Why AI Isn't Taking Over the World Anytime Soon

When I tell people I work with machine learning and artificial intelligence, I often get asked "Are robots going to take over the world?" Many people have expressed their fears I AI taking over jobs that are currently being done by people and increase unemployment. They fear a change in their current way of life and a lack of privacy. With AI and smart devices rapidly becoming such an integrated way of life in the past few decades, who's to say it cannot expand to an ultimately unstoppable force? Since technology has made exponential leaps, it would seem a robot apocalypse would be just around the corner. Well...it's not.

Experts have categorized AI into two categories: Artificial Narrow Intelligence (ANI) and Artificial General Intelligence (AGI). These categories are also referred to as weak and strong AI, respectively. What we have currently is considered ANI. Some examples of this are the algorithms that suggest what you should watch next on Netflix based on your previous watch history, your Alexa and Siri smart assistants, and the artificial chess player that beat the number one human chess player in the world. These are considered narrow AI because each one was built to complete one specific task. The Netflix algorithm cannot tell you if your friend's birthday is coming up and the chess AI cannot recommend shows you should watch. They weren't built to handle multiple tasks, but rather be very efficient in handling a narrow task.

Even the robots in interviews that have been circulating the internet lately are examples of ANI, even though they are meant to appear as an example of AGI. If you notice, the interviewer sticks to a set of general questions to ask and the robot circulates through a list of general responses based on the question asked. For example, Sophia, made by Hanson Robotics, is really good at giving answers closely related to the questions given to her, but she cannot tell you what Netflix shows you should watch. She would most likely tell you some of the most popular Netflix series at the moment instead of giving a personalized answer. Sophia is more like an Alexa wrapped in a humanoid exterior than a human.

With this in mind, it doesn't make ANI any less impressive. ANi has brought us self-driving cars, chatbots, autofill search browsers, social media algorithms, song and tv recommendations, and so much more. It has become a crucial part of our daily lives and a useful tool in developing new technologies. As diverse and useful as ANI is, it is leaps away from AGI.

AGI is the type of AI we see in movies. AGI is AI that can generalize as humans can; it is able to problems solve based on past experiences and process emotions. We aren't close to achieving this level of AI and I don't think we will be for a long time because there are many limiting factors. Neural networks that make up AI are modeled after the neural pathways in our own brains. While we understand a lot about how our brains and cognitive processes function, a lot still remains unknown. I think to be able to replicate our intelligence we first need a better understanding of how our own brains work. This would allow us to have mnmore control over machine learning instead of treating an AI like a black box. We understand that humans learn by experiencing an emotion or event, taking previous experiences, and applying this information to new and unfamiliar situations. In machine learning and AI, we call this transfer learning. This helps generalize AI by having it learn from a few basic pieces of data instead of needing an exact example of every possible situation and a unique response for each. This also takes up a lot of memory which is physical space in computer hardware. We are limited by how much data we can store and how many processes can be conducted at once. Even the most advanced supercomputers cannot currently compete with the complexity and speed of the human brain.

There are many advancements that would need to occur to make AGI a possibility. I think a better understanding of neural processes and quantum computing would need to come about first. For now, we can enjoy Siri telling us what song we are humming and watch on in awe as we build self-righting rockets. There's no reason to panic about a robot apocalypse. At least, not yet.