Prague, Czech Republic, August 19, 2016
Josef Gemela, Associate Vice President, Consulting, IDC CEMA
In the famous Terminator movie, a super-intelligent computer network gained artificial consciousness and decided to wipe the human race from the face of the earth. In The Matrix movie — a similarly dystopian vision of the future — intelligent machines enslaved humans to use their bioelectricity for power. Two years ago, Stephen Hawking warned the world that the development of full artificial intelligence (AI) could spell the end of the human race. Bill Gates and Elon Musk have articulated similar concerns. Today, we are witnessing the rapid deployment of a new generation of computer systems, together with the fast uptake of Internet of Things (IoT) technologies, smart products, robotics, and other forms of interconnected intelligent machines. Does this mean that AI is progressing to the final phase in which it can pose an existential threat or take control over our lives? The answer is both yes and no.
The progress AI has made over the last 60 years is quite disappointing. The history is full of hypes and overenthusiasm regularly connected with the introduction of more advanced computer systems or novel computational models (e.g., self-learning artificial neural networks), followed by periods of disappointment, disillusion, and criticism. Many AI scientists believed that the first general intelligence machines would be constructed in the 1970’s, but all the efforts and research grants yielded computer programs capable of solving only very simple problems. However, smart cars, smart homes, smart healthcare, smart utilities, and other smart/intelligent industries are today’s reality, and AI is (again) a major obsession for many technology companies. Let’s take a deeper look at what is really happening right now.
To put it simply, the human brain consists of several layers with its “front-end” system controlled by sensory and motor neurons. These are responsible for collecting and understanding inputs received by sensors and triggering an appropriate action through sending the right signals to our muscles. The process is predominantly undertaken at the unconscious level, as all our routine reactions become “wired” into our neurons and synapses over time. This wiring is enabled by the brain’s ability to learn from repetitive situations and tasks. For example, after several years of routine, you can drive your car from home to the office pretty much “automatically”, and your consciousness is only awoken in situations when something unexpected arises that disrupts or contradicts the model wired into your brain.
In line with the fast deployment of IoT technologies, we are seeing AI make tremendous progress in the area known as human-assisted intelligence. Human-assisted technologies are helping us with many front-end tasks. Today, we have better-connected and mobile electronic sensors, we have more advanced systems for understanding the signals (voice, image, or mental state recognition), and we have developed better decision-making applications that are able to analyze high volumes of data. When looking at the different components of the IoT (sensors, connectivity, data processing, and decision-making engines), none of them is revolutionarily new. However, the key breakthroughs that IoT has brought about are a) the ability to perform the whole sensors-decision-action process in real time and b) the ability to integrate the sensors or the whole system into various mobile products, such as vehicles and smartphones. As a result, IoT systems have enabled the pervasive introduction of various “smart” products or environments, with electronic systems able to perform specific sensors-decision-action tasks equally as well or even better than humans. However, the AI-powered computational algorithms used in IoT decision-making systems are still very simple, involving either “if-then” rule-based models, or more advanced expert systems but with limited self-learning capabilities. Beyond the front-end system, the human brain is even more complicated, and composed of numerous components responsible for thinking, consciousness, formulating aspirations and objectives, anticipation, and social and cultural behavior. These are all aspects of self-awareness that Skynet and the Matrix possessed, but also areas in which AI research has stalled. Today, we are not even able to seriously estimate how far away our computer systems are from resembling Skynet, and when human-level AI can or will be accomplished. While a number of possible research limitations have been formulated, let’s focus on two of them.
According to IDC, approximately 11 billion devices are connected to the internet around the globe today, with this number expected to increase to 30 billion by 2020, and to 80 billion by 2025. Ten years from now, 152,000 new devices will be connected every minute. The human brain, on the other hand, has approximately 100 billion interconnected neurons with a total power consumption of 20–40W and occupying 1.4 liters of volume. Even though this comparison is rather simplistic, it reveals the huge gap between today’s computer technology and brain technology. In addition, we still don’t fully understand the brain's computational architecture. While our computers are based on the sequential and deterministic cooperation of different subsystems, it seems that the brain's architecture is based on a set of back-looped and often overlapping and competing agents. Assuming this model is valid, emulating the functions of the human brain on computer hardware can probably only bring about limited progress, and greater computational power and memory capacity may not be the key to intelligence.
Another concern is equally serious. Just as our body has many physical limits (for example, we will never be able to run 100 meters in under 5 seconds), the human brain, like any other system, definitely has its own cognitive limits. The question here is whether constructing human-level or above-human-level AI is within or beyond our intellectual skills and capacity. Let’s assume, for example, that we have managed to build a super-intelligent machine and are now waiting for it to produce something super intelligent. In the famous book, The Hitchhiker's Guide to the Galaxy, written by Douglas Adams, the Deep Thought supercomputer, when asked to provide the Answer to the Ultimate Question of Life, The Universe, and Everything, answered “42”, which was a confusing, shocking, and cryptic statement. But what if a super-intelligent machine produces a more serious answer like E=mc2? Einstein's theory of relativity was the most ground-breaking, but at the same time, most misunderstood and disputed concept, which took scientists decades to validate. Will our brains understand super-intelligent statements, concepts, and theories? Or will they be considered absurd and meaningless because our brains don’t comprehend them? And how can our brains construct a super-intelligent machine if its outcomes couldn’t be understood?
If Skynet is not a genuine or impending threat, is there anything else we should be afraid of? Definitely, yes. Our everyday life will become more and more dependent on computer systems. Any malfunction, downtime, or hacker attack could have dire consequences, including the spread of health hazards or even mass loss of life. Moreover, military systems will be no exception, and are actually at the forefront of IoT development, with automated and semi-autonomous weapons systems having already become a reality. Secondly, more data about us will be collected, stored, and analyzed to understand our lifestyle and preferences, as well as to predict our behavior. In other words, Big Brother will be getting bigger and fatter.
+420 221 423 220