machines can conceivably do. One of the few main examples of this type of thinking in machines is the recently-developed Google DeepMind artificial intelligence system called the Differential Neural Computer. A hybrid system that combines conventional data storage and the learning methods of neural networks, it has been reported to “build on [what is] already in its memory.” The Differential Neural Computer also incorporates an optimization system that “constantly…[compares] its results with the desired and correct ones,” which increases its accuracy over time. This machine therefore acts like a human brain in many aspects.
In theory, this seems a clear-cut case of artificial intelligence demonstrating a human capability for learning and thinking. The truth remains more complex. The Differential Neural Computer incorporates a combination of a neural network and conventional data storage to allow “learning” from a data bank. But it must be retrained for every task it must accomplish, which means it is highly specialized, and is thus unable to “learn multiple tasks.” In addition, learning is dependent on the amount of data with which the Differential Neural Computer is provided. For increased accuracy, this represents a massive amount of data. It thus cannot fully reason as humans do.
However, this type of system is still vastly preferable to conventional computing for solving more ambiguous problems – problems for which humans do not yet have the full answers. Problems involving ordinary differential equations, for example, could be well-handled by a conventional computer, since humans already have the capability to solve these questions and can create an algorithm for the computer. But neural networks could solve problems that involve more unknown questions, such as those regarding patient diagnoses and effective treatments. The challenges lie in broadening the range of tasks artificial intelligence can accomplish and reducing the amount of data required to accurately train such machines.
Another major hurdle artificial intelligence (and indeed, technology in general) experiences is an inability to detect and express intent. Consider, for example, humans – a person who engages in conversation with others has some understanding of social cues and norms, and can detect, to some extent, the intent of the person or people he or she is conversing with. He or she can express his or her intents as well. But machines still struggle with understanding intent, although “recognizing affective feedback” is crucial to interaction. An entire subfield of research (called plan recognition) is dedicated to this field, underscoring its importance to machines. Currently, researchers in this field recognize that technology in general incorrectly assumes that the activity of a user can be modeled “with one dataset sampled from a fixed period of time.”
In addition, machines have difficulty in understanding intent in specific contexts. This is easily demonstrated through the generic spell-check function on computers and smartphones, among other devices. The spell-checker is a complex product of computational linguistics and relies on “a built-in dictionary of words to detect errors” and a probability model “to perform error correction.” However, the dictionaries used in these programs do not necessarily cover all words present in a certain language, which leads to incorrect detection of otherwise valid words by the spell-checker. Context-sensitive spelling error correction programs attempt to combat this issue by determining the “grammatical and semantic contexts” in which errors have occurred. Intent is considered by the program as a result of such contexts, though this concept is still being developed. In the end, although intent is a recognized necessity for the true development of artificial intelligence, it has yet to be fully implemented in technology in general.
The greatest of these unanswered questions is whether or not artificial intelligence will one day have the ability to synthesize and demonstrate emotions. This has been considered as an essential function of any artificially intelligent machine that will work with humans. It has been well-characterized as part of the research surrounding the artificial emotional intelligence field. Emotion is integral to human life. All of human communication requires input from emotions. Many argue that artificial intelligence cannot truly be considered intelligent if it does not have the ability to appropriately detect and respond to human emotion. However, others argue that emotion models require further development of artificial intelligence before they can be implemented.
A model, the OCC (Ortony, Clore, and Collins) model, has been developed to computationally express emotions from 22 different categories. It also has the capability to determine an appropriate intensity of emotion for each type. But it is not fully realistic, partly because the physical structure of machines prevents them from emoting properly. The human face involves 25 muscles to express emotion. The number and quick activation of these muscles in humans creates a barrier to incorporation into machines. And even though some machines do have high expression processing capabilities, they are not capable of understanding those emotions. Thus, it is critical that artificial intelligence is developed further, both physically and computationally, if emotion will one day be incorporated.
The field of artificial intelligence represents a vast and interconnected area of study, one that presents special challenges to all involved in its conception and development. In this paper, I detailed the current status of technology and the frontiers technology has crossed until the present-day, identified specific devices potentially exhibiting human-like intelligence, thinking, and/or emotion, and discussed the barriers (critical thinking, intent, and emotion) artificial intelligence has yet to break. As seen throughout this paper, the term “artificial intelligence” remains a misnomer since all devices with such a label are incapable of fully incorporating human-like thinking and emotion into their functions. Once these barriers have been crossed, however, artificial intelligence will become entrenched into the fabric of society, as ubiquitous as cellphones and Internet are today.