Artificial General Intelligence: From the Goalposts of Human Intelligence to Implications for Society
Artificial General Intelligence (AGI) is being discussed by leading AI researchers and companies as the next leap in artificial intelligence, but what is AGI? What does it mean to be intelligent and how close are current AI models to human-level performance on a variety of subjects? When will AGI be achieved and what could it mean for our society? We’ll be exploring these questions and more in three parts, first, starting with the basics — what is intelligence? It’s commonly misunderstood as a score on an IQ test but it’s much more complex than that. We will discuss the basics of intelligence to set up the comparison for artificial versions (and hopefully dispel a few myths along the way).
The idea of intelligence first brought me to psychology where I studied and researched cognitive development — how do our brains and abilities develop over time to become what we’re using right now? It’s a fascinating topic and helps build the foundation for thinking about human intelligence — how did this lump of brain cells come to write this article? How did billions of years of evolution and our early lives as children shape our current brain? My brain tells me this is an important topic (but it is biased on the matter).
Let’s start with the basics — what is intelligence? I consider it the ability to learn from experience and adapt to new situations, which applies to bugs, crows, dogs, and neural networks, all at differing levels. Human-level intelligence also includes the ability to understand and handle abstract concepts like truth and justice as well as use knowledge to manipulate our environment. However, my very general definition doesn’t clearly depict the many facets of our intelligence, so here’s a better one:
“Intelligence is the capacity for abstraction, logic, understanding, self-awareness, learning, emotional knowledge, reasoning, planning, creativity, critical thinking, and problem-solving. It can also be described as the ability to perceive or infer information; and to retain it as knowledge to be applied to adaptive behaviors within an environment or context.”
Gardner suggests eight distinct forms of human intelligence — since our intelligence is so multi-faceted, we have a greater advantage in learning from and adapting to our environments. While there may be different dimensions of intelligence, there is also some evidence that the dimensions are correlated and there is a central g-factor of intelligence.
Since we like to think and talk about AI, let’s first talk about the hardware that your mind runs on. The human brain is a structured collection of approximately 86 billion neurons, with trillions of connections between them. Your brain is a massively parallel computing device. The individual neurons, like circuits, transmit electrical and chemical signals across synapses that connect to other neurons, exchanging and modulating information, all at an amazing microscopic scale. The complex interplay between neurons and their connections enables the brain to process and store information and allows us to be smart.
The brain’s organization also improves our intelligence — it’s organized into distinct regions, each with specific roles in supporting our multi-faceted intelligence. For example, the prefrontal cortex is involved in higher-order cognitive functions such as planning, decision-making, and problem solving. The cerebellum plays an important role in coordinating movements, while the occipital lobe in the back of your head is responsible for visual processing. Even more, the interconnectedness of these regions allows for the integration of information, for example the sensory and motor cortices are right next to each other, and the emergence of complex cognitive functioning.
Our cognitive processes are the basis of our intelligence, allowing us to perceive, attend to, process, and manipulate information. Our perception allows the integration of sensory information from our surroundings, which is then processed and interpreted by the brain. Attention enables the selective focusing on relevant stimuli (for those of us with ADHD), while filtering out irrelevant information. We have memory systems, including working memory and long-term memory, which are essential for storing and retrieving information. Our problem-solving and decision-making abilities involve the application of knowledge and reasoning skills to generate solutions and make informed choices. Our emotions serve as motivational drivers, guiding attention, memory, and decision making. They help us prioritize info, make meaningful connections, and respond appropriately to social settings. Creativity enables us to generate new ideas and find innovative solutions to our problems. We pull these systems together for decisions of whether to scroll to the next video, the best chess move, or whether I should purchase the newest VR headset (the answer is “yes”).
How did all this impressive cognitive machinery get on top of our shoulders? The evolution of brains is a fascinating story that spans over 500 million years, beginning with the emergence of the first nervous systems in simple organisms. In the early Cambrian period, around 520 million years ago, the first primitive brains appeared in simple animals, such as flatworms. These early brains were clusters of nerve cells that allowed for more complex behaviors and increased chances of survival. As animals evolved and diversified, so did their brains. In vertebrates, the hindbrain, midbrain, and forebrain began to differentiate, with the hindbrain controlling basic functions like breathing and heart rate, while the midbrain and forebrain enabled more advanced sensory processing and decision-making. The evolution of the forebrain, particularly the telencephalon, was a significant step in the development of complex cognition. In mammals, the telencephalon gave rise to the neocortex, which is responsible for higher-order cognitive functions such as perception, spatial reasoning, and even conscious thought. Primates underwent a rapid expansion of the neocortex, enabling the development of advanced social cognition, language, and problem-solving skills. And our human brains are much bigger and more complex than other great apes.
