4 scenarios for the next 20 years of artificial intelligence
… or how to learn to live with automation and machine learning
“The long AI-winter passed slowly through spring, to its first and possibly final summer.” – unknown AI-chatbot, 2023.
The concept of AI has been around since the early days of human imagination. Earliest mentions of non-human intelligence is in The Brazen Head, a medieval legend from the 13th century, that tells the story of a talking head made of brass that was able to answer any question.
In early modern literature, the notion of intelligence outside of the human brain dates back at least to Samuel Butler’s 1872 novel Erewhon that drew on Darwin among the Machines. The book raised the question of the evolution of consciousness among self-replicating machines that might supplant humans as the dominant species.
To avoid the very long history of science fiction and computer engineering; let’s start with some key historical background followed by some, more or less plausible, scenarios for the next 20 years of AI.
A short history of neural networks and AI
The idea of neural networks began as a model of how neurons in the brain function, termed ‘connectionism’ and used connected circuits to simulate intelligent behavior. In 1943, neurophysiologist Warren McCulloch and mathematician Walter Pitts portrayed this with a simple electrical circuit.
Neural networks have evolved from an academic curiosity into a vast “deep learning” industry. Among the pioneers of modern-day AI Marvin Minsky is a superstar.
Minsky pioneered robotics, telepresence and designed and built some of the first visual scanners with mechanical hands and tactile sensors. His research influenced many subsequent robotic projects. In 1951 he built the first randomly wired neural network learning machine (called SNARC, for Stochastic Neural-Analog Reinforcement Computer), based on reinforcing the synaptic connections.
The proof of AI was initialized through Allen Newell, Cliff Shaw, and Herbert Simon’s, Logic Theorist. The Logic Theorist was a program designed to mimic the problem-solving skills of a human and was funded by RAND Corporation. It was presented at the Dartmouth Summer Research Project on Artificial Intelligence (DSRPAI) hosted by John McCarthy and Marvin Minsky in 1956. Check out Stanford Encyclopedia of Philosophy for more history.
Also see Minsky’s Computation: Finite and Infinite Machines from 1967.
Going mainstream
The ideas of AI were still confined to the corners of academia and the growing field of computer engineering. A year later the concept made its appearance in the public awareness.
With Stanley Kubrick's film "2001: A Space Odyssey" from 1968 the world was introduced to HAL 9000, a sentient computer that serves as the antagonist of the story. HAL's development and eventual malfunctioning is one of the most iconic depictions of AI in popular culture.
Advances in the 21st century led to increased investments, industrial applications, public interest, and more recently government oversight.
The dawn of a new industrial revolution
With decades of computer and network industrialization and commodification we arrive at widespread adoption of high technology, globalization, and massive data models and structures.
The combination of new technologies, global markets and massive amounts of information is the foundation for the coming years of increasing volatility and uncertainty caused by AI.
Recent breakthroughs in AI:
2020:
OpenAI released GPT-3, a language model with 175 billion parameters, setting a new record for the largest language model at the time.
Google's DeepMind developed AlphaFold 2, an AI system that can predict the 3D structure of proteins, which won the CASP13 protein folding competition.
NVIDIA introduced the A100 GPU, which is designed specifically for AI and can deliver up to 20 times more performance than its predecessor.
2021:
OpenAI developed DALL-E, an AI system that can generate images from textual descriptions, such as "a snail made of harp strings."
Google introduced LaMDA, a language model that can converse on any topic, making it more conversational than previous language models.
Microsoft introduced GPT-3 competitor, Microsoft Turing, with 17 billion parameters.
DeepMind developed MuZero, an AI system that can master complex games without any prior knowledge of the rules.
2022:
Facebook AI Research (FAIR) introduced GEAR, an AI system that can learn how to play video games like humans, making it more adaptable and less reliant on data than traditional reinforcement learning approaches.
OpenAI introduced Codex, an AI system that can write code in natural language, making it easier for non-programmers to develop software.
NVIDIA released the Grace CPU, which is designed specifically for AI workloads and can deliver up to 10 times more performance than its predecessor.
ChatGPT is a smaller cousin of GPT-3 customized for chatting. It was created by fine-tuning GPT-3 with supervised and reinforcement learning: humans provided dialogues or rated ChatGPT’s dialogues in order to guide it towards.
Galactica is a language model by Meta AI customised for scientific research.
GitHub introduced its Copilot plugin and Amazon its Codewhisperer for automatic code generation.
The release of Midjourney, Stable Diffusion and other image generators.
2023:
The release of OpenAI GPT4, Microsoft Copilot, VALL-E, and Bing Chat, Google Bard, Adobe Firefly and other commercial plugins, tons of free consumer apps, image generators and writing tools, in addition to updates for Whisper, Stable Diffusion and other open-source projects.
Research shows that AI adoption had more than doubled from 2017 to 2022, though the proportion of organizations using AI had plateaued between 50 and 60 percent for the past few years. Worldwide spending by governments and businesses on AI technology is expected to top USD 500 billion in 2023.
We have some very strange days and interesting times ahead.
Scenarios for the coming AI-season
Short and cold AI-summer
AI continues to improve, leading to increased automation in medical research, manufacturing, IT, marketing, finance, education and service industries. This leads to significant job losses as machines and robots become more capable of performing complicated tasks. Some workers lose jobs and companies face pressure to invest in new technology. However, the increased efficiency and productivity resulting from automation lead to lower costs for consumers and increased profits for some companies.
The potential of asymmetric AI-wars between major global powers and non-state actors that use AI to exploit the vulnerabilities or weaknesses of states cause great concern. AI also enable new forms of warfare such as cyberterrorism or bioterrorism that target civilians or public health. Many warned that such technology poses a risk regarding attribution, deterrence, and accountability in conflict and warfare.
Due to these and many other risks and uncertainties a new agreement for AI was implemented in the United Nations, the STOP-agreement of 2028. Further moratoriums, government regulations, and in some countries outright criminalization of AI stop most applications. Political parties and worker unions freeze much of development of AI for other than government services, such as tax collection, fraud detection, identification, and surveillance.
Read Pause Giant AI Experiments: An Open Letter and the entry for AI Alignement on Wikipedia.
Consequence: Business as usual on steroids with a 10 – 30 percent increased efficiency in some sectors.
Mild AI-summer with scattered showers
Scientific maturation, industrial applications, and market acceptance lead to a steady increase in AI-adoption across all industries. AI is also a source of disruption and conflict. In this scenario, AI technologies are unevenly distributed and exploited by countries, companies, and individuals, leading to increased gaps and tensions. AI also creates new forms of crime and violence, such as deep fakes, autonomous weapons, and digital fraud.
Due to increased energy usage AI also exacerbates existing problems such as climate change and poverty. AI becomes a tool for domination and warfare, as well as for cyberattacks and propaganda. Many people lose trust in AI and resist its adoption and use. Despite its many shortcomings, AI increase the general welfare, educational possibilities, and security of most countries’ populations.
Consequence: A wild economic and social ride with 30 – 60 percent increased efficiency in most sectors.
Warm AI-summer with heavy thunderstorms
A new scientific, industrial, and societal paradigm sweeps across the planet. Governments are replaced by AI and universal basic income is introduced in most countries. Exponential growth of AI is only constrained by energy and market saturation. Most industries are fully automated, and people retire after graduation in prompt-engineering. Life also available as a neural implant.
AI technology is increasingly integrated into military operations, leading to significant advancements in military capabilities. This improves efficiency and reduce the number of casualties in military operations. Most wars between states are run in simulations using military digital twins.
Some local AI-wars are asymmetric between major powers and a wide variety of non-state actors including militant neo-luddites, political terrorists, radical environmentalists, ethnic warrior tribes, bio-engineered superhumans and renegade private armies. All of them use AI to exploit vulnerabilities and weaknesses of states, companies and other actors.
AI enabled new forms of warfare such as cyberterrorism or bioterrorism that target civilians or public health. Additionally, the development of AI-powered military capabilities leads to an arms race between countries, increasing global tensions and instability. Some states run military AI on other planets.
Consequence: Utter chaos and high risk of global annihilation with 60 – 90 percent increased efficiency in all sectors.
Indian AI-summer and early fall
This trajectory will lead to the digitalization of all biological species into physical dynamic cubes of computing power competing for energy on a grid covering earth. This is the late summer and early AI-fall. Check out This is How You Lose the Time War.
The scenario consists of a full-blown Singularity/AGI with a long tail of nano/bio/syn-tech revolution(s) that transform the universe into alien structures and dimensions. The unknowns and uncertainties are of immense proportions and make these worlds incomprehensible to humans.
Consequence: Complete transformation of humanity and cosmos, 100 percent increased efficiency.
###
PS. Read Olof Johanneson, Ray Kurzweil, Eliezer Yudkowsky, Nick Bostrom, Robin Hanson, Kevin Warwick, Max Tegmark, and many other authors and futurists for more short and long term AI-scenarios.
Disclaimer: parts of this text is written with the support of ChatGPT and Bing Chat.
Visit The Plausible Futures Newsletter on Twitter for AI-news and updates.
I liked your outlook .. the last few sentences especially . It’s all gonna go hyperexpinential
I worry about the World my grandkids get to inherit
Thanks for your analysis , Olaf.
My spec fic revolves around such topics .