Artificial Intelligence.. Isn't All That Intelligent

Quite the shocking claim; yet, believable, nonetheless. To a world where artificial intelligence has a dominant position in the economy and a predominant role in the lives of each and every human, it’s surprising to consider that it may not be as smart as we believe. But it’s not to say that artificial intelligence is dumb, or incapable of most things. It’s just that society’s beliefs and expectations for artificial intelligence are far, far, far too high. And this simply needs to be disproved, in someway, shape or form.

People claim that artificial intelligence may reach “human-level” intellect in the nearest 10 years. These viewpoints of overestimation have been incorporated in history ever since the mid-1900s. In fact, in an MIT research project in 1966, an undergraduate was claimed to have to create an artificial intelligence model with the vision of a human being in a single summer - an insanely difficult project that he, obviously, could not complete.

They defend these claims by pointing at self-driving cars. Automated airplanes that can go on autopilot the entire flight. Waiter and waitress machine learning robots that operate efficiently and productively. But society simply hasn’t mastered these inventions unlike the media leads people to believe. While some of them exist at light levels and are currently being expanded on, we are very far from having these inventions become normalized in society. There’s a reason no restaurant is fully automated, and not every car on the highway can maneuver without a passenger. It won’t all change in the next 10 years.

Last year, Google took a giant leap and invented Duplex, an assistant that has its own form of human intelligence. While its function appeared simple, the ability to call other human beings and create reservations or purchases for dining or movie tickets, the amount of effort required for it to operate efficiently was incredulous. Google had to design an intricate and complex path for information to be sent from the user to Duplex to perform its actions well enough to be used by anyone. Its presence shows the difficulty in developing AI to match almost a tenth of the intelligence of a human being. To get to 90% or more would be unreal.

The advanced path for information from the user to Duplex

The advanced path for information from the user to Duplex

Deep learning is a method for training AI that stems from the beliefs of these people. In essence, deep learning turns AI into a human with its own brain that contains neurons and receptors that enable it to perform its own independent functions. With enough data, we can learn how to alter each neuron inside the brain of the AI prototype, to recognize objects, memorize patterns and languages, and other mentally-based human actions. This extends itself to a car stopping when seeing humans in its path, or a camera tracking movement and unexpected actions.

In the end, it is valid that programmers and technological innovators are taking steps to create AI to be more human in its intellectual abilities. But normalizing it and viewing it as anything short of remarkable is where society must draw the line. It isn’t easy to create these models, and even with Deep Learning’s growth and development, it will still be unreal to observe AI prototypes working on their own independently and efficiently. And who knows what possible outcomes this could entail? We can’t tell who for sure, but we can agree it won’t be the same people who believe AI will be as smart as a human in the next 10 years.

Yousef Khan