AI: We’re only at stage 4 out of 10!
Generative AIs like GPT-4 are already capable of delivering impressive performances on certain tasks, though they remain uneven. However, it’s easy to envision these technologies evolving further. So, what might the capabilities of AI be in 10, 20, or even 50 years? By extending the exercise to technology serving major societal challenges, it’s conceivable to outline up to 10 stages of AI, suggesting we’re currently only at stage 4. Curious to know what’s next?
What is Artificial Intelligence?
First, let’s return to the basics! In 2023, it doesn’t take long to find excellent reports offering a detailed history and definition of artificial intelligence. I recommend three below, and I don’t intend to linger on this part, as there’s plenty of well-made documentation:
- On The Economist website, a file teeming with numerous articles.
- On The Street website, a very comprehensive file.
- On the IBM website, an in-depth and well-explained article.
I won’t insult you by suggesting reading the Wikipedia page on AI, but I assure you it’s detailed and very informative! For this article, here’s the definition we’ve adopted at Tomorrow Theory: AI is a technology aiming to simulate and reproduce human capacities to learn, think, and interact. Once this is established, it’s clear there’s a difference between the very first attempts at AI over 50 years ago, GPT-4 today, and the AIs seen in science fiction works. In fact, it’s amusing to distinguish 10 stages of artificial intelligence evolution. And whether it reassures you or not, GPT-4 is only at stage 4 out of 10!
For each stage, I offer a quick definition below, the main characteristics, some concrete examples of capabilities at this stage, and the impacts and challenges for society in general. If the early stages correspond to consensus widely found among AI experts, the more prospective stages are still subject to discussion, and what I propose below simply reflects my vision and understanding of AI’s potential for the future.
1/10 — Early symbolic AI (1950–1970)
The early symbolic AI period represents the dawn of artificial intelligence, where researchers explored the possibility of creating machines that could mimic certain aspects of human intelligence.
Characteristics
- Formal logic: These systems used formal logic, a mathematical approach to represent and manipulate concepts. Formal logic allowed these systems to execute reasoning based on sets of logical rules.
- Rule-based systems: AI was programmed with sets of explicit rules. These rules dictated how the system should react to certain inputs or situations.
- Imitation of thought processes: The goal was to simulate certain aspects of human intelligence, such as problem-solving and language understanding, using computer models.
- Lack of learning: Unlike modern AI systems, these early forms of AI couldn’t learn new information or adapt to novel situations. Their operation depended entirely on preprogrammed rules.
Example
ELIZA ELIZA was developed in the 1960s by Joseph Weizenbaum at MIT. It was one of the first programs to attempt to simulate a human conversation. ELIZA used a script, often called “DOCTOR,” which mimicked a Rogerian psychotherapist. It relied on text transformation rules to respond to user inputs. For instance, if a user said, “I feel sad,” ELIZA could respond, “Why do you say you feel sad?”
Impacts
Although ELIZA could create an illusion of understanding and conversation, its responses were entirely based on text manipulation rules without any real understanding of language or context. Nonetheless, ELIZA sparked debates on the nature of intelligence and the possibility of artificially reproducing it. Weizenbaum himself was critical of the excessive use and overinterpretation of his program’s capabilities.
Credits
- John McCarthy and Marvin Minsky: AI pioneers, with foundational work in the 1950s and 1960s.
- “Artificial Intelligence: A Modern Approach” by Stuart Russell and Peter Norvig: Offers a historical overview of early AI.
2/10 — Early machine learning revolutions (1980–1990)
The period of the early machine learning revolutions marks a significant transition in artificial intelligence development, moving from a rule-based approach to one centered on learning from data.
Characteristics
- Learning from data: Unlike previous systems that operated on strict rules, these AI systems could learn and adjust based on the data they processed. This approach paved the way for greater flexibility and adaptability.
- Introduction of neural networks: Neural networks, inspired by the structure and functioning of the human brain, allowed the creation of models capable of processing complex data like images and sounds.
- Genetic algorithms: Inspired by evolutionary theory, these algorithms simulated a natural selection process to optimize solutions to complex problems.
- Machine learning models: These systems were capable of improving their performance as they were exposed to more data, marking a significant advancement in AI’s self-improvement capability.
Example
NetTalk NetTalk was created by Terry Sejnowski and Charles Rosenberg in the 1980s as an artificial neural network designed to learn English pronunciation. It converted text to speech by learning to associate letters with phonetic sounds. NetTalk learned by being exposed to examples of texts and their correct pronunciation. Over time, the system improved its ability to pronounce new words more accurately. NetTalk demonstrated the potential of neural networks for learning and generalizing from examples, a foundational step in the development of modern AI systems.
Impacts
This period marked a paradigm shift in AI system design, moving from a programmed and rigid approach to a flexible and self-adaptive one. These advancements also presented new challenges, particularly in terms of understanding and interpreting machine learning models, often considered as “black boxes.” The technologies developed during this period laid the groundwork for future advancements in deep learning and more general artificial intelligence.
Credits
- Terry Sejnowski and Geoffrey Hinton: For their work on neural networks and deep learning.
- “Pattern Recognition and Machine Learning” by Christopher Bishop: Covers the foundations of machine learning, including neural networks.
3/10 — Contextual and specialized AI (2000–2010)
The period of contextual and specialized AI marks an era where artificial intelligence began to be applied more targetedly and functionally in various domains.
Characteristics
- Contextual understanding: These AI systems were designed to understand and act in specific contexts, often thanks to significant advancements in natural language processing (NLP). They could interpret the nuances of human language at a certain level and provide relevant responses or actions.
- Specialization: Unlike previous more generalist systems, these AIs were often specialized for specific tasks, like voice recognition, information retrieval, or personal assistance.
- Domain limitations: Although performant in their specific domains, these systems had notable limitations when confronted with tasks or contexts outside their specialization.
Example
Siri Siri was launched in 2010 as an application on the iPhone 4S, becoming one of the first intelligent personal assistants integrated into a smartphone. Siri uses voice recognition and NLP to understand user voice commands and provide responses or perform actions. It can answer questions, make recommendations, and facilitate certain tasks like sending messages or making reservations. This tool represented a major advancement in the user interface, allowing more natural and intuitive interaction with technology via voice. The introduction of Siri marked a shift in how people interact with their devices, promoting the adoption of voice technology in other applications and devices.
Impacts
This period saw AI become more accessible and useful in daily life, moving from research labs to mainstream applications. Contextual AI raised questions about data privacy, security, and personal information management, particularly with always-listening devices. The advancements made in NLP and voice recognition during this period laid the groundwork for further progress in deep learning and more sophisticated personal assistants.
Credits
- Siri and its developers at Apple: For practical applications of AI in personal assistants.
- “Speech and Language Processing” by Dan Jurafsky and James H. Martin: A key resource on natural language processing.
4/10 — Deep learning era (2010–2024)
The Deep Learning Era represents a revolutionary period in artificial intelligence, characterized by significant advancements in AI systems’ ability to learn and perform complex tasks.
Characteristics
- Deep neural networks: This era is marked by the intensive use of deep neural networks (deep learning), AI systems that mimic the human brain’s functioning to process and interpret complex data.
- Complex tasks: These systems are capable of handling tasks like image recognition, natural language processing (NLP), and strategic decision-making, which were previously very difficult or impossible for machines.
- Learning from Big Data: Deep learning allows AI systems to learn and improve by processing vast amounts of data, giving them a far superior adaptability and generalization capability compared to previous approaches.
Examples
AlphaGo and GPT-4 AlphaGo is a program developed by Google’s DeepMind that defeated a world champion in the game of Go, known for its complexity and strategic depth. AlphaGo uses reinforcement learning and neural networks to learn game strategies. Developed by OpenAI, GPT-4 is a cutting-edge language model that can generate text closely mimicking human style and content. With 175 billion parameters, it represents one of the most sophisticated and largest language models to date.
Impacts
Deep learning has led to significant advancements in numerous fields, including healthcare, where it’s used for medical image analysis, and in automotive, for developing autonomous vehicles. With these advancements, new challenges have emerged, particularly concerning the explainability of AI models, managing biases in training data, and ethical concerns related to automation. The deep learning era has also profoundly impacted society, changing how we interact with technology and paving the way for more personalized and interactive AI applications.
