In Brian Aldiss’s short story Super-Toys Last All Summer Long, we see a world of intelligent robots that very well function as human beings. They mimic the same string of emotions and even strive to understand the meaning of love. In fact, the realm of science fiction has been so long obsessed with sentient machines from Philip K. Dick’s Do Androids Dream of Electric Sheep to William Gibson’s cyberpunk masterpiece Neuromancer. The idea of benevolent machines serving humanity or the opposite—of wiping it out—has its roots deep in the human conscience.
“Infrastructure capabilities needed to implement AI are coming together at a faster rate than expected. Soon, AI will be an integral part of everything that we do and eventually take over most of the activities that require human ability today. Just like every other transformative idea, AI is a double-edged sword that can be used for the good of mankind or for destructive purposes. It is very important that we aim at regulating AI and start working on new social contracts as machines will start producing everything that we need to live our day-to-day life.” — Varghese Samuel, CEO, Fingent.
It’s quite peculiar to think of intelligence in general. We attribute our current level of intelligence and cognitive capabilities to millions of years of painstaking evolution. Our vast cerebral capacity has enabled us to surpass all other species. Now, we’re using our intelligence to replicate intelligence computationally. Revolving around the term artificial intelligence, computer scientists and statisticians across the world are hard at work, attempting to create machines that learn from the world around them more efficiently and reliably.
AI research is catching up with an outpouring of remarkable developments. There are neural networks capable of producing dream-like imagery (Deep Dream), machine learning models that decipher human speech in real-time, algorithms that beat seasoned Go players and recognize images and what they contain much, much more quickly than humans.
Nonetheless, AI in this sense only applies to specific tasks. It’s often called “Narrow AI.” A machine that matches human cognition is referred to as “Artificial General Intelligence (AGI).” However, our focus here is to examine whether AGI will ever equal human intelligence (or even exceed it) in the near future, taking into light the vast progress made in the field but also the massive obstacles that remain.
How Far Away Are We From AGI?
AI isn’t some distant dream but a reality today, even though most AI systems today are relatively narrow and limited—certainly not the AGI for which we’re searching. Most of our smartphones comes readily integrated with some form of AI to assist us, whether it’s capturing the perfect photo or understanding our speech inputs.
Services like Siri and Alexa are certainly narrow rather than general AI. They have the ability to execute tasks based on the verbal commands you provide. They can interpret your voice commands and do a host of things from searching for nearby restaurants to playing your preferred tracklist or giving you directions to your office on the quickest route. They can even tell you a joke or two. But impressive as those feats are, that’s about it.
A true AGI could theoretically perform any task you serve it while learning any new task you may throw at it. That single AI could generalize to new domains distinct from the data on which it was trained.
Breakthroughs in the Search for AGI Show Promise
Several organizations and individuals have started developing algorithms that simulate human-like intelligence, with some notable developments. Here are some of the breakthroughs in AI that hold promise:
- DeepMind, a Google project-turned-company, came out with the Impala architecture. It proved to be a breakthrough in the search for AGI, with its deep reinforcement learning model.
- DeepMind was behind AlphaGo, the world’s first computer program that defeated the world champion Lee Sedol at the ancient Chinese strategy board game, Go. Being a more complex game than Chess, the AlphaGo program utilized a combination of human and AI opponents to train itself of every single move. It’s similar to the historic chess match in 1997, when IBM’s Deep Blue beat the reigning champion, Garry Kasparov.
- IBM Watson is another example of major breakthroughs. It’s chiefly a question-answering system that excels in natural language processing. In 2011, Watson took part in the game show Jeopardy and beat two of the leading champions.
- The ImageNet challenge is yet another major breakthrough in AI research that assesses object detection and image classification capabilities of algorithms on a large scale. Technology behemoths like Microsoft and Google have put to use their own systems that are able to pinpoint and identify images much faster than humans.
A key breakthrough in AGI is underway with a redefined approach to developing neural networks. The Whole Brain Architecture Initiative, with its objective of supporting research at simulating human cognitive abilities artificially, is going a step forward with its novel approach in neural networks. The approach has been to build a computer system that can mimic human brain activity that allows it to learn and think by itself, as well as to develop intelligence gradually. AGI will emerge after all, but as we have said, it will take more time than expected.
How Far Away Is “The Singularity”?
Futurists are predicting a full-scale disruption in the near future. Ray Kurzweil, a famous futurist, thinks that AI will enhance our abilities rather than displace us. Amongst the wave of predictions that he has made over the years, Kurzweil believes that machines will surpass human intelligence by 2045, at which point we’ll become synthetic, hybrid beings—part machine, part human. He calls it the “Technological Singularity.” However, he’s largely optimistic about computers attaining that level of intelligence. Kurzweil claims that by 2029, AI will pass the Turing test and achieve human-level intelligence.
“I set the date for the Singularity — representing a profound and disruptive transformation in human capability — as 2045. The nonbiological intelligence created in that year will be one billion times more powerful than all human intelligence today.” —Ray Kurzweil, “The Singularity Is Near: When Humans Transcend Biology.”
These claims are most definitely up for dispute. Elon Musk and the late Stephen Hawking voiced their concerns over the looming existential threat of AI. Whatever the concerns and optimism are, in this article we’ll keep things in a positive light. There may or may not be a Terminator-like scenario, in which machines try to stomp out the human race. However, these things are not only impossible for the near future; they’re also quite unlikely in the more distant future.
In fact, the claims of AI one day superseding the capabilities of the human brain are largely exaggerated. As far as anything that mankind has yet discovered or invented, the human brain is the most complex thing that exists in this universe. Between our ears sits an organ weighing just three pounds, comprising some 80-100 billion neurons. It would take a supercomputer about 40 minutes to replicate one second of brain activity.
Of course, computing is getting more powerful. In the future, we may see that time lag decreasing and ultimately becoming negligible. Computer science has progressed to new heights, so it’s now possible to replicate specific areas of brain activity more efficiently. For instance, in areas of gathering sensory inputs and information processing, computers are way ahead of the human brain. It’s edging closer to achieving complicated movements, processing visual data, figuring out language and reasoning.
However, the human brain still has the edge when it comes to creativity, empathy, emotions, planning and ultimately consciousness. Computers up to now have not been able to simulate any of the above. The most advanced AI today clearly mimics these vital activities and makes it seem cognitive, but it’s actually not.
“We are at a similar point in the consideration of machine intelligence. Machines are just passing over an important threshold: the threshold at which, to some extent at least, they give an unbiased human being the impression of intelligence.” —Carl Sagan, The Dragons of Eden
Will AGI Ever Equal Human Ability?
The short answer is no. Although it’s made tremendous progress, AI is still sorely lacking the flexibility and creativity that we possess as a species. Even though computer scientists and futurists claim AI can surpass human intelligence, it simply can’t. Medical science still struggles to wholly understand the brain apart from the billions of neurons engaged in crosstalk.
The way the brain perceives, remembers and processes information is something with which scientists struggle today. How can we even know when we’ve reached human intelligence in a machine when we barely know what human intelligence is in humans?
There’s also the ultimate mystery of “consciousness.” It’s puzzling, and it’s baffled philosophers and now neuroscientists for centuries. If we set out to create machines that think, the fundamental thing is to center our understanding of ourselves and to discover what truly constitutes consciousness. Is it just that consciousness is a byproduct of the firings between neurons at various regions of the brain, or something else?
Well before we attain AGI from a technological perspective, there are a whole host of questions we need to answer.
Written by Tony Joseph, a Software Developer at Fingent.