An industrial revolution unfolds as a series of waves of creative destruction. An emerging wave is supposed to offer better alternatives to existing products, often at less cost. New technology cores power such creative waves. For example, a solid-state image sensor fueled the wave of creative digital camera Innovation, causing destruction to film-based cameras. Similarly, a multi-touch user interface-based smartphone caused destruction to previous designs around a physical keyboard and/or stylus. Like all other industrial revolutions, the Fourth Industrial Revolution also needs a technology core to fuel waves of innovations. Artificial intelligence technology is looked upon as the technology core. Is this technology strong enough to unfold the next Waves of Innovation? Or, is there a possibility that the underlying weakness of AI impedes the fourth industrial revolution?
However, this technology offers the potential to add human-like intelligence to physical machines, terming them as cyber-physical systems. It’s envisioned that these cyber-physical systems will operate by themselves, making our productive activities free from human touch. Subsequently, we will be able to offer a better alternative to many goods and services at a lower cost. Consequentially, we will move to another phase of industrial revolutions for offering a higher quality of living standards.
Creative waves—what it takes for AI to power the fourth industrial revolution?
The role of technology to delegate roles from humans to machines is not new. It started millions of years ago. Generating ideas for creating and recreating products and systems to get jobs done better at less cost is the inherent characteristic of human beings. In ancient philosophical writings, it appeared as Praxis. In the long journey of creating better means, human beings developed water wheels and sailboats. The march continued with the invention of the steam engine. Subsequently, this steam engine fueled waves of innovation, causing creative destruction to a number of products and industries. Hence, we call it the first industrial revolution.
The invention of the internal combustion engine, light bulbs, electric motors, job division, and production lines, among other technologies, led to the formation of a set of innovation waves. These waves caused destruction to products, firms, and industries created during the first industrial revolution. This invention of the Transistor in 1947 led to the starting of the 3rd industrial revolution. During the 3rd industrial revolution, we have better alternatives like compact computers, cellular communication, digital cameras, and also robots working in the industry.
At the dawn of the 21st century, we started envisioning the emergence of the fourth industrial revolution. Unlike the past, we have not witnessed the invention of any new technology core, though. However, we are anticipating that our progress in sensor, computing, wireless connectivity, and software, and their fusion, will lead to a new technology core. We call it artificial intelligence. We are under the impression that this technology will allow us to add human-like sensing, perception, and resining capability to physical machines. Subsequently, machines developed during the 2nd and 3rd industrial revolutions will operate without having a human in the loop, and they will serve purposes better. Let’s look into two examples.
Lesson from autonomous vehicles–current state of AI is not strong enough
Every industrial revolution witnessed iconic innovation waves. For the fourth industrial revolution, an autonomous vehicle is one of them. For a long time, human beings are dreaming to have autonomous vehicles. The journey started in the 1920s. The past three attempts failed to offer us a dream machine. At the dawn of the 21st century, the dream remerged. The idea is to add sensors to automobiles for imitating human-like intelligence. These sensors will keep providing data about the surroundings. The software will be processing and figuring out the situation and safely driving it without the need for a human driver in the loop. Moreover, low latency wireless connectivity will allow the vehicle to remain in sync with other neighboring vehicles. Hence, there will be highly coordinated movements of vehicles, making driving safer, and improving the throughput of road networks.
So far, more than a dozen demonstration vehicles have already shown up. They are equipped with more than a dozen diverse sensors like ultrasonic, camera, RADAR, and LIDAR. High-performance onboard software works in cooperation with cloud-based learning modules. Millions of images and data are fed to neural network-based learning algorithms to learn how to understand situations and decide about safe driving action.
In the beginning, there was rapid progress in learning. But the progress kept saturating before reaching the target—making autonomous modules better than a human driver. On the one hand, human drivers deal with millions of variations. Nobody knows exactly the detailed science. On the other hand, the memorization-based technique through weight adjustment of the underlying neural net-based learning machine starts oscillating after reaching some maturity. Subsequently, after the investment of more than $80 billion in R&D, autonomous vehicles have basically stalled, before rolling out. Such reality underscores the observation that the weakness of AI impedes the fourth industrial revolution
Lesson from ASIMO–weak AI ended the life of ASIMO before rolling out
The 2nd big hope has been in Humanoid robots. Over the last more than 2000 years, human beings have been after the dream of recreating themselves to get many purposes served better than ever before. Along with Robot Sophia, Honda’s ASIMO created hope of realizing the dream. ASIMO is the technologically most advanced humanoid robot ever developed. Honda started R&D in the 1980s to develop a human-like machine to offer service. One of the target sectors was the elderly care market. Over the decades, Honda’s R&D team made successes in nurturing ASIMO with strong mechanical capability. It was able to walk, run, stand on a single foot, and even could kick a soccer ball. It has numerous sensors, including a pair of cameras.
Unfortunately, Honda’s team could not develop adequate human-like Innate abilities in ASIMO. Hence, ASIMO could not qualify for the elderly are jobs. Subsequently, Honda management stopped further R&D on ASIMO, after 32 years and spending of $500 million, in 2018. The premature end of ASIMO also indicates that weak AI impedes the fourth industrial revolution.
Review of AI–Memorisation based AI technique is a core weakness
There are five basic capacities we need to imitate human-like cognitive capabilities in machines. The first one is the ability to memorize, retrieve, and communicate data. Well, technology appears to be better than human beings. The ability to sense the environment is the 2nd capacity. Our progress in developing diverse sensors like cameras, radar, and LIDAR is already quite high. The 3rd capacity is to extract information from sensory data and interpret the situation. Despite progress, we have a high-level deficiency. The next challenge is predicting the likely movement or dynamics of objects like humans and other living and non-living objects in the scene. The fifth one is deciding about having a safe, effective, and efficient move. Despite the success of computers in outperforming the world’s best chase player, our capability to build the 4th and 5th capacities in machines is highly primitive.
Some of the underlying technologies that we use are i. search and optimization, ii. logic, iii. probabilistic methods for uncertain reasoning, iv. Classifiers and statistical learning methods and v. Artificial neural networks. Of course, they help. But they are highly incapable in predicting the likely movement of neighboring moving objects and human beings.
Moreover, their behavior keeps changing. For example, once the rain starts falling, the behaviors of pedestrians rapidly change. Although the neural network-based deep learning algorithms help us, the weight-based memorization technology of connectionistic network does not allow us to develop a set of mathematical equations to imagine the unseen future. On the other hand, a 5-year old upon recognizing a horse attains the capability of simulating all sizes of horses. As a result, he or she does not require further training to recognize horses of all sizes and shapes—whether cub, skinny or fatty. Hence, existing AI technology is extremely weak.
Robot hands prevent taking over jobs, and also the unfolding of fourth industrial revolution
Another weakest point of developing a human-like machine is robot’s hand. In order to perform different tasks at home or other service delivery centers, a robot service provider should be able to handle multiple types of objects. Some of them will be soft, hard, deformable, fragile, solid, liquid, cold, or hot. So fa, our progress in developing robot hands over the last 300 years in making them suitable for delivering services is extremely poor. Therefore, in addition to AI, primitive robot hands will prevent the fueling of many envisioned innovations to mark the unfolding of the fourth industrial revolution.
As opposed to codified capability, human beings’ innate abilities are a major barrier to the realization of human free productive activities. In fact, exiting AI technology is not strong enough to imitate any of the 52 innate abilities of human beings. We apply multiple of them subconsciously in performing even a simple task. Hence, it does not sound unfair that the underlying weakness of the current state of AI impedes, or perhaps will stall, the unfolding of the fourth industrial revolution.