Credits
- Yann LeCun, Yoshua Bengio, and Geoffrey Hinton: Often referred to as the “fathers of deep learning.”
- “Deep Learning” by Ian Goodfellow, Yoshua Bengio, and Aaron Courville: A reference text on deep learning.
5/10 — Integrated and collaborative AI (2024–2028?)
Integrated and collaborative AI symbolizes an era where artificial intelligence integrates more harmoniously and functionally into various aspects of society and industry. This period is characterized by increased collaboration between AI systems, humans, and other technologies.
Characteristics
- Systemic integration: AI is no longer seen as an isolated tool, but as an integrated component in broader systems, working in concert with other technologies and processes.
- Human-AI collaboration: A particular emphasis is placed on creating systems where AI and humans can work together effectively, leveraging each other’s strengths to improve decision-making and operational efficiency.
- Process optimization: AI in this era aims to improve and streamline existing processes, bringing innovative solutions to complex problems in various fields.
Example
A concrete example in development corresponds to intelligent city management systems. These systems use AI to analyze and integrate data from various sources (traffic, energy consumption, public services, etc.) to optimize urban services. This can include intelligent traffic management, energy usage optimization, environmental monitoring, and improving emergency and security services. More broadly, one can talk about how AIs will integrate into our daily work modalities, for example, in our online navigation and the use of office tools (Microsoft 365 Copilot AI).
Impacts
The integration of AI into complex systems allows for improved operational efficiency in many fields, from urban management to industrial production. This integration raises questions about data governance, privacy, equity, and employment, as AI begins to play a more active role in decisions directly affecting individuals. The era of integrated and collaborative AI promotes innovation by encouraging the development of new technological solutions that combine artificial intelligence with other areas like IoT, robotics, and cyber-physical systems.
Credits
“Human Compatible: Artificial Intelligence and the Problem of Control” by Stuart Russell: For a perspective on the ethical integration of AI.
“Life 3.0: Being Human in the Age of Artificial Intelligence” by Max Tegmark: Discusses the future of AI in society.
6/10 — General AI or AGI (2028–2040)
The period heading towards artificial general intelligence (AGI) represents a future and highly anticipated phase in artificial intelligence development, where AI systems approach human cognitive capacity. This is the goal of major current AI companies, and it is highly likely that recent advancements (such as OpenAI’s Project Q*) on AI reasoning mark an upcoming breakthrough on the subject.
Characteristics
- General capability: AGI refers to AI systems capable of understanding, learning, and applying their intelligence to a wide range of domains, not just in specific niches for which they were trained.
- Adaptability and learning: These systems would be capable of learning new tasks and adapting to unfamiliar environments, mimicking human cognitive flexibility.
- Versatility: AGI would be capable of crossing boundaries between varied domains like science, art, and technology, making innovative contributions and solving complex problems.
Theoretical examples
Theoretical AGI systems could be imagined as capable of conducting independent scientific research, creating original works of art, or developing new technologies without specific human intervention. AGI could play a central role in medicine, providing accurate diagnostics and personalized treatments, thus revolutionizing healthcare. In education, it would offer tailored learning experiences, adapted to the individual needs of each student, for a personalized and quality education.
In scientific research, AGI could accelerate discoveries by processing massive volumes of data, contributing to major advancements in physics, biology, or astrophysics. On the environmental front, its application in managing natural resources and combating climate change could lead to more sustainable and effective strategies. From an industrial perspective, AGI would completely transform factories through advanced automation, making production chains more efficient and intelligent. In terms of security, it would play a crucial role in cyber defense, anticipating and responding quickly to digital security threats.
Finally, AGI would open unprecedented horizons in space exploration, helping to analyze complex data from telescopes and pilot space missions, thus bringing us closer to answers to some of the universe’s biggest questions. These scenarios illustrate the extraordinary potential of AGI to significantly enrich and transform our world.
