Even though inventor, computer scientist and futurist Ray Kurzweil has famously predicted that we will achieve technological singularity by 2045 and perhaps by 2029, AI will pass a valid Turing Test, the internet is already abuzz with wacky predictions of having a winning shot at Singularity in 2018. Before we dive into the 2018 predictions, let’s define technological singularity — a hypothetical moment in time when AI will have progressed to the point of a greater-than-human intelligence.
Interestingly, the term “singularity” was first used by was by mathematician John von Neumann. But it was popularized by science fiction writer Vernor Vinge, who posited that artificial intelligence, human biological enhancement, or brain-computer interfaces could be possible causes of Technological Singularity. According to futurist Ray Kurzweil, Singularity will occur around 2045 whereas Vinge predicts a swifter timeline — sometime before 2030. Leading tech companies, scores of startups and AI researchers are racing to develop a swathe of AI solutions that are significantly augmenting human intelligence. Recent developments in the AI world are in a way giving us a picture of our Singularity future that is fast coming into view
AI enthusiasts and proponents of Singularity have thrown in several plausible arguments that put us within a shouting distance of achieving visions of singularity in 2018:
Google Multi Model ushers in narrow AGI: This neural network from Google Brain learned how to perform eight tasks such as image and speech recognition, translation and sentence analysis simultaneously with striking accuracy. This new approach from Google takes us one step closer to Artificial General Intelligence. Another key highlight of MultiModel is using its knowledge of one task to use it to solve a different problem. The model shows transfer learning from tasks with a large amount of available data to ones where the data is limited.
Google AutoML designs Neural Nets better than human scientists, ushers AI inception: Google AutoML. It’s already designing better neural nets than the best human scientists when it comes to image recognition. Earlier this year, the Mountain View search giant rolled out AutoML project – a machine learning model that can create other machine learning models through a technique called reinforcement learning. And when applied to a computer vision problem, the results bested the best state-of-the-art computer vision systems created by humans. The researchers applied AutoML to the ImageNet image classification and COCO object detection data set — two of the most respected large-scale academic data sets in computer vision. NASNet can address “multitudes of computer vision problems we have not yet imagined,” the researchers write.
According to Google researchers, NASNet surpassed the Inception models built by the Google team earlier. CEO Sundar Pichai had an interesting take on AutoML, he joked that they had achieved “AI Inception” wherein AI systems create better AI systems. AutoML could bring massive improvements in self-driving technology, image recognition technology utilized in Google Lens and even speech recognition tech.
Advances in AI Hardware converge on Strong AI: There are several advances in AI Hardware that are unleashing the arrival of Strong AI. Increased focus on quantum computers, rethinking chip architecture and novel computing devices is acting as a catalyst for the next wave of AI advances. According to IBM research, the new equation for AI innovation hinges on new hardware capabilities that accelerate training and running of Deep Learning models. However, the real breakthrough will come from the new AI hardware developments – the neuromorphic chips and quantum computing systems, which can help create future AI systems that can mimic human reasoning. According to IBM research, while this might not happen in the very next wave of AI innovation, it’s in the sights of AI thought leaders.
Researchers have discovered a roadmap for General AI: According to Numenta Co-founder Jeff Hawkins, AI researchers have discovered the secret to Strong AI. Despite the breakthroughs, researchers understand the limitations of current AI techniques and although Deep Learning networks have achieved spectacular accuracy for many applications, a minor failure can be catastrophic. According to Francois Chollet, a Deep Learning pioneer, “You cannot achieve general intelligence simply by scaling up today’s deep learning techniques.”
It is because of these failures, Hawkins notes, that AI proponents are adopting a different approach. That’s why, much of current research has gravitated towards brain theory. Hawkins observes in his blog that the “Roadmap for general AI is rapidly forming in the world of brain theory. It may take several years for the discoveries in brain theory to be fully integrated in AI, but the roadmap for how to get there is clear.”
Discovery of Generalized Learning Algorithms will lead to Strong AI: This one is definitely a far short and is known as the crown jewel of AI that once discovered, will lead to a wave of Strong AI. This breakthrough — discovering Generalized learning algorithms (GLAs) will allow strong AI to be realized. A research paper cites that GLAs should not be confused with artificial general intelligence. In fact, GLA is not a single algorithm required to give build strong AI implementation, but a foundation. The paper postulates that once GLA is discovered, we will have exited the era of narrow artificial intelligence and conventional machine learning.
Today, there are slew of companies and startups such as Opencog, Numenta, Allen Institute for Artificial Intelligence, Google-owned Deepmind, OpenAI and even IBM that are working towards building a Strong AI. Despite the spectacular technological breakthroughs, there is a group of researchers who believe we will never exceed human intelligence. A renowned neuroscientist from Max Planck Institute for Brain Research in Frankfurt claimed that one can never surpass human intelligence, only approach it asymptotically approach it. And there is another set of believers (Elon Musk, Stephen Hawking) whose arguments fall on the side of fear-mongering. Despite all the arguments, one things is clear – recent developments are putting us within reach of the not-so-distant future when machines surpass human levels of intelligence.
Try deep learning using MATLAB