Another way to think about the question of how we got here is ontogenetically; that is, through the development of our brain from the first cells to what is up there now. From conception to adulthood, the human brain undergoes significant growth and change. During the prenatal period, the brain develops at an amazing rate, with neurons forming and migrating to their designated locations. At birth, a baby’s brain contains nearly all the neurons it will ever have, but it is still only about a quarter of its adult size. Over the first few years of life, the brain experiences rapid growth and development, with synapses forming at an incredible pace. This period of synaptic overproduction is followed by a pruning process, where unused connections are eliminated, allowing for the refinement of neural networks for speedier processing of commonly used pathways. As children engage with their environment, their brains continue to develop, with experiences shaping the structure and function of neural circuits. The brain’s flexibility and plasticity enable it to adapt and learn throughout childhood and adolescence, with the prefrontal cortex, responsible for executive functions and decision-making, being one of the last regions to fully mature in early adulthood. Our brains need a lot of time to grow and experience the world before becoming fulling formed.
Earlier, I mentioned intelligence isn’t as simple as an IQ test — it’s a complex and controversial topic. Historically, intelligence quotient (IQ) tests have been the most widely used method for assessing cognitive abilities. These tests typically measure a range of skills, including verbal comprehension, perceptual reasoning, working memory, and processing speed. However, critics argue that IQ tests are often culturally biased, focusing primarily on skills valued in Western societies, and fail to capture the full spectrum of human intelligence. Moreover, IQ scores can be influenced by factors such as education, socioeconomic status, and test-taking experience. In addition to measuring the various dimensions in Gardner’s Theory of Multiple Intelligences measure such as emotional intelligence (EQ) and practical intelligence, emphasize the importance of non-cognitive skills in overall intellectual functioning. However, measuring intelligence remains an imperfect science, and no single test or theory can fully capture the complexity and diversity of human cognitive abilities. Whether we’re measuring across people or across species, using multiple measures of intelligence is important.
Given this explanation of human intelligence, what happens when computers can reach our levels of intelligence? That’s AGI. Although there is some disagreement, Artificial General Intelligence (AGI) refers to an artificial intelligence that can understand, learn, and perform any intellectual task that a human being can. Unlike narrow AI systems that are designed for specific tasks, such as image recognition or language translation, AGI would possess the flexibility and generality of human cognition. An AGI system would be able to reason abstractly, solve novel problems, learn from experience, and adapt to new situations without requiring explicit programming or training for each specific task. Like us, it would have the ability to combine knowledge from multiple domains, draw insights, and make decisions based on incomplete or uncertain information. AGI could also be capable of exhibiting human-like qualities such as creativity, emotional intelligence, and even self-awareness. It is currently a matter of debate among experts regarding its feasibility, timeline, and potential implications for society. Some researchers believe that AGI could lead to transformative breakthroughs in science, technology, and human progress, while others caution about the potential risks and challenges associated with creating machines that can match or surpass human intelligence.
Artificial General Intelligence: Recognition through Benchmarks and Evaluations
‘“In from three to eight years, we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, have a fight. At that point the machine will begin to educate itself with fantastic speed. In a few months it will be at genius level, and a few months after that, its powers will be incalculable.”
There are many ways to measure intelligence, whether it is logical problem-solving ability, vocabulary size, juggling ability, trivia knowledge, or being able to recognize when someone is feeling sad. We will need to apply the same approach with AI to understand where it equals or even surpasses human abilities with different benchmarks and evaluation methods to measure and track progress. When AI has reached human-level performance, then Artificial General Intelligence (AGI) may be achieved and going beyond humans would be Artificial Superior Intelligence (ASI). Although, as with human intelligence, there is still much debate about what exactly AGI is and how we will know if and when it happens.
A good way to start defining AGI is to contrast it with “narrow” or “weak” AI which is designed to perform specific tasks within a limited domain — think self-driving cars. They are very good at visual recognition, spatial mapping and driving-decision making. They are not good at hypothesizing potential protein structures, writing python code, playing chess, or mimicking a person as a chatbot. These are all examples of narrow AI and are very different from us — our nervous system can do all these things and much more, not just one task. AGI should be more like us, able to perform any intellectual task as well as or better than a human can. That’s a significant challenge!
When we get there, AGI will be able to think, learn and understand on a similar level to humans and do what we can do (and soon thereafter better). Some likely capabilities of AGI include:
· Human-like intelligence: AGI would have the ability to reason, learn, and adapt in a way that is like human cognition.