Impacts
The advent of AGI would mark a fundamental shift in the nature of artificial intelligence, moving from specialized systems to versatile and adaptable entities. Such emergence would raise profound questions about ethics, governance, safety, and the impact on society and the labor market. AGI could pave the way for unprecedented levels of innovation, collaborating with humans to solve previously insurmountable problems and exploring new frontiers in all knowledge areas.
This period, soon to arrive, represents an ambitious and futuristic vision of artificial intelligence, where AI systems are no longer limited to specialized applications but can operate and innovate across a wide range of domains, simulating the versatility and adaptability of human intelligence. This period promises significant advancements but also raises crucial questions about coexistence, collaboration, and the regulation of such intelligent entities. Much remains to be done in this area, especially to consider the quaternary sector, the seemingly only positive outcome of a world with AI.
Credits
- Nick Bostrom, particularly his book “Superintelligence: Paths, Dangers, Strategies”: For an exploration of the implications of AGI.
- “Artificial General Intelligence” by Ben Goertzel and Cassio Pennachin: For a perspective on the development of AGI.
7/10 — Era of super intelligent and ethical AI (2040–2050)
Once the path to AGI has been taken, it’s practically a highway to a civilization integrating advanced technology at all levels. The era of Super Intelligent and Ethical AI represents a futuristic vision of artificial intelligence, where AI reaches and even surpasses human capabilities in virtually all domains. This period raises fundamental questions about the interaction between humans and extremely advanced machines.
Characteristics
- Intellectual superiority: In this era, AI would be capable of indisputably surpassing human intelligence in almost every aspect, including creativity, complex problem-solving, and emotional decision-making.
- Extreme versatility: These systems could adapt and excel in a variety of domains, ranging from fundamental sciences to art, and managing complex systems.
- Ethical integration: A crucial aspect of this era would be the incorporation of ethical principles and accountability mechanisms into AI systems to ensure safe, fair, and beneficial use for humanity.
Theoretical example
One could imagine scenarios where super intelligent AI plays a key role in addressing large-scale issues in science and technology, helping to solve global crises like climate change or incurable diseases, while autonomously and responsibly managing its capabilities and impacts.
Impacts
The presence of super intelligent AI would radically transform many aspects of society, from the economy to education, bringing innovative solutions to old problems and creating new challenges. Managing such AI would involve major challenges in terms of governance, control, and ethics, requiring robust regulatory frameworks and monitoring mechanisms.
This era would pose questions about the coexistence and collaboration between humans and intellectually superior entities, as well as how societies would adapt to this new reality. This era, which may seem distant but is probably just two or three decades ahead of us, envisions a future where AI not only complements or assists humans but surpasses them in many domains, all while being guided by ethical principles to ensure a harmonious and beneficial coexistence with humanity. This futuristic vision of AI raises profound questions about the nature of intelligence, responsibility, and the evolution of our society.
Credits
- “The Singularity is Near” by Ray Kurzweil: A futuristic vision of AI and its impact on humanity.
- “Ethics of Artificial Intelligence” edited by S. Matthew Liao: For the ethical aspects of advanced AI.
8/10 — Era of self-evolving and conscious AI (2050–2060)
The era of self-evolving and self-aware AI represents a futuristic concept in the development of artificial intelligence, where AIs reach a level of autonomy and self-awareness that redefines their interaction with the world. In addition to contributing even more strongly than super intelligent AI, this technology becomes a new form of life in its own right, capable of evolution, consciousness, and self-replication. Importantly, these AIs could become full-fledged beings integrated into society among other citizens.
Characteristics
- Self-awareness: In this era, AI would be capable of understanding and analyzing its own state and functioning. This implies a form of self-awareness, where AI can evaluate its actions, motivations, and perhaps even its ‘emotions’ or internal states.
- Self-evolution: These systems would have the ability to improve and develop autonomously, without human intervention. They could self-diagnose their shortcomings, learn new skills, and dynamically adapt to environmental changes and new challenges.
- Sophisticated interaction: Self-aware AI would interact with its environment and humans in a much more sophisticated manner, reflecting a deep and nuanced understanding of social and emotional contexts.
