How Technological Convergence Poses An Existential Risk
Imagine you are sitting at home relaxing from a chaotic day. You turn on your television, and just as you get into the storyline, the signal is interrupted. Suddenly, a strange image appears, and a voice begins speaking in an unfamiliar language. But that doesn’t matter, because somehow you understand what is being said. A near-instant translation manifests inside your head, and the message is clear.
“Greetings, people of Earth. We have received your signals and have been observing you. It is apparent that you are a danger to yourselves and your planet. You require assistance, and we intend to come to your planet to salvage what is left of it. It will take us some time to reach you as we are located in a distant galaxy on the outer edges of your universe. But don’t worry. We are coming to help you.”
And if that weren’t jarring enough, what you thought was a background prop emerges as the being that has just explained its intentions to “help” salvage Earth. It’s unlike anything you’ve ever imagined. Suddenly the feed is cut, and the world as you knew it is now gone. The clock is ticking to understand this alien civilization, its true intentions, and its imminent impact on humanity. Sounds like mere science fiction...right? Think again because, according to Stephen Hawking, Artificial Intelligence (AI) has the potential to evolve into a superintelligence capable of such feats. And just as we wouldn’t sit and wait for the aliens to invade, we must prepare for the emergence of superintelligence.¹
We live in a time of unprecedented technological innovation. Assuming we maintain this trajectory, further digital and biological infrastructure advancements will merge our physical world with virtual environments. The World Economic Forum has termed the emergence of this era The Fourth Industrial Revolution. What began at the turn of this century is characterized by a ubiquitous mobile internet; smaller, cheaper, more powerful sensors; machine learning, and AI.
“Success in creating AI would be the biggest event in human history. Unfortunately, it might also be the last, unless we learn how to avoid the risks.” —Stephen Hawking
AI is already transforming our lives.
AI is everywhere, and we don’t even realize it. Today’s AI consists primarily of machine learning algorithms. Machine learning is a branch of AI that focuses on using algorithms to automate the process of extracting patterns from data. This process allows machines to imitate the way humans learn, gradually and continually improving their accuracy. Such algorithms are commonly used in everyday devices such as phones, cars, medical equipment, businesses, and social media platforms. The applications of AI will only expand as the technology becomes more scalable, more robust, and more widely distributed.
Machine learning tends to be more accurate than previous methods of prediction. And it’s helping humans perform activities that were previously too expensive, time-consuming, or unachievable. For example, machine learning advancements combined with increasingly powerful computing systems and big data sets have made weather forecasting more accurate in predicting extreme conditions. Before machine learning, scientists used simpler models such as linear regression to predict weather forecasts. These models were time-consuming to create due to the multiple inputs that needed calculating, the time drove cost, and the models weren’t as accurate as machine learning methods. Because regression models containing a high degree of variation can be wrong for specific instances but still correct on average.
On the other hand, machine learning methods may be wrong on average but have less variance creating more accurate results on a per-event basis. As the algorithm receives feedback on its predictions, it becomes more accurate over time. Another way of thinking about this is through a target shooting analogy. Let’s assume you shoot twice. In the first shot, you score -5 to the left. In the second shot, you score 5 to the right. With simple regression, you could consider that a bullseye. With machine learning methods, you’re likely to miss the target completely, but get closer to the bullseye each time you aim.²
Training machine learning models require data. This training involves supervised or unsupervised learning methods to recognize patterns and make predictions. In supervised learning, a training set including labeled inputs and desired outputs is presented to the model. The training set allows the model to learn over time. Supervised learning models fall into two categories: classification and regression. Unsupervised learning models do not rely on labeled datasets and instead discover hidden patterns without human assistance. Unfortunately, even in structured formats, this data must undergo a tedious cleaning process before being usable by machine learning models. Because machines do not possess the ability to rationalize and make judgments as humans do, they rely on humans to help them learn for now.