Path to Artificial General Intelligence

The rapid development of artificial intelligence (AI) in recent years has inspired new visions of our future, many of which are being developed by Big Tech and thousands of startups worldwide. Now, artificial general intelligence (AGI) beckons as a thrilling and challenging frontier to explore, offering many possibilities and transformations across most industries.

AGI can be defined as the creation of machines that possess the ability to understand or learn any intellectual task that a human being can do. However, there is no strict definitional consensus. Some experts, such as Sam Altman, CEO of OpenAI, describe AGI as “AI systems that are generally smarter than humans.”i Still, simply put, AGI is a machine’s capacity to perform tasks and solve problems with a level of cognitive ability comparable to—or exceeding—human intelligence.

AGI encompasses various abilities, such as reasoning, learning, problem-solving, perception and language understanding. These capacities enable an AGI system to be versatile and adaptive, granting it proficiency in diverse, wide-ranging tasks, without needing explicit programming for each one.

AGI’s promise lies in the ability of digital systems to help humanity solve its most complex challenges. Environmental degradation, income inequality, drug discovery, space exploration, material science, food production and mental health are areas in which AGI can make significant contributions.

However, it’s unclear how to create an AGI-capable system.

Examples of narrow AI include language translation algorithms, facial recognition systems and self-driving vehicle technologies. These AI systems are designed for a specific purpose and cannot perform tasks beyond their designated domains. In contrast, AGI aims to equip machines with the cognitive flexibility to apply their intelligence widely, emulating human-like strategic thinking and problem-solving, which are required to resolve complex issues.

The concept of AGI can be traced back to the early days of AI research in the 1950s. AI’s founding fathers—John McCarthy, Marvin Minsky, Allen Newell and Herbert A. Simon—believed in the possibility of creating machines that could perform any intellectual task a human could. The term “artificial intelligence” was coined to describe machines or systems “exhibiting behavior at least as skillful and flexible as humans.”iii

Since then, AI research has mainly focused on creating ANI systems, producing many successful and transformative specialized AI applications. Achieving AGI has remained the ultimate goal but has proven more elusive than initially anticipated. Recent advancements in deep learning, neural networks and reinforcement learning have rekindled the hope of progressing towards AGI, but there is still a long way to go.

Convolutional Neural Networks

One of the most widely used methods in image processing and analysis is the Convolutional Neural Network (CNN). Inspired by the visual processing mechanisms of the human brain, CNNs excel in tasks such as facial recognition by learning to identify patterns through training on large datasets.

A typical CNN consists of an input layer, multiple hidden layers and an output layer. The input layer receives arrays of picture pixels, while the hidden layers, composed of “neurons,” utilize mathematical operations to extract features from the image. These hidden strata include, among others, convolution, pooling and fully connected layers.iv

The convolutional layer is usually first to extract features from an input image. Once the image passes through the hidden layers, it hits the output layers, which classify and deduce results from the image processing.

Key AGI Approaches & Techniques

Various methodologies have been proposed to develop AGI, including:

  • Reinforcement learning is an approach in which an AI agent (software) interacts with its environment and learns to achieve its goal through trial and error, guided by a reward system.
  • Deep learning is a subset of machine learning (ML) that employs artificial neural networks to learn hierarchical representations of data, thereby enabling computers to perform tasks with minimal human intervention.
  • Knowledge-based systems involve a symbolic approach that uses formal rules and logic to represent and manipulate knowledge, allowing machines to reason about complex problems.
  • Evolutionary algorithms are a class of optimization algorithms, inspired by natural evolution, that uses techniques such as selection, crossover and mutation to solve optimization problems.

Challenges & Limitations

Despite promising advancements in AI research, several challenges must be addressed before AGI can become a reality. Some of the most significant include:

  • Scalability: Developing an AGI system that can scale its learning and understanding across multiple domains, tasks and environments remains a significant hurdle.
  • Transfer learning: AGI requires the ability to transfer knowledge from one domain to another, a feat that remains challenging for current AI systems.
  • Explainability: As AGI systems become more sophisticated, understanding the decision-making process behind their actions becomes increasingly crucial yet difficult.
  • Resource constraints: Developing AGI requires vast computational resources, posing a challenge to those with limited access to cutting-edge hardware and expertise.

As mentioned above, CNN architecture consists of input and output layers, along with multiple hidden layers. CNNs are designed to detect patterns within the whole image, which are crucial for calculations such as classification. In this analysis, we utilize four convolutional layers. Images, in any format, are processed through the first convolutional layer where they undergo max pooling, which selects the brighter pixels, and then are converted into the 128×128 pixel format size. Then these images are sent through the second convolutional layer where essential features, such as facial geometry points, are extracted, again by applying max pooling. The final convolutional layer resizes these images into the 32×32 pixel size and converts them into arrays to accelerate measuring the distances between illuminated pixels to make the images recognizable by the human eye.