· Flexibility: AGI would be able to apply its intelligence to a wide range of tasks and domains, just like humans can.
· Learning from experience: AGI would be capable of learning from its experiences and improving its performance over time.
· Generalization: AGI would be able to take knowledge and skills learned in one context and apply them to new, unfamiliar situations.
· Creativity and problem-solving: AGI would be able to think creatively and come up with novel solutions to problems, much like humans do.
· Self-improvement: AGI will be able to design better neural networks, better AI hardware, and better training datasets, all to improve itself.
When we get there, it might change everything (coming next in part 3 of the series).
Just as we’ve taken a multi-faceted approach to measuring the various dimensions of human intelligence, we need to take the same approach with AGI. Benchmarks and evaluations will help measure the capabilities of AI systems and determine their progress towards AGI. Without these objective measures, it becomes hard to compare different approaches, identify areas for improvement, and quantitatively track the advancement of the field. They become common measures for researchers and developers to evaluate their models, share their findings, and collaborate towards next steps towards AGI.
For decades, the benchmark for AGI was the Turing Test, developed by Alan Turing to assess a machine’s ability to exhibit intelligent behavior indistinguishable from a human. However, simple mimicry can be pretty convincing and today’s Large Language Models (LLMs) like ChatGPT are preferred by humans. The Winograd Schema Challenge, introduced in 2012, evaluates a system’s ability to resolve ambiguities in natural language based on common-sense reasoning. While this benchmark addresses some of the limitations of the Turing Test, it still focuses on a narrow aspect of intelligence. The General Language Understanding Evaluation (GLUE) benchmark, developed in 2018, assesses a model’s performance on a range of natural language understanding tasks. While GLUE provides a more comprehensive evaluation of language understanding, it does not cover other essential aspects of intelligence, such as reasoning, problem-solving, and creativity. Newer benchmarks like GAIA test for practical, real-world knowledge and reasoning across domains like science and general knowledge and AGIEval looks to measure general abilities of AI models relevant to human cognition and problem solving.
Google’s Deepmind has proposed “Levels of AGI” which organizes AI’s performance, generality and autonomy, similar to levels used in autonomous driving technologies. True AGI will be flexible and general, with both strong performance across domains, not just specialized in a particular domain. According to the chart below, we have made huge strides in Narrow AI, surpassing human levels at Level 5. However, for General AI, we are only at Level 1, “Emerging AI” with systems like ChatGPT and Claude. Soon, we may hit Level 2, “Competent AI” that is as good as the average human (cue relevant George Carlin quote).
According to Stanford’s AI Index Report, current AI models are as good as or better than humans on most measures. There are fewer and fewer domains that we are better than AI, even for domains like creativity, which we thought were pretty uniquely human.
Looking across standardized tests from high-school level to graduate level, GPT-4 exceeds 80th percentile human scores on several tests, including most AP courses, but more impressively the Bar Exam, LSAT and GRE exams.
Some experts believe that we’re starting to see the beginnings of AGI in today’s LLMs or even full-blown AGI. However, there is no consensus about what AGI is, let alone when we might get there. In 2022, half of the over 300 experts expected AGI before 2061 (in 40 years) and 99% thought it would happen in the next 100 years.
Recent advancements have pushed some people’s timeframe up. Mustafa Suleyman, author of The Coming Wave, suggests AI will achieve this human-level performance soon, saying “within the next few years, AI will become as ubiquitous as the Internet”. He adds, “The coming wave of technologies threatens to fail faster and on a wider scale than anything witnessed before”.
Improvements in neural network architectures, better training data, and faster hardware may quickly take us to AGI, which further underscores the importance of appropriate benchmarks and measurement. Once we start seeing real glimpses of AGI, full-blown AGI won’t be far behind, so we need to heed the warnings.
Skeptics like Meta’s Turing Award Winner Yann LeCun argue that we don’t have the right tools now for AGI, saying:
“It’s astonishing how [LLMs] work, if you train them at scale, but it’s very limited. We see today that those systems hallucinate, they don’t really understand the real world. They require enormous amounts of data to reach a level of intelligence that is not that great in the end. And they can’t really reason. They can’t plan anything other than things they’ve been trained on. So they’re not a road towards what people call “AGI.” I hate the term. They’re useful, there’s no question. But they are not a path towards human-level intelligence.”
