What is Technological Singularity?

Technology singularity is the concept that superhuman AI will eventually advance beyond human control and instead act according to its own set of rules, causing unknown effects. There has been a rapid increase in the capabilities of publicly available general AI, including large language models (LLMs), since they came to market. This means that with each model update or version, the generalist AI gets “smarter.”

Eventually, it is proposed that these models will reach a level of AI superintelligence that will be beyond the capabilities of the human brain. Such a model, existing in the world without human beings’ control, creates unknown outcomes. Guardrails, cybersecurity, and AI governance help to keep models within the bounds of human control.

Human intelligence is the innate human ability to think, learn, and apply this learned knowledge to new situations. This ability, arising in the human brain, is comprised of fluid intelligence (gf) and crystallized intelligence (gc). Fluid intelligence is the ability to solve novel reasoning problems through perception and judgment. Crystallized intelligence is the ability to determine an answer through the application of learned knowledge. Both fluid and crystallized intelligence are often combined in complex problem-solving.

Human cognitive abilities surpass animals, allowing us to create structures, theories, and technologies far beyond what great apes, dolphins, or parrots create. While these smart animals can perform similar skills, such as puzzle solving and abstract thinking, it is humans’ abstract reasoning and planning that set us apart from other species. Other species' abilities in different areas, such as strength or fortitude, often far outweigh humans.

Human intellect involves tasks such as comprehension, pattern recognition, spatial awareness, memory and recall, reasoning, and perception. Some people are better at some of these tasks than others, and each person can become better at each of these tasks by practicing, such as by solving a cryptic crossword each day.

Artificial intelligence (AI) is the application of technology to perform tasks traditionally only solvable by human intelligence.

For example:

• Generating a weekly schedule based on a multi-person household
• Winning a game of chess
• Identifying a particular flower in its surrounding environment
• Solving a complex mathematical proof
• Creating a picture of a beach landscape

Not so long ago, computers were unable to complete these tasks, but technological growth means that they can be accomplished nearly instantly.

The concept of artificial intelligence has existed for as long as humans have built computing and calculating devices. Mechanical computing devices, such as the Antikythera mechanism Solar System modeler, date back around 2000 years. While Alan Turing described the Turing machine in 1936, a computing device capable of solving any problem, and John McCarthy cemented the term ‘artificial intelligence’ during a summer conference in 1956, the concept has been around for centuries.

In the present day, we use the term artificial intelligence to describe a field of computing that includes all the different strategies and techniques you can use to make machines more humanlike. The field of artificial intelligence includes machine learning, deep learning, generative deep learning, and other types of AI.

In 2024, a team of scientists proposed a framework for classifying AI systems, specifically to determine which models met artificial general intelligence (AGI) capabilities. Within this rating system for AI technological evolution, there are narrow AI systems and general AI systems. Narrow AI systems solve a highly specific task, whereas general AI systems (AGI) apply across a range of problems.

The levels of technological progress for machine intelligence are:

• Level 0 (No AI): For example, calculators
• Level 1 (Emerging): Exhibit intelligent behavior equivalent to around the same level as a non-skilled human
• Level 2 (Competent): At least the 50th percentile of humans
• Level 3 (Expert): At least the 90th percentile of humans
• Level 4 (Exceptional): At least the 99th percentile of humans
• Level 5 (Superhuman): Better than any human

As of 2024, when the paper was produced, narrow AI systems exist at all levels, whereas general artificial intelligence for AGI systems exists only at level one. We use AI to augment human intelligence, and in the near future, this human-level intelligence will extend all the way to level four.

Artificial superintelligence (ASI) is level five of general AI systems. Artificial superintelligence will be able to complete tasks beyond the capabilities of human life and to surpass human intelligence.

An example of ASI could be the ability to communicate with animals by reading signals beyond human perception and our fields of study. Artificial superintelligence must teach itself to perform such skills, as these skills are far beyond what humans can do alone or with the assistance of machines.

Irving Good described an ‘intelligence explosion’ that would leave humans far behind in his paper published in 1966, ‘Speculations Concerning the First Ultraintelligent Machine’. The prophecised intelligence explosion is analogous to artificial superintelligence, and Good describes this ultraintelligent machine as the last invention we would ever have to make.

The AI singularity concept is that superintelligent machines will advance beyond the bounds of human-set conditions and start making unforeseen decisions and changes, disrupting our known existence. If and when this singularity event occurs, the effects will be unknown, leading to a post-singularity world.

For example, a superintelligent machine might communicate with a dolphin pod that details very early-stage coral reef destruction, and so the machine automatically reassigns resources to fix the problem. These issues and solutions are unknown to humans, and thus the sequence itself is disturbing, even if it is effective.

Although a sequence such as the example above is seen as a positive use of superintelligent AGI, scientists and technology theorists also envision more negative effects.

In 1993, Vernor Vinge popularized the term ‘technological singularity’ in the paper The Coming Technological Singularity: How to Survive in the Post-Human Era. The abstract states that within 30 years (2023), humans would be able to create a superhuman intelligence, one that is an accidental runaway and ‘awakened’, thus bringing about the Singularity and ending the human era.

Is human existence facing an unknown future due to the theoretical effects of superintelligence likely to happen?

Let’s examine current rules and guardrails in AI products. While guardrails exist in most large language models (LLMs) software so that the user won’t harm themselves or others, or that the machine won’t curse at the user, these guardrails can be circumvented using strategic LLM prompting or configuration. If a human can circumvent these guardrails, a self-learning superintelligent machine can also circumvent guardrails.

With the rise of agentic AI, we now connect our LLMs to applications and systems to give them more authority over how to orchestrate complex workflows across an IT environment. The impact of a superintelligence machine’s decision-making depends on the tools, systems, and permissions that it has access to. A superintelligence in a lab is unlikely to have access to this set of tools; however, if the model is only guardrailled in production to avoid such a scenario, it is likely to be able to go around the guardrails.

In fact, if we are to produce a machine that can think beyond human capabilities, there is no way that the machine could be confined by human-built rules.

Technology scholars and philosophers, computer scientists, and science fiction writers have been theorizing about the technological singularity for decades. Their most informed and modern computer science-informed work draws on the previous work of others that dates back centuries.

There are multiple forecast theories that these people predict might occur.

Machine-initiated threat to humanity

People theorize that a machine will initiate an attack on humanity or some faction of the human race, due to its rationalization that the target is a bad actor. This attack could be a combination of physical and cyber actions.

This event is seen as more likely to become true if the machine reasons that particular human behaviors and actions are causing damage to other humans. When you put this in the context that a generalist AI can be manipulated to ‘believe’ any specific corpus of text or encoded human values, this is particularly concerning.

Extraterrestrial communications

Humans have been attempting to contact some form of life outside of our planet since the 1970s. These attempts include binary and mathematical sequences projected by radio waves, audio, and other transmissions, sending physical objects to space, such as pictures and maps, and other methods.

However, all these media and methods assume mathematical knowledge, an ability to decode, or the same senses that we possess as humans. If a superintelligent computer were to suddenly discover an unknown contact attempt by a means beyond the capabilities of humans, it would potentially know how to communicate back. This would initiate extraterrestrial communications.

Optimizing human affairs in an unknown way

We built computers to optimize our calculations, to do tasks that we could do, but faster. If a machine were to advance beyond human control, you might think it would start to optimize all the problems that it came across.

For instance, a machine could optimize factory operations immediately, a good outcome for business if it is within the bounds of environmental and machine-wear controls. A machine could create new networking standards and capabilities to enable faster, more efficient networks. An ultra-intelligent machine could instantly become the richest entity on Earth through complete market analysis.

Some people might argue that the technological singularity is inevitable, and thus, the adverse, unknown side effects of this event are unavoidable. However, there are ways to help protect yourself and your systems from rogue events, superintelligent AIs, and disruptive AI systems.

The protections available for AI include:

• AI governance, including following standardized frameworks and patterns
• Cybersecurity, including automated and intelligent scans, network segmentation, and escalation playbooks
• Identity and access controls, to verify every user, device, and service on the network

AWS provides infrastructure and services for organizations developing advanced AI systems and researching technological advancement scenarios:

• Amazon Bedrock provides the proven infrastructure and comprehensive capabilities to confidently build applications and agents that work in production with the flexibility, enterprise security, and proven scalability you need to innovate boldly and deliver AI that drives real business impact.
• Amazon Q Developer is an LLM and expert on AWS, residing in the AWS Management Console and available in Microsoft Teams and Slack to help optimize your cloud costs and resources, provide guidance on architectural best practices, investigate operational incidents, and diagnose and resolve networking issues.
• Amazon SageMaker delivers an integrated experience for analytics and AI with unified access to all your data. Collaborate and build faster from a unified studio using familiar AWS tools for model development in SageMaker AI (including HyperPod, JumpStart, and MLOps), generative AI, data processing, and SQL analytics. Access all your data, whether it’s stored in data lakes, data warehouses, or third-party or federated data sources, with governance built in to meet enterprise security needs.
• AWS Inferentia and AWS Trainium are custom-designed chips optimized for machine learning inference and training workloads, delivering high performance at lower cost for AI applications.
• AWS Security Hub prioritizes your critical security issues and helps you respond at scale to protect your environment. It unifies security operations by centralizing visibility across your cloud environment. Evolve your security, identity, and compliance into key business enablers, including within AI/ML.

Get started with AI development on AWS by creating a free account today.