Originally published on
As our society’s technological progress marches forward, we’ve become ever more fascinated with the concept of artificial general intelligence (AGI).
From IBM’s Jeopardy-playing computer, Watson to television programs like Westworld, we’ve collectively begun exploring and philosophizing about the potential of AGI. Of course, most discussions about AGI in our popular culture are focused on the future, and not the current realities of the present when it comes to artificial general intelligence. Below, we’ll discuss the current realities of AGI and what breakthroughs we’re on the cusp of in 2018.
How Close Are We To True Artificial General Intelligence?
Software programs built into your iPhones and other hardware like Siri or Amazon’s Alexa may lead you to believe that we’re very close to true artificial intelligence. But in order to actually assess that in a valuable way, we have to come to a common understanding of what true AGI means. After all, how can we hit a target if we don’t know what we’re aiming for?
The Definition Of True Artificial General Intelligence
Now, every piece of technology we own has some sort of “artificial intelligence,” but they’re mostly only equipped to handle certain tasks. Siri can listen to your voice commands and perform some tasks for you. But the tasks it can perform are limited to the box that Apple creates for Siri. It can only pull information from applications on your phone. Siri can give you weather information, browse the web for you, play music, etc. This, while impressive, is not true AGI.
A machine with true AGI would be able to perform any intellectual task a human being can. This means if you asked a robot with AGI to hammer a nail, it wouldn’t need to be programmed to do so. It would try — and possibly fail — on its own. It would be able to learn from its mistakes and try until it got it right.
A human doesn’t need to be taught how to walk. You just figure it out through trial and error. A computer with AGI could theoretically learn in the same way. To put it simply, a computer today still will only do what you, or a programmer, tell it to do — and nothing more. It can’t learn from mistakes or intuit anything through common sense. But that may soon change.
Possible AGI Breakthroughs In 2018
While we are still at least two decades away from true artificial general intelligence, there are some incredible people and organizations working on ways to make computers function like the human brain.
Hiroshi Yamakawa And Whole-Brain Architecture
You’ve probably heard of technical terms like “neural networks” and “machine learning” in the news and other media the past few years. Machine learning simply refers to the idea that if you give a computer a large enough amount of data and a large enough amount of ways to interpret and direct this data, it could “think.”
IBM’s Watson, by using the entire internet as its database, was able to answer Jeopardy questions thanks to machine learning. A neural network is meant to create a type of machine learning that effectively mimics the network of neurons in the human brain. This field is in its infancy, but Hiroshi Yamakawa and his Whole-Brain Architecture Initiative are working on revolutionizing the neural network. As of right now, neural networks that exist today, despite how complex they are, are still primed to complete a certain task. Watson answers Jeopardy questions — others are designed for facial recognition or to mimic human handwriting. Essentially, they mirror a single aspect of the human brain’s functionality. Yamakawa argues that we need to expand our neural networks further, interconnecting them and allowing them to feed off each other like the human brain does.
Yamakawa’s whole-brain neural network will allow a computer to complete tasks and “think” in ways that were not considered in the design phase of the computer. It will theoretically learn new things and perform new tasks of its own free will. Ideally, Yamakawa argues, that since the neural network will be designed after the human brain, it will be easy to communicate with and relate to the computer when it becomes truly artificially intelligent.
A big hurdle to mimicking the human brain is capturing its creative essence. At their core, traditional computers are pretty dumb. They take things super literally and can only answer true or false questions. This makes computers very good at math or performing any intellectual task that requires a right or a wrong answer.
But what if you wanted a computer to perform a task that had no correct answer at all? What if you wanted a computer to paint or write a novel? A painting can’t be true or false. It just is. Getting a computer to understand that can be difficult.
But what was previously thought impossible is now a reality. Recently, a Japanese AI utilized a complex neural network to write a book that nearly won a literary award. Granted, it still required some help from humans. The team that created the AI still had to give the computer the plot, the characters and their genders in order to write the novel. But the sentences — their structure and flowery language — were written autonomously. The computer also still required a database of previous writing to get started, but that isn’t much different from how the human brain works. And we can only expect the technology to become more sophisticated as time goes on.
Want To Learn More About AI?
Hopefully, this article cleared up some misconceptions about artificial general intelligence and gave you a clear idea of what we can expect from our robot counterparts in the future. We’re still a long way away from true artificial intelligence, but every day we come closer.