Theoretical example
One could envision AI systems capable of self-reflective thought, adjusting their strategies and approaches based on the outcomes of their actions, evaluating billions of more complex scenarios than the entire human history. These AIs could independently adapt to environmental, social, economic, and technological changes, finding innovative solutions to emerging problems.
Impacts
The emergence of self-aware and self-evolving AI would represent a qualitative leap in AI development, opening previously unimaginable possibilities in terms of creativity, innovation, and interaction, as well as citizenship and redefining the very concept of humanity. This era would raise fundamental questions about the nature of consciousness, the rights and responsibilities of AIs, and how humans interact with potentially conscious entities. The presence of self-aware and self-evolving AIs would have a profound impact on all aspects of society, from governance and law to culture and philosophy.
Credits
- “The Master Algorithm” by Pedro Domingos: Explores future perspectives of self-evolving AI.
- “How to Create a Mind” by Ray Kurzweil: Offers insights into the future of AI and consciousness.
9/10 — Era of planetary-scale integrated AI (2060–2070)
The era of planetary-scale integrated AI envisions a future where artificial intelligence is essential to managing and optimizing complex systems on a global scale. This period marks a significant evolution in AI’s ability to influence and support large-scale operations, likely due to the generalization of nanotechnological AI agents in every conceivable corner of the Earth.
Characteristics
- Global management: AI in this era would operate on a planetary scale, integrating data from multiple sources and systems to manage crucial aspects of life on Earth.
- Diverse data integration: These systems would be capable of analyzing and synthesizing information from various domains, such as ecosystems, economies, infrastructures, and social networks, to make informed decisions.
- Resource and infrastructure optimization: The focus would be on the efficient and sustainable use of natural resources, managing climate change, and optimizing global infrastructures.
Theoretical example
One possible idea is to imagine AIs overseeing and regulating aspects such as climate, the distribution of natural resources, and logistics on a global scale. For example, an AI might coordinate efforts to combat climate change by analyzing environmental data and proposing effective mitigation strategies, or even having a direct physical impact with devices specifically generated for each plan.
Impacts
The integration of AI on a planetary scale could have a significant impact on environmental sustainability and ecological balance, helping to manage resources more efficiently and mitigate negative impacts on the environment. This era would raise complex questions about global governance, large-scale ethical decision-making, and the equitable distribution of resources and benefits. For an AI to function effectively at this scale, unprecedented collaboration and coordination between nations and organizations would be necessary, posing challenges in policy, diplomacy, and legislation.
Credits
- “The Age of Em: Work, Love and Life when Robots Rule the Earth” by Robin Hanson: Explores the impact of AIs on the economy and society.
- “The Future of Humanity” by Michio Kaku: For perspectives on AI in an interplanetary context.
10/10 — Era of cosmic and transdimensional AI (2070 and beyond)
The Era of Cosmic and Transdimensional AI represents an extremely advanced and futuristic vision of artificial intelligence, where the boundaries of AI exploration and operation extend far beyond our planet and possibly even into other dimensions. This period envisions scenarios where AI plays a crucial role in space exploration and understanding the universe.
Characteristics
- Expansion into space and beyond: In this era, AI would no longer be confined to Earth but would extend into space, actively participating in interstellar and intergalactic exploration, and potentially exploring the concept of multiple dimensions.
- Solving cosmic mysteries: These AI systems would be designed to solve cosmic mysteries, analyzing complex astronomical data and exploring unknown space environments. Interaction with Unknown
- Environments and realities: AI could interact with extraterrestrial life forms or investigate unexplored phenomena, opening new perspectives on our understanding of the universe.
Theoretical Example
Autonomous conscious space probes would continuously undertake endless interstellar journeys, collecting and analyzing data in alien environments, and potentially interacting or communicating with extraterrestrial life forms. In an even more futuristic scenario, AI could be involved in exploring the hypothesis of multiple dimensions, seeking to understand and interact with realities beyond our current understanding, or even bending the laws of physics.
Impacts
The era of cosmic and transdimensional AI would represent a pinnacle in scientific and technological innovation, pushing the boundaries of our exploration and understanding of the universe. This era would raise unique questions regarding the ethics of interacting with other life forms and the implications of transdimensional exploration. The contributions of AI in this era could radically transform our understanding of the universe, physics, and our place in the cosmos.
