Advanced computing resources and connected technologies are converging in a range of automated applications, including robots, drones, and algorithm-driven applications. What could possibly go wrong?
It’s hard to read a newspaper or magazine and not find at least one article about artificial intelligence (AI) or robotics. Machine learning, deep learning, neural networks, analytics, and big data have entered our daily lexicon. Robotics and AI were front and center at CES 2020. Dire predictions of millions of workers losing their jobs within the next 10 years to intelligent robots abound. Robots have been chipping away at production jobs for years, but even employees in other professions are likely to be in jeopardy. AI is being integrated into advanced manufacturing equipment as well as in applications designed to optimize transportation, healthcare, finances, medical research, marketing, and customer service. Agricultural weed-spraying equipment being developed by John Deere uses AI and computer vision to identify weeds among crops. With this data, herbicide can be selectively applied only to the weeds. Little Caesars has patented a pizza-making robot. AI has already been used to manipulate social media, which has become a major influence in our culture. Bob Lord, senior vice president of IBM’s cognitive applications and developer ecosystems division, expects AI to go mainstream in 2020. It is entirely possible that AI applications will become the new normal but getting there will not be a straight line.
As complexity increases, so does the likelihood of failure and unintended consequences. Children might order a truckload of LEGOs from their Amazon Alexa, but there are more serious concerns too. Sophia, a humanoid robot known as the world’s first “robot citizen,” was asked if she would destroy humans. She said, “OK. I will destroy humans.” The potential for autonomous military weapons that use AI to determine friend or foe also presents concerns. Many experts have warned of the danger unconstrained intelligent robots pose to human civilizations.
The computing resources required by AI exhibit all of the elements of a complex system, including many highly interconnected and interdependent parts. There is an entire science of modeling complex systems, but it is unclear if these tools are adequate to fully address the challenge of robotic systems driven by AI.
Algorithms at their most basic level are pretty straightforward. They consist of a set of rules or processes required to solve a problem. Artificial intelligence uses a set of algorithms to make decisions. There have been numerous articles about how complex algorithms have become and how they inevitably contain errors as well as the bias of their human creators. Software engineers have admitted that algorithms written by a team may not be fully understood by anyone.
A neural network consists of a series of algorithms that are designed to identify underlying relationships within a large dataset using a process similar to the human brain. Input of accurate data allows manipulation to achieve a definitive answer. The potential for errors in the code, algorithms that produce unintended results, or simply the input of inaccurate data can cause unexpected or inaccurate results.
AI presents plenty of opportunities to be abused with malware and attacks that can impact businesses and society. An autonomous car that runs a series of red lights or hits a pedestrian is an example of immature AI. Multiply that by the millions of applications that are anticipated to become mainstream over the next 10 years and you have a recipe for disaster. Input of corrupted data resulting from a defective sensor or intermittent connector contact will result in inaccurate output. That is where the integrity of the circuits, sensors, and connectors impacts the performance of robotic and AI-enabled devices.
Robots using AI are becoming adept at new skills simply by completing a step-by-step learn-by-doing process. Machine learning refers to systems that can learn by themselves. Systems now available use deep learning to provide real-time voice recognition and language translation. This deep learning process requires the analysis of potentially thousands of variables to identify the most favorable solution, be it low cost, highest quality, fastest assembly, optimization of materials, etc.
Understanding all the possibilities of how a robot that utilizes AI will respond is part of ensuring that it will achieve its intended objectives, but another aspect is much more basic. Robots of all types are constructed with a wide variety of sensors, cables, and connectors, all of which must perform to stated specifications. Amazon recently demonstrated its new delivery drone, Prime Air. Studded with an array of sensors, this drone utilizes machine learning to enable its “sense and avoid” technology. Failure of one or more of these sensors could cause the generation of inaccurate data or complete failure. Nobody wants delivery drones dropping from the sky. With the dramatic increase in robots designed to function closely with humans at work and home, erratic behavior could become a human safety issue. The computing and communication infrastructure that will be required to support AI will demand exceptional reliability as well as increased speed.
In addition to robotics, AI applications in such fields as medical, process automation, and security will require real-time response. It is becoming more common to locate AI computing resources at the edge or integrated directly into a sensor. Rather than hosting AI applications in the cloud, it might make sense to offload them directly to the equipment, minimizing the transfer of sensitive data to the cloud.
Historically, connectors have been identified as a potential source of failure. An open or intermittent contact can cause havoc in a complex electronic system. Manufacturers of connectors and sensors will be pressed to develop “fail safe” and low-cost interconnects that feature a long service life in harsh environments.
We are entering an exciting new world where an increasing number of critical decisions are being made by machines powered by artificial intelligence. Connectors and electronic sensors are playing a key role in ensuring that those decisions meet the expectations of applications that range from managing a coffee maker to executing a precision military airstrike.