Breakthroughs & Milestones

Although AGI has yet to be achieved, some notable breakthroughs in AI research have inched us closer to realizing this goal. For example, large language models (LLMs), such as OpenAI’s GPT-4, Meta’s Llama 2 and NVIDIA’s Megatron-LM, which generate human-like text, have demonstrated impressive abilities in multiple language tasks.

Adam stochastic gradient descent: In our convolution layers, we employed the Rectified Linear Unit (ReLU) activation function. For the output layer, the Softmax activation function was used. To optimize this model, we utilized the Adam algorithm, a variation of stochastic gradient descent known for its adaptive learning rate adjustment. This entire process is detailed above.

The capabilities of LLMs are hugely impressive and continue to improve daily. Now, OpenAI is developing and releasing integrations that will allow its language models to interact with other applications and the internet in general. This will bring new and exciting capabilities across a variety of industries—especially health care, transportation and finance—and will influence a renaissance in web technologies and scientific research.

In The Singularity is Near: When Humans Transcend Biology, Ray Kurzweil discusses linear versus exponential growth. He describes the concept of “knee of the curve,” the point at which growth rapidly increases and the change from linear to exponential growth is observed.v The launch and widespread use of LLMs will mark the turning point where technological innovation moves from linear gains to exponential gains. As more users have access to capabilities that they would not normally have, they will be able to rapidly develop solutions and collaborate with others to achieve even more than they otherwise would alone.

DeepMind’s AlphaGo and AlphaZero programs have showcased expert-level gameplay and self-learning in competitive board games. These milestones provide valuable insights into the potential capabilities of AGI, inspiring further research and development.

Potential Applications of AGI

The successful development of AGI could revolutionize almost all industries and professions, accelerate scientific discovery and contribute to solving complex global challenges. This section highlights the transformative potential of AGI across different domains.

Transforming Industries and Professions

AGI could significantly impact industries such as healthcare, finance and manufacturing, by offering innovative solutions and streamlining processes. For example, AGI systems could assist physicians in diagnosing and treating complex medical conditions, help financial analysts find optimal investment strategies, or revolutionize supply chain management by optimizing production and logistics.

Advancements in Scientific Research

One of the most intriguing aspects of AGI is its potential to accelerate scientific research and discovery. AGI systems could assist researchers in finding groundbreaking solutions to long- standing problems—for example, harnessing nuclear fusion that would provide the world with limitlessly clean and self-sustaining energy—as well as providing innovative ideas and techniques to various scientific fields, such as physics, chemistry and biology.

Solving Complex Global Problems

AGI has the potential to help humanity address some of its most pressing challenges, including environmental degradation, poverty and inequality. By providing new insights and solutions, AGI could contribute to more sustainable policies, resource management and equitable distribution of knowledge.

Ethical Considerations & Implications

AGI development raises serious ethical concerns revolving around its alignment with human values, potential misuse and governance.

AGI Alignment with Human Values

As we develop AGI, it is critical to ensure alignment with values that serve humanity’s interests. Researchers and developers must prioritize the safe and responsible design of AGI technology, incorporating fairness, transparency and inclusiveness to minimize potential harms and maximize benefits.

As AGI advances, the power, autonomy and rationality of AGI systems will increase and approach goals that may risk misalignment with positive human values and create new conflicts of interest. To moderate this, it is crucial to develop methods to control and influence AGI’s decision-making processes. AGI could be used for malicious purposes as already shown with ANI applications, such as surveillance systems that infringe on privacy rights.

The power of AGI comes with the potential for misuse, and it is crucial to consider how to safeguard against unintended consequences and malicious exploitation. Establishing strict security protocols, fostering international collaboration and devising robust countermeasures are crucial measures to mitigate the risks posed by AGI.

Ensuring responsible use requires robust safeguards and regulations that should be developed with international regulation, shared legal frameworks and comprehensive oversight to prevent deployment that causes harm to human individuals, institutions or societies.


A critical aspect of this journey involves the evolving capabilities of large language models (LLMs). While the advancements in LLMs and other generative AI systems are impressive and continue to improve, they have significant environmental costs. For instance, “training GPT-3 in Microsoft’s state-of-the-art U.S. data centers can directly consume 700,000 liters of clean freshwater, and the water consumption would have been tripled if training were done in Microsoft’s Asian data centers.”vi

Of even greater concern, “[c]ompanies began sounding the alarm about data center power consumption five years ago at the annual Hot Chips semiconductor technology conference by predicting that worldwide compute demand could exceed the total world electricity power generation within a decade.”vii For a thorough analysis of AI’s energy usage, please refer to “AI’s Energy Appetite: Voracious & Efficient” in the Fall/Winter 2023 issue of Dakota Digital Review.viii

Generative AI workloads will increase as more individuals, organizations, businesses and government agencies identify its value. Different approaches to building AI and AGI systems will need to be developed to ease water and energy consumption.

Debate on AGI Regulation & Governance

To ensure the responsible and ethical development of AGI, there is an ongoing debate surrounding its regulation and governance. Policymakers and AI researchers must engage in a continuous dialogue to create comprehensive frameworks and policies that balance innovation with ethical concerns, thereby guiding the future of AGI in a manner that benefits humanity.

AGI will undoubtedly change the nature of work as it currently exists today. Research suggests that in five to 20 years, a third of jobs could disappear because of LLMs.ix Lawmakers and governments worldwide will have to carefully study the impact LLMs have on the employability of its citizenry, so that a strategy is put in place to address the resulting disruption. A basic starting point could be the creation of state-funded, job-retraining programs that identify individuals affected or displaced by AGI and LLMs.

Corporations and governments will also have to come to terms with the capabilities and limitations of AGI—as examples, rapid preprocessing of large datasets and the challenges of data dependence in establishing the contexts needed for comprehension—to ensure that safeguards are in place to prevent models from producing damaging content that promotes discrimination, hatred and violence. This will require careful collaboration between business and government to ensure that legislation and regulations safeguard the public yet do not stifle innovation.

Looking forward, a hypothetical AGI system called Dishbrain might prove very helpful, harmless and honest, for example, when used as a personal assistant. Its broad intelligence could revolutionize everyday life as it learns from vast online datasets, offering customized advice, managing schedules and even providing emotional support. While its applications in education, healthcare and creative fields are promising, responsible development and user control are crucial to guard against bias and manipulation and to ensure alignment with principles beneficial to humanity. ◉


i https://openai.com/blog/planning-for-agi-and-beyond

ii Ramuš Cvetkovi?, Iva, and Marko Drobnjak. “As Above so Below: The Use of International Space Law as an Inspiration for Terrestrial AI Regulation to Maximize Harm Prevention.” Artificial Intelligence, Social Harms and Human Rights. Cham: Springer International Publishing, 2023. 207-238.

iii Prentice, Catherine. “Demystify Artificial Intelligence.” Leveraging Emotional and Artificial Intelligence for Organisational Performance. Singapore: Springer Nature Singapore, 2023. 25-40.

iv C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2015, pp. 1-9.

v Kurzweil, R. The Singularity is Near: When Humans Transcend Biology, Penguin Books, 2005, p. 10.

vi Li, P., Yang, J., Islam, M. A., & Ren, S. (2023). “Making AI Less ‘Thirsty:’ Uncovering and Addressing the Secret Water Footprint of AI Models.” ArXiv (Cornell University). https://doi.org/10.48550/arxiv.2304.03271

vii McGregor, J. (n.d.). “Generative AI Breaks the Data Center: Data Center Infrastructure and Operating Costs Projected to Increase to Over $76 Billion By 2028.” Forbes. Retrieved May 24, 2023, from https://www.forbes.com/sites/ tiriasresearch/2023/05/12/generative-ai-breaks-the-data-center-data-center- infrastructure-and-operating-costs-projected-to-increase-to-over-76-billion- by-2028/?sh=40da55667c15

viii https://dda.ndus.edu/ddreview/ais-energy-appetite-voracious-efficient/

ix Zarifhonarvar, A. (2023). “Economics of ChatGPT: A Labor Market View on the Occupational Impact of Artificial Intelligence.” EconStor Preprints.

Atif Farid Mohammad
Atif Farid Mohammad
Atif Farid Mohammad, PhD, is the Global Head of GenAI R&D and Chief Data Officer at Global Technology Solutions, as well as an Adjunct Professor of Artificial Intelligence at University of the Cumberlands and an Adjunct Professor of Cyber Security at Texas Wesleyan University. He is also the Quantum Computing Dissertation Chair at Capitol Technology University in Washington, D.C. Prof. Mohammad has authored the Tech Tips column at several web platforms and extensively published at IEEE, Springer and other publications and conferences since 2008. He completed his undergraduate degree at the University of Karachi, Pakistan, and his master’s degree at Queen’s University, Kingston, ON. Prof. Mohammad completed his first PhD in Cyber Security at the University of Quebec and his second PhD in Scientific Computing at UND.

Most Popular

Recent Comments