The Thermodynamic Revolution in AI

Why Embracing Randomness Could Change Everything

What if the future of AI isn’t about building bigger and faster machines but about rethinking how we compute?

For decades, we’ve leaned on Moore’s Law, trusting that processing power would double every two years. And it has—until now. Transistors, the building blocks of computers, are now so small they’re almost atomic in scale, and at that level, things get unpredictable. Thermal noise—tiny, random energy shifts—begins to disrupt calculations, leading to greater heat and power demands. Today’s high-powered AI is even driving companies to explore nuclear-powered data centres to keep up with demand. So, is there a smarter way to evolve AI that doesn’t involve brute force?

Possibly. It might come from the most efficient computing system we know: the human brain.

“Randomness is not a bug; it’s a feature.” — Nassim Taleb, Fooled by Randomness

Let’s dive into how embracing randomness, probability, and thermodynamics might give AI its next big breakthrough.

The Scaling Crisis: Why Traditional Computing Needs an Upgrade

Digital computers work with bits—strict 0s and 1s. But as we continue cramming more transistors onto chips, the limits of this approach become clear. Thermal noise and other atomic-level forces interfere with our carefully designed systems, causing instability. This scaling crisis means we can’t rely on brute-force processing power alone to advance AI. We need a paradigm shift.

Traditional Solutions to Increasing Compute and Managing Energy Demand

As AI models grow in complexity, so does the demand for computing power. Companies like Alphabet (Google) and Nvidia have been leading the charge to develop more powerful hardware and optimise energy use, but each of these approaches faces limitations:

1. Graphics Processing Units (GPUs): Nvidia, for example, has pioneered GPUs, which excel in parallel processing and are widely used in AI training. GPUs are highly efficient for handling large datasets, but as models grow, so does the energy demand, pushing the limits of what even advanced GPUs can sustainably manage.
2. Tensor Processing Units (TPUs): Developed by Alphabet’s Google, TPUs are custom-built for machine learning tasks and offer efficiency improvements over traditional processors. However, even TPUs struggle to keep pace with the exponential growth in AI compute needs without generating substantial heat and requiring significant energy.
3. Data Centre Optimisation: Companies have invested heavily in cooling systems and more energy-efficient data centres. Techniques like liquid cooling and AI-based workload management help reduce energy consumption, but they only go so far. As demand grows, traditional cooling methods struggle to keep data centres efficient, driving up both financial and environmental costs.
4. Renewable Energy Sources: To offset these demands, tech giants like Google and Amazon are investing in renewable energy sources for their data centres. While this reduces carbon footprints, it doesn’t fully address the need for new, energy-efficient computing architectures to reduce overall energy consumption.

These methods represent incremental improvements, but they ultimately bump up against physical and environmental limits. This is where thermodynamic computing and probabilistic AI enter the picture as a way to rethink computing from the ground up. By harnessing randomness and natural energy fluctuations, thermodynamic computing could be a game-changer, potentially offering a path forward that aligns with sustainability goals and efficiency needs.

Learning from Nature: The Brain’s Efficient Randomness

Think of your brain. Despite weighing only about 1.5 kilograms, it uses around 20 watts of power—barely enough to power a light bulb. Yet, it performs complex tasks like decision-making, pattern recognition, and problem-solving. Supercomputers need millions of times more energy for similar feats. The key lies in how the brain leverages randomness and probability, adapting to its environment with incredible efficiency.

Unlike traditional computers, the brain doesn’t need precise, rigid calculations for everything. Instead, it uses randomness and probability to “guess” and adapt, making decisions even with incomