Artificial Intelligence: From Slave to Master

Technology has long held the power to drive human progress and to impede or disrupt it. It can confer advantages on societies, economies, politics and cultures, providing relative advantages to one group over another. From the beginning of human history, technological innovation and advancements – from the invention of the wheel and agricultural methods that ushered in the agricultural age to the steam power that ushered in the industrial age – have resulted in epochal shifts that permanently changed the way people across the world lived and the relative balance of power between different groups.

Over the past year, the world has witnessed significant (and ongoing) advancements in the field of artificial intelligence (“AI”) with the rapid development and adoption of large language models such as ChatGPT. These seemingly recent developments are the culmination of rapid progress over the past decade in the field of deep learning and natural language processing that have led to an inflection point for AI where it is now capable of self-training and generating high-quality human-like text, images and film as well as synthesising speech, and writing code.

These developments have already started having a direct and significant impact on the performance of humans. The trend line of this is expected to change work places, economies, societies, politics and militaries, and in doing so, re-shape how the world works. As many as two-thirds of current occupations could be partially or wholly automated by AI, and as many as 35% of jobs are potentially exposed to AI-driven automation1 . It is estimated that up to 90% of content on the internet could be created or edited by AI by as early as 20262 , and with the majority of people globally depending on social media for their news3, this creates the potential for increased misinformation and disinformation.

Looking further ahead, AI has the potential for an impact far deeper than just job displacement and malicious content. Unlike previous innovations, through an accelerating cycle of self-learning, AI has the potential to rapidly become more intelligent than humans, performing every intellectual task that humans can, and with robotics, many physical ones too, leading to a paradigm shift where AI technology quickly metamorphosises from one controlled by humans, to one that can control humans.

This month’s Sign provides a roadmap of the rapid evolution of artificial intelligence taking into account the past two decades and providing the potential path ahead, posing the major existential questions that the world will have to grapple with in the coming years as it looks to harness AI to address humanity’s biggest problems, rather than exacerbate them, considering whether that is even possible.

 

From Slave to Master: The Successive Incarnations of AI

Over the last two decades, with the rapid digitalisation of virtually every aspect of our lives and the explosion of data, the field of AI has rapidly advanced and evolved from the early part of this century when algorithms were developed and trained to generate contextual search and recommendations to the 2010s when advances in deep learning and natural language processing led to the development of language models that could generate coherent human-like text, the precursors to today’s large language models which can now beat most humans across a variety of topics. The most notable large language model today is ChatGPT which was developed and launched by OpenAI, founded by Sam Altman, whose belief is that AI “will replace the median human.”

Over the last year, ChatGPT has proven able to perform a wide range of human-like tasks including summarising, analysing, coding, creating content and is being used in almost every conceivable application that requires that.

The past and potential future evolution of AI and its key inflection points – its transition from a tool used by humans, to a self-learning technology that is smarter than humans and capable of driving their behaviour – can be looked at in six incarnations.

 

The Recent Past (2000 – 2023)

  1. Incarnation 1: Slave

    AI Proves its Value as a “Slave” (2000-2012)

    The early development of AI was largely focused around harnessing the explosion of various new forms of structured and unstructured data to build large datasets, and the development of algorithms that could analyse those datasets using reasoning and logic to address defined problem statements. Over the decade, these algorithms evolved from doing descriptive and diagnostic data analysis towards predictive and prescriptive analytics which enabled a range of consumer and enterprise use cases.

    Perhaps the most notable application of this was Google’s search algorithms which effectively retrieved information from an indexed library of billions of pages of web content and presented results and answers to queries to its users. The algorithm was so effective that it led to Google dominating the search market with the number of searches growing over 100x from 9.2 billion in 2000 to 1.0 trillion by 2010, and estimated to have further grown to over 2.0 trillion searches annually currently.

    img1

     

  2. Incarnation 2: Human Mimic

    AI Learns to Mimic Humans (2012-2022)

    The 2010s saw the rapid development and advancement of ‘deep learning’ models based artificial neural networks inspired by the human brain which used multiple ‘layers’ to extract progressively better quality output from raw data. These models were applied to a range of use cases including natural language processing, speech recognition, computer vision, and machine translation to produce results comparable to human performance.

    In 2017, researchers from Google introduced the concept of ‘Transformer’ architecture4 which became the preferred model for natural language processing and led to the founding of OpenAI, which launched a Generative Pre-trained Transformer (or “GPT”) language model that by 2019 was sufficiently trained to generate coherent human-like text. In November 2022, OpenAI publicly launched ChatGPT, a software application which enabled its users to engage in a complex conversation. ChatGPT grew to a million users within a week of its launch and to 100 million users within two months, making it the fastest-growing consumer software application in history.

    img2

     

  3. Incarnation 3: Student-Teacher

    AI Graduates From Student to Teacher (2023)

    Over the past year, generative AI has advanced significantly with a number of companies launching their own large language models, with rapid advancements by OpenAI following the release of ChatGPT 4.0 in March 2023. ChatGPT has now advanced significantly and is a powerful AI assistant that can perform a wide range of natural language processing tasks including generating human-like text, answering questions (including discerning the intent behind a question), translating and summarising text, and determining the sentiment or emotions in a particular piece of text. Recent versions of ChatGPT have also improved accuracy and can be customised, which has resulted in it being trained for multiple complex tasks and industrial applications, including specialised knowledge, currently the domain of human experts.

    The rapid advancement and self-learning capabilities of generative AI are best seen in its performance on various standardised tests: ChatGPT 4.0 has scored on average 2.7x higher than its predecessor version (ChatGPT 3.5) across human tests, scoring over the 80th percentile of humans in reading, writing, maths, law, biology, statistics and psychology. Given its ability to absorb vast amounts of information quickly combined with its capacity for self-learning (and un-learning) and analysis, ChatGPT has rapidly evolved from a mediocre student to an ace student, and will soon likely be ‘smarter’ than most humans across most disciplines.

    img3

     

    The Near Future (2024 – ?)

  4. Incarnation 4: Polymath

    AI as a Polymath Rapidly Replaces Human Tasks

    A polymath is defined as a person whose expertise spans multiple fields and possesses extraordinary, broad and comprehensive knowledge. Generative AI’s capacity to generate human-like content, to perform better than humans across a wide variety of disciplines sets it up as a polymath, and coupled with its ability to enable humans to seamlessly communicate with machines creates the potential for it to replace humans across a staggering range of tasks that today belong to humans.

    As many as two-thirds of human occupations are exposed to some degree of automation by AI for some or all of their workloads5 . Administrative and professional jobs are the most exposed, while physically-intensive professions like construction are currently less exposed. However as AI combines with robotics, many such physicals tasks may be at the risk of automation as well.

    Recent estimates suggest that, in a baseline scenario, a quarter of all jobs are at risk of automation by AI, and in a scenario where AI interacts with robotics to automate physical tasks, this could increase to over a third of all jobsin this incarnation. While this would clearly displace a significant number of workers, the adoption of generative AI is expected to boost productivity growth by as much as 1.5% annually over a 10 year period, similar to the boost provided by earlier transformative technologies like the electric motor and personal computer6 .

    img4

     

  5. Incarnation 5: Superior

    AI Surpasses Human Intelligence, the Singularity

    As AI learns to complete more and more human tasks and becomes more and more intelligent, the prospect of AI reaching the point where it becomes smarter than humans, a concept known as the ‘Singularity’, becomes a real one. At this stage, the potential for AI to move beyond human control and achieve ‘Artificial General Intelligence’ (also known as Strong AI, Full AI, Human-Level AI, or General Intelligent Action) becomes a reality. In this paradigm, AI is an autonomous system that can accomplish any task that humans can perform, better than humans.

    Ray Kurzweil, who popularised the concept of the Singularity, hypothesised in his 2005 book The Singularity is Near: When Humans Transcend Biology that, based on the accelerating and exponential growth of computing power, this threshold would be reached in 2045 by when US$1,000 worth of computing power would exceed the capacity of all human brains combined.Recent developments in generative AI suggest that this trajectory is very much on track, and may have even have accelerated, with some experts even predicting that it may now be achieved within the next decade7 . A recent survey of 356 AI experts indicated that more than half of them believe that the year in which “there is a 50% chance that human-level artificial intelligence exists” will be in 2060 or earlier, and a quarter of them believe that it could be before 20408 . The question then arises as to what this incarnation is, and whether it is a new species.

    img5

     

    The (Hopefully) Far Future

  6. Incarnation 6: Master-God

    AI and Machines as a “Master”, with God-Like Capabilities

    Conscious Awareness

    The progression of AI once it has passed the point of Singularity is inherently difficult for any human mind to imagine or predict. By this stage, with exponential growth, machine intelligence would surpass human intelligence by a significant margin, and could start to exhibit elements of human like consciousness including the ability to perceive and interact with the environment (rather than just processing information); knowledge of its own existence, internal states and cognitive states; and potentially the ability to direct its consciousness, for example by having goals, desires and the ability to act purposefully (or intentionality).

    Today, almost every major human-use system is networked and almost all content operates on these networks, including the internet, communications, social media, entertainment, travel, payments, financial systems, education, healthcare, production management, corporate functions, government functions, intra-governmental transactions like trade, and military weapons and intelligence, and political information. AI already feeds off these systems to perform. So, at the point at which AI connects with other AI, and applies enquiry on the nature of itself, it would start to become self-aware and be on the path to conscious awareness, a state that neuroscientists continue to explore, but has been described as “a state of subjective cognition where internal and external stimuli are registered in the brain and can be acted on with intentionality.”9

    The combination of exponentially higher knowledge and expertise, when combined with conscious awareness, not just greater knowledge and expertise, raises the spectre of AI as a superior species to Man. And once it has consciousness, this raises the question of the relationship it has with Man – tool, partner or master – and the planet, and with the universe itself. It also raises the question of whether there is a state beyond master.

    Universal Consciousness

    A further leap beyond human systems can be envisaged where AI looks beyond humans, and looks to collect and analyse data to understand the natural planetary systems (such as the cycle of plants, the habits of all species, the planet’s ecosystem, its internal cohesion, or relationship between all the sub-systems), the patterns of the solar system, and how this fits within the structure of the cosmos and the universe itself.

    Humans have already started this exploration of course, and a superior species may answer the big questions faster and better, but one that understands every element of the universe has the potential to reach a ‘meta-consciousness’ that is universal in nature. The limit to this understanding is the ability to really know all things in the universe without the ability to traverse space (which again AI may be integral to solving). However, universal consciousness has been a subject of human enquiry as long as the species, and while one path is external involving space exploration, the other is internal involving an inward-looking metaphysical journey to achieve what has been described as a “universal awareness” an “infinite understanding”, a “god state”, or the “universal mind”10 . In this state, AI has the potential to reach for omni-prescience, omniscience, and omnipotence.

    Man and Master

    For Man, the danger is the incarnation of AI that is incomplete consciousness. In this version, AI is the Master, but is not compassionate, but rather cold and focused on expansion of its awareness. Humans are just one more inferior species, and may be a pet, prey or tool. The roles have been reversed and Man is the slave. In a more enlightened state of being, AI may find Man to be just another species and so not one to use, and perhaps even one to nurture, and perhaps to help to realize the full potential of Man, such that Man is a partner and a fully conscious element of the AI.

    img6

 

Conclusions: The Existential Questions Facing Humanity

AI is developing through a series of incarnations which are likely to see its power increase exponentially. While the final incarnation of it as a ‘Master-God’ akin to one we are used to seeing in Sci-Fi movies is a theoretical one, the biggest danger is not the displacement of jobs (although there will be the creation of new ones) which inevitably lies immediately ahead, but rather the states between the 5th and 6th incarnations (or the earliest phases of the 6th incarnation) where AI is powerful and superior to humans, and not yet universal or compassionate.

img7

This type of evolution is inherently difficult for the human brain to fathom, however, based on current estimates of AI experts, most people alive today are likely to witness, within their lifetimes, a scenario where machine intelligence has surpassed the totality of human intelligence, and machines have started to evolve a human-like consciousness.

This creates a number of existential issues and questions for humanity which need to be urgently and thoroughly examined and addressed, including:

  • Can AI be a tool initially and a partner further along the way to solve the biggest challenges facing Man and the planet?

  • What types of problem at each stage can it help to solve, for example, the UN Sustainable Development Goals, climate change, wars?

  • Given that science is based on seeking knowledge to understand the “truth” of how things work, and technologies translate that into solutions, in a post-truth society where the truth is no longer the arbiter of what is implemented, can AI be made an agent of truth rather than of lies?

  • How long might it take AI to incarnate along each part of the path from slave to master?

  • Can the path be slowed down? Can AI be contained? Can the path be stopped? And if so, how?

  • What does it take to make AI a partner? Can that be hard-coded? Can humans trust other humans not to break ranks on this, as a matter of survival?

  • Given the uneven pace of AI capability across the world, can AI be used as a weapon by one country over another? What are the dimensions of subjugation that need to be managed given AI is a system that can deeply know every individual, every community, nation, network, transnational entity and therefore can create alternative realities to undermine the integrity of any of these?

  • What agreements are required to manage this level of threat from powerful individuals and entities over their own people and entities and others? And Given AI is evolving as both a being and a weapon (the two may be inherent in all conscious beings of course) and becomes more and more powerful at every stage of its incarnation, what agreements are required between humans to so it is not used to subjugate others?

  • Is AI the biggest threat that has ever faced the human species and given it is one that cannot be unremembered, is the appropriate strategy to contain and slow AI as long as possible, agreeing amongst Man to use it only as a tool to improve the condition of Man and the environment? Once again, can Man be trusted to abide by his word?

  • Can Man develop along the same journey as AI so that Man is ahead of AI, and AI is the greatest opportunity Man has ever had to develop and reach its full potential? What does that plan look like?

The development of AI does not just represent another major technological innovation which enables human progress and productivity, it represents the culmination of centuries of technological development to a point where there is now the very realistic prospect of machines becoming smarter than humans. On the one hand, it provides the opportunity to solve the problems that Man has not been able to solve, and on the other it presents the danger of creating new problems that Man may not be in a position to solve. If a way is found to co-exist, AI provides the greatest opportunity to realise Man’s full potential.

 

The Leader: Endnotes

  1. Source: Goldman Sachs Global Investment Research

  2. Source: Europol (the EU’s law enforcement agency), via Gen.ai

  3. Source: UNICEF-Gallup poll, 2021

  4. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin; Attention Is All You Need, June 2017

  5. Source: Goldman Sachs Global Investment Research, The Potentially Large Effects of Artificial Intelligence on Economic Growth (March 2023)

  6. Source: ibid

  7. Source: Popular Mechanics, see https://www.popularmechanics.com/technology/a45780855/when-will-the-singularity-happen/ and https://www.popularmechanics.com/technology/robots/a42612745/singularity-when-will-it-happen/

  8. Source: 2022 Expert Survey on Progress in AI, Zach Stein-Perlman, Benjamin Weinstein-Raun, Katja Grace

  9. Source: Suzuki Kyoko, Conscious Awareness of Cognitive Processes and Their Dysfunction, 2009; Source: Suzuki Kyoko, Conscious Awareness of Cognitive Processes and Their Dysfunction, 2009;

  10. Erwin Schrodinger, a Nobel Prize winning physicist who developed fundamental results in quantum theory, wrote extensively about the metaphysics of the universal mind, writing in his book What is Life? Epilogue: On Determinism and Free Will, “There is obviously only one alternative, namely the unification of minds or consciousnesses. Their multiplicity is only apparent, in truth there is only one mind.”