There’s no doubt about it. Technology continues to advance at impressive rates. And while the novelty of technologies such as self-parking cars and robotic vacuums have worn off, we are still many years away from the age of computers capable of human thought. Well, that was true before the development of Artificial Neural Networks (ANN), of course.
ANN is one of the only techniques currently available for training machines to truly think like people, and it is a tool used within the deep learning space.
Artificial intelligence, defined broadly, is the field of training machines to autonomously perform tasks normally thought to require intelligence. Beneath that umbrella is machine learning, in which machines autonomously learn new tasks, and deep learning is a further subcategory of machine learning. It involves training machines to process information in a way that mimics the way the human brain learns new things. Enter artificial neural networks.
The nitty gritty of ANN
A biological neural network refers to the neurological system through which humans and animals process information. If we think of the human brain as a computer, each system contains neurons for sensing and collecting new information or data. After new data is collected, it is communicated through the synapses, or data pathways, of the brain. When that information is received, it is processed, and the brain determines which new neurons if any, that data should be sent next. This process is called Edging.
After Edging, that data is used as an input to an inherent non-linear function, which makes sense of the data. Data is not weighted equally. For example, input data that states a shoe is untied would be sent to a decision-making faculty of the brain – do we tie or not tie the shoe? Input data that states a tiger might be lurking in a nearby bush, however, will be sent straight to the fight-or-flight part of the brain, and all other bodily systems will be notified of the threat.
Understanding the human brain is critical for understanding ANN. With ANN, artificial systems mimic the same functionality of the human brain. Instead of neurons, artificial systems rely on nodes for data collection. New datasets, typically an input of numbers, are put through an initial algorithm for preliminary processing and/or organization. ANNs are typically at least six layers deep but often are deeper. Each layer takes an input and further processes it, allowing the machine to make an informed decision about what task to perform next.
Let’s say the input data refers to a user’s music selection on a single day. That data will be processed, organized, and sent to the next relevant algorithm via edging. In this example, the next relevant algorithm might compare today’s musical selections to the selection of the previous day, week, month, or year. From there, a third algorithm might make predictions about that user’s musical preferences by genre, artist, and song. The final step in this process might be adding those songs to a suggested playlist. As a user listens to that playlist and likes new songs, the ANN will continue to train itself, determining whether the original song suggestions were correct and making informed adjustments along the way.
In this way, ANNs are highly trainable, scalable, and effective.
Applications and concerns
ANNs are used in a number of applications today. Some examples of relevant applications include music suggestion algorithms for platforms like Spotify and Pandora; ad suggestion tools based on the buying habits of an individual user for platforms like Etsy and Amazon; and complex industrial applications such as predictive algorithms that monitor the potential for on-site emergencies (e.g., nuclear meltdowns, malfunctions of critical machinery, and injury of personnel).
ANNs will only continue to get more powerful as Big Data grows, which gives rise to some concerns. Some theorists argue ethics must be built into ANNs, which prioritize the value of things like human life, to avoid developing a machine like that depicted in the movie Terminator.
Perhaps this is because some mathematicians understand that if our existence were to be decided upon based on factors like impact on the planet and threat to other species, we would be guilty on all accounts. However, that is not to say we human beings are not worthy of life. We do much good alongside our habits of destruction. Movies like Arrival argue this point. Still, machines are far from being able to understand, value, and appreciate the concepts of beauty, truth, and love. For that, it is essential to develop these algorithms with ethics in mind.
Other things come into play with regard to the ethics of ANNs, such as the importance of keeping personal information secure from both corporations and governments. We live in a world where cameras with facial recognition processing boards line street corners. For persons with Smart devices in their homes, that data could be used to determine when someone is home or not (could this information be used in a court of law? And would its use be ethical to use this information if someone were accused of murder and it conflicted with an alibi?).
These are the questions the developers of today are tasked to answer. The best answer we have now might be to encrypt all of this data, lest it gets into the wrong hands. But who is to say how data will impact the ethics of our future society? At least for now, the way we develop new algorithms might have a say in how these data are used in the future. It’s a meaningful conversation to begin having now, and thankfully it is being discussed in depth. Otherwise, if our future does not look like Terminator, it might instead look like 1984. And both of those outcomes seem grim, at best.