“So the mission of FAIR [Meta’s Fundamental AI Research team] is human-level intelligence. This ship has sailed, it’s a battle I’ve lost, but I don’t like to call it AGI because human intelligence is not general at all. There are characteristics that intelligent beings have that no AI systems have today, like understanding the physical world; planning a sequence of actions to reach a goal; reasoning in ways that can take you a long time. Humans, animals, have a special piece of our brain that we use as working memory. LLMs don’t have that.”
As Meta pursues embodied intelligence, which is closer to how children learn about the world and develop their intelligence — a much deeper understanding of our reality than pure text could ever capture. Google is similarly pursuing the combination of AI and robotics, which brings autonomy and action in the world. By connecting a LLM to a robot, the system can plan and execute commands to achieve its goals in the real world like cleaning up a spill on the counter by grabbing a towel and wiping up (the video still trips me out…)
As AI systems develop, benchmarks will need to develop too. They should be an interdisciplinary effort involving AI researchers, cognitive scientists, neuroscientists, and philosophers. With diverse perspectives and concrete benchmarks, we can trace the development of AI’s pursuit of human intelligence and pinpoint the moment when it surpasses us.
We previously discussed the goalposts of AGI, human intelligence in its many facets. However, as AIs approach and surpass human-level capabilities on many tasks, the goalposts may shift. That’s why we need concrete and reliable benchmarks to trace AI development and provide some warning that AGI may be approaching. The field is advancing so rapidly that benchmarks can’t play catch up and need to stay ahead of the curve to accurately measure future advancements and provide clear comparisons to human abilities.
What does achieving AGI mean for us? What capabilities will it enable? What does it mean for our society? What does it mean for our species?
Potential Benefits of AGI to Society
“Our mission is to ensure that artificial general intelligence — AI systems that are generally smarter than humans — benefits all of humanity. If AGI is successfully created, this technology could help us elevate humanity by increasing abundance, turbocharging the global economy, and aiding in the discovery of new scientific knowledge that changes the limits of possibility. AGI has the potential to give everyone incredible new capabilities; we can imagine a world where all of us have access to help with almost any cognitive task, providing a great force multiplier for human ingenuity and creativity.”
-OpenAI CEO Sam Altman on AGI
All of this discussion about intelligence leads us to the ultimate question of “So what?” The so what is Artificial General Intelligence (AGI) which could have transformative effects across our society. It could revolutionize the way we live, work, and solve complex problems of every type. When we create intelligent machines that can think, learn, and reason like humans, innovation will be unleashed on science, the economy, climate change, even social progress. And when these intelligent machines can improve themselves by designing better neural network designs and more complex AI chips, their capabilities will far outpace ours in a short time. Imagine having a million tireless super-geniuses working on our biggest issues — innovation is sure to take off.
AGI has the potential to accelerate scientific research and discovery by helping scientists analyze huge amounts of data, generate and test hypotheses, iterate, and identify patterns and insights that would take humans experts years to uncover, if ever. Think about the complexity of identifying intricate protein structures to fight disease or analyzing all of the cosmic data coming from the James Webb Space Telescope (JWST) for keys to how the universe works. In fields such as medicine, materials science, and space exploration, AGI could help us develop new treatments, create innovative materials, and explore the universe in ways we never thought possible. By augmenting human intelligence with the speed, accuracy, and relentlessness of AGI, we could achieve scientific breakthroughs at an unprecedented rate across fields. And AGI would also look inwards and work on improving itself, leading to an exponential increase in capability.
“Superintelligence … is the last invention we will ever need to make.”
AGI could also help us solve complex global challenges that have been plaguing us such as climate change, epidemics, and even poverty. With the ability to process and intelligently analyze massive amounts of data from various sources, AGI could help us develop more accurate climate models, optimize resource allocation, and identify effective strategies for mitigating the impacts of climate change. To help fight poverty, AGI could assist in creating personalized interventions, improving access to education and healthcare, and encouraging economic development in underprivileged communities. AGI can help us accelerate progress towards the United Nations’ Sustainable Development Goals and create a more equitable and sustainable world.
AGI can enhance both our society at a macro level but also our many individuals. Personalized education powered by AGI could adapt to each person’s unique learning style, pace, and interest, providing bespoke content and support to maximize everyone’s learning potential. In healthcare, AGI could revolutionize diagnosis, treatment, and drug discovery by analyzing patient data, identifying risk factors, and developing personalized treatment plans. AGI-assisted technology could also help individuals with disabilities, providing greater accessibility and independence. AGI can potentially help everyone thrive in their own unique way.
AGI will drive significant economic growth (and thus the high-stakes race to be the first) and job creation in new industries. Just as the Industrial Revolution and the internet transformed the global economy, the development of AGI would create a new era of innovation and productivity. New businesses and industries will emerge to manage and harness the power of AGI, creating job opportunities in fields such as AI development, data analysis, and robotics. AGI would help automate repetitive and dangerous tasks, freeing up human workers to focus on doing more fulfilling work. We are already using AI to do repetitive tasks like clean/ recode datasets, writing basic code, which frees up our employees to spend their time feeding their curiosity.
Turning AGI’s superior reasoning abilities towards governance and decision-making could provide data-driven insights to help leaders make more informed and effective choices. AGI could help policymakers identify trends and develop evidence-based solutions to complex problems. AGI could also assist in monitoring the implementation and impact of policies, allowing real-time adjustments and improvements. We can create more responsive and accountable governmental systems that better meet and serve the needs of their citizens.
Some additional potentially transformative effects of AGI:
· Safety and Security: From predicting and managing natural disasters more effectively to enhancing cybersecurity, AGI has the potential to protect societies and potentially predict disasters ahead of time.
· Quality of Life: Everyday life could be markedly improved with AGI through smarter cities, personalized technology assistance, and by automating mundane tasks, allowing people more leisure time and a higher standard of living.
· Ethical and Fair Decision-Making: Though this is a complex and contentious area, there’s potential for AGI to assist in making unbiased decisions based on comprehensive data analysis, benefiting governance, justice, and societal norms.
“I believe that the rapid progress of AI is going to transform society in ways we do not fully understand and not all of the effects are going to be good.”
-Geoffrey Hinton, “Godfather of AI” on AGI
However, once we achieve AGI, or “The Singularity”, growth becomes uncontrollable and irreversible. As we pursue this goal, we need to consider the ethical implications and ensure it is developed responsibly. The creation of intelligent machines raises important questions about privacy, security, fairness, and accountability. We need to have clear guidelines and regulations to govern the development and deployment of AGI in decision making, ensuring that it aligns with human values and promotes society’s wellbeing. There will likely be unintended consequences, ones that we can’t foresee, but by staying ahead of development with ethical considerations, we can maximize its benefits and minimize its risks. Additional concerns include:
· Loss of Jobs: One of the most immediate concerns is the automation of tasks currently performed by humans, leading to widespread job losses and economic inequality. While new jobs may be created, there is uncertainty about the pace of this transition and whether the new jobs will be accessible to those displaced.
· Security Risks: AGI systems could be weaponized or used in cyber-attacks, presenting unprecedented challenges to national and global security. The autonomous nature of AGI could make controlling such systems difficult if they are deployed maliciously or without sufficient safeguards.
· Ethical and Privacy Concerns: The deployment of AGI could lead to increased surveillance and erosion of privacy. Decisions made by AGI systems might not align with human values or ethics, especially in sensitive areas like criminal justice, healthcare, and social services.
· Loss of Control: A fundamental risk of AGI is the possibility that it could become uncontrollable or make decisions that are incomprehensible to humans, leading to unintended consequences. This includes the so-called “alignment problem,” where AGI’s goals diverge from human values and intentions.
· Economic and Social Inequality: The benefits of AGI might not be evenly distributed, exacerbating existing inequalities. Those who control AGI technologies could gain disproportionate wealth and power, leading to greater social stratification.
· Dependency: Over-reliance on AGI systems could erode human skills and capacities. In critical sectors like healthcare, transportation, and defense, this could have serious implications if systems fail or are compromised.
· Existential Risk: Perhaps the most extreme risk is the hypothetical scenario where AGI surpasses human intelligence and autonomy, potentially leading to human extinction or a radical alteration of human life.
To best realize the beneficial effects of AGI, we need collaboration between researchers, policymakers, and our society at large. It shouldn’t be mega corporations or academics deciding alone — it requires input and engagement from diverse stakeholders, including governments and the public. With an open dialogue, we can ensure that AGI reflects the needs and values of our entire society. And with a more productive future, we can share the benefits worldwide.
Once AI systems are as smart or smarter than our smartest humans, the benefits to society will be immense and far-reaching. From accelerating scientific discovery, to solving global climate change, to technological innovation, AGI has the potential to transform our work and personal lives. However, to get there, we need a responsible and collaborative approach to AGI development, one that prioritizes the well-being of humanity first and foremost. Together, we can harness the potential of AGI for a better future for us and future generations of humans. As we stand on the precipice of a monumental achievement, it becomes increasingly important to shape the right future. AGI isn’t just a technical challenge, but a social and philosophical one too. It requires us to grapple with fundamental questions like the nature of intelligence, our relationship with machines, and the potential for machine consciousness. We may still have a little time to figure all of this out before it’s too late…