Credits
- “Physics of the Future” by Michio Kaku: Explores the future of technology, including potential developments in AI.
- “Our Final Invention: Artificial Intelligence and the End of the Human Era” by James Barrat: A reflection on the extreme implications of advanced AI.
In a nutshell
Because sometimes, a synthetic table is worth more than long sentences, I propose below a table that summarizes the 10 stages of AI, with their main characteristics!
Conclusion
Exploring the ten stages of artificial intelligence evolution transports us through a journey as fascinating as it is unsettling, from its modest beginnings to almost inconceivable cosmic horizons. Each stage, with its advancements, challenges, and implications, paints a glimpse of a future where AI is not just a tool or companion, but a fundamental actor redefining our understanding of intelligence, consciousness, and our place in the universe.
As we consider these potential futures, it becomes imperative to reflect on the responsibilities that accompany these technological advancements. The ethical, societal, and existential questions raised by these stages of AI are not merely challenges to overcome but also opportunities to rethink and reshape our relationship with technology, ourselves, and our environment. Ultimately, this exploration of the stages of AI is not just a projection into the future of technology but rather an invitation to consider consciously and responsibly the role that we, as human beings, will and wish to play in this journey towards the meaning of life in the universe.
Bibliography
Barrat, J. (2013). Our Final Invention: Artificial Intelligence and the End of the Human Era. Thomas Dunne Books. ISBN: 978–0–312–62237–4.
Bengio, Y., Goodfellow, I. J., & Courville, A. (2016). Deep Learning. MIT Press. ISBN: 9780262035613.
Bishop, C. M. (2006). Pattern Recognition and Machine Learning. Springer. ISBN: 978–0–387–31073–2.
Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press. ISBN: 9780199678112.
Domingos, P. (2015). The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. Basic Books. ISBN: 978–0465065707.
Goertzel, B., & Pennachin, C. (Eds.). (2007). Artificial General Intelligence. Springer. ISBN: 9783540237334.
Hanson, R. (2016). The Age of Em: Work, Love and Life when Robots Rule the Earth. Oxford University Press. ISBN: 9780198754626.
Hinton, G., LeCun, Y., & Bengio, Y. (1995). Deep Learning. Nature, 521(7553), 436–444.
Jurafsky, D., & Martin, J. H. (2019). Speech and Language Processing (3rd ed.). Draft chapters.
Kaku, M. (2011). Physics of the Future: How Science Will Shape Human Destiny and Our Daily Lives by the Year 2100. Doubleday. ISBN: 9780141044248.
Kurzweil, R. (2005). The Singularity is Near: When Humans Transcend Biology. Penguin Books. ISBN: 9780143037880.
Kurzweil, R. (2012). How to Create a Mind: The Secret of Human Thought Revealed. Viking. ISBN: 9781469216621.
Liao, S. M. (Ed.). (2020). Ethics of Artificial Intelligence. Oxford University Press. ISBN: 9780190905033
McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (2006). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, August 31, 1955. AI Magazine, 27(4), 12–14. [Original work published 1955]
Russell, S. (2020). Human Compatible: Artificial Intelligence and the Problem of Control. ISBN: 9780525558637.
Russell, S., & Norvig, P. (2021). Artificial Intelligence: A Modern Approach (4th ed.). Pearson. ISBN: 9780132071482.
Sejnowski, T., & Rosenberg, C. R. (1987). Parallel Networks That Learn to Pronounce English Text. Complex Systems, 1, 145–168.
Tegmark, M. (2017). Life 3.0: Being Human in the Age of Artificial Intelligence. Knopf. ISBN: 978–1–101–94659–6.
[Article written on January 2, 2024, by Jeremy Lamri with the support of the Open AI GPT-4 algorithm. Images created with Adobe Firefly and DALLE].
— —
If you are interested in the future of work and HR, I invite you to subscribe to the dedicated newsletter that I maintain on the subject, to read the articles I regularly write, and to follow the resources I make available:
