Artificial intelligence (AI) has transitioned from a future fantasy to a current working capability. It is expanding faster than it inspires and frightens. With the emergence of agentic AI, generative AI, cognitive AI, affective AI, and many other AI forms, some say the world is moving toward building AI smarter than humans. It draws a direct parallel to the invention of the atomic bomb in its potential for irreversible impact. As governments rush to harness AI efficiency, malicious actors build AI systems that may soon exceed human control. 2024 Nobel Prize winner Geoffrey Hinton, the "Godfather of AI," issued a stark warning in an interview with BBC Radio 4's Today program that aired on December 27, 2024. Hinton revisited his prediction that AI would lead to human extinction in the next 30 years, saying the risk was closer than we think. This article critically examines the regulatory approaches taking shape across jurisdictions, including the European Union, the United States, China, India, and Canada. We also explore the specific incentives behind their legislative activities, which include human rights protection, international competitiveness, geopolitical dynamics, and social concerns regarding emerging technologies. Our analysis reveals that despite these efforts, current national laws and rules still create a fragmented patchwork rather than a cohesive and proactive global framework. This particular fragmentation distinctly highlights not only convergent aims, such as prioritizing public safety, but also divergent fundamental assumptions, particularly concerning the ideal balance between comprehensive state control and promoting rapid market innovation. While countries continue to craft their regulations tailored to local needs, in order to make AI safe for everyone we must establish global AI principles and cross-border rules to ensure consistency, foster trust, and manage risks that transcend national boundaries. By pairing strong local governance with harmonized international standards, we can co-create AI systems that adhere to ethical principles like transparency, accountability, and professional oversight to promote societal well-being. The world needs a "Geneva Convention for AI" to guide this collaborative effort
Each "Communication Corner" essay is self-contained; however, they build on each other. For best results, before reading this essay and doing the exercise, go to the first essay "How an Ugly Duckling Became a Swan," then read each succeeding essay. If you take writing seriously, then you almost certainly recognize the name George Orwell(real name Arthur Blair, 1903--1950). Undoubtedly, Orwell is one of the most widely celebrated authors of the 20th century as both a novelist and essayists. Certainly, two of his most enduring works of fiction are Animal Farm and Nineteen Eighty-Four, both of which were made into major motion pictures. So, when I came across a reference to Orwell's six rules of writing, I simply had to search them out to read them. You should, too. But with a certain degree of skepticism.
GenAI is coming for the comfortable, not the creative minds Generative AI provides organizations with conversation (an always-on, endlessly patient interface for any simple question), distillation (summarizing vast volumes of text), and fabrication (generating the flood of standard documents and compliance material that modern bureaucracies seem to need). Together, these capabilities can automate "bullshit jobs" while raising the bar for what counts as genuinely original work, as companies rethink how they recruit, train, and create value.
AI systems that fail early are visible; AI systems that succeed often fade into the background. In conventional software, once a system has been validated against explicit specifications, it typically settles into stable operation and becomes part of the invisible infrastructure that organizations rely on. Many AI systems do not settle in the same way. Even after successful deployment, their behavior can shift with context, usage, and interpretation, introducing a different class of post-deployment risk. Because AI systems generate outputs from learned statistical patterns rather than fixed rules, early success can encourage forms of reliance that exceed the system's original design intent. Systems that appear to be "working" are reused, extended, and embedded into workflows that carry operational, financial, or safety consequences, even as their behavior continues to evolve. The examples in this article illustrate recurring patterns in how AI systems differ from traditional software after deployment, including authority expansion, reliability drift, and gaps in operational control. Using public incidents, the article argues that reliability after deployment depends less on launch accuracy and more on stewardship. Here, stewardship refers to the operational practices required to keep AI systems reliable and dependable over time: defining clear usage boundaries, continuously evaluating real-world behavior, and embedding governance mechanisms such as monitoring, override controls, and accountability directly into system operation.
Each "Communication Corner" essay is self-contained; however, they build on each other. For best results, before reading this essay and doing the exercise, go to the first essay "How an Ugly Duckling Became a Swan," then read each succeeding essay. Pithy quotations are wonderful things. The very best of them illuminate thoughts and concepts in ways that intrigue and stay in the mind. What could be better than that?
Fault-tolerant quantum computing (FTQC) has the potential to reliably execute complex calculations by overcoming the problems raised from the errors and noise inherent in quantum systems. This paper builds upon a 2025 report issued by the National Academy of Technologies of France. After recalling the principles that underly the so-called "quantum advantage," the use of error-correcting codes in the design of fault-tolerant quantum computers is discussed. Also explored are the five most advanced physical technologies that can be used to build such computers and the challenges of scaling up to a sufficient size to run useful applications. After highlighting the limits of monolithic qubit integration, the focus turns to increasing the number of available physical qubits by interconnecting several quantum processing modules. The paper concludes with a discussion of technical and economic environments, their performance benchmarks, and their future coexistence with other computing technologies and with supercomputers.
This article examines the transformative impact of generative AI on law as a text-centric system. It argues that the computational rendering of legal language through embeddings and large language models promises a double access revolution: lowering barriers for litigants and increasing efficiency for legal professionals and courts. Yet these gains come with acute risks, from hallucinated authorities and skewed outputs to prompt variance, docket overload, epistemic contamination, and professional monoculture. The article advocates a jurisprudence of augmentation in which human legal actors retain non-delegable responsibility for verification, judgment, and accountability.
In this interview, Dr.-Ing. Asif Ali Khan speaks with Ubiquity senior editor Dr. Bushra Anjum about computing-in-memory (CIM), a promising paradigm that enables memory devices to perform computation in addition to their primary functionality of data storage. CIM has the potential to revolutionize the future of computing system design, making systems not only more scalable but also significantly more energy efficient and sustainable. However, there are numerous challenges that must be addressed before this architecture can be widely adopted, which require interdisciplinary efforts across device technology, architecture, system software, and the compiler communities .
The general idea behind machine learning is that, instead of people writing programs to extract information from data, machines extract information by learning from examples. Neural networks are widely used in machine learning, and this paper tries to give an accessible introduction. Currently, machines don't learn like people. A central idea is that things can be associated with lists of numbers, and that similar lists of numbers are associated with similar things. A list of numbers can be considered to be points in a multidimensional data space. The points associated with similar things cluster in multidimension space. Neural networks are very good at separating out the clusters. A major application of machine learning is classification, and this has enormous commercial value, e.g., many problems require a classification between "yes" and "no"---take an action or don't take an action. Any intelligence associated with neural networks comes from (i) the humans who design the network, (ii) the humans who collect and preprocess the training data before it enters a network, and (iii) the humans who postprocess and interpret the numerical outputs of the network. An open problem is that many machine learning systems are black box classifiers and cannot explain their decisions or recommendations. As such, their use may be unethical or even illegal in some jurisdictions. Machine learning is a powerful and valuable technology that can be understood by everyone.
This is the second part of a two-part article. The first surveyed the history of computing with real numbers up to about 2015. After decades of the IEEE standard for floating-point format, the industry is now in a state of flux again with many different formats being proposed, especially for efficient machine learning and inference workloads. The posit format has significant advantages over the IEEE standard, and it may become the new standard for representing real numbers .
We have predicted the emergence of Intelligent Enterprises, which will spawn a digital economy, impact societal structure and behaviors, and mark the beginning of the sixth technology epoch, dominated by the spread of cognitive systems---AI-augmented tools and systems.
Cryptocurrencies have given rise to extreme excitement and controversy. There is a continuing debate on whether they will flourish or disappear. This essay argues that an even more interesting and important issue involves the implications of the cryptocurrency mania for the evolution of our society. What matters about cryptocurrencies and related subjects, such as blockchain, is not really their technology. That technology is neither novel nor revolutionary. The key factors are trust and crowd psychology, which are driving us toward the post-truth world in which groupthink helps create "alternate realities." The cryptocurrency scene provides insight into these developments, which are consistent with those in many related areas, such as the growth of disinformation, deepfakes, and AI.
The ability of a machine to store numbers and perform calculations on them is fundamental to computing. Many novice and expert users wrongly assume that number representation is a long-solved problem. Modern computers use binary encodings, and nothing more needs to be said, right? Wrong! For example, irrational numbers like pi are approximated; rational numbers like 2.0 may overflow when multiplied by a larger number; and (1/x)*x should equal 1.0 but often does not! Modern floating point evolved rather than being designed! John Gustafson has studied this and related problems of computing for many decades. He is the inventor of posits---a superior way to represent numbers in binary. In this two-part article, Gustafson surveys the history of machine number representations and illustrates by example how even the most modern number representations used today often fail. The IEEE 754 floating point standard is on shaky grounds, lacking a sound basis, and yet we persist in using it. How did we get here, and where might it lead? This is the fascinating story of computer numbers. --- Ted G. Lewis , Ubiquity senior editor
The case study of driverless vehicles shows that AI has limitations. Understanding this can lead to successful systems, while not understanding it has led to expensive failures. Despite billions of dollars invested over many decades, no vehicle entirely controlled by AI can operate on public roads. Success comes from an appropriate combination of AI doing what it does best and humans doing what they do best. The examples of a grocery delivery system and driverless taxis show that viable systems can be produced when vehicles drive partly using AI but request remote human assistance when the AI cannot cope. Such successes can give the public the impression that the AI can solve much harder problems than it can---to operate in the highly complex real world AI systems need humans in the loop. There are various aids for drivers such as AI controlling cars on freeways, but these all require the driver to take back control when instructed. Other lessons for engineering AI system include: the level of autonomy for intelligent machines and systems should be well defined; a systems approach that codesigns AI systems with their operating environment can lead to successful systems capable of incremental improvement; new and unanticipated problems can emerge on applying AI; the highest standards of conventional software engineering are required for robust and safe AI applications; driverless vehicles challenge the idea that regulation stifles innovation, and provide examples of companies putting profit before safety---regulation is essential for safe applications of AI .
Each "Communication Corner" essay is self-contained; however, they build on each other. For best results, before reading this essay and doing the exercise, go to the first essay "How an Ugly Duckling Became a Swan", then read each succeeding essay. In a song from the classic musical comedy "My Fair Lady," Eliza Doolittle storms "Words! Words! Words! I'm so sick of words!" She isn't wrong. Used correctly, words shed light and generate interest and understanding. Used incorrectly, they confuse and generate mistrust and even animosity This is particularly true in the teaching of math and science in primary education (kindergarten-12th grade). But it doesn't need to be this way. In fact, it can be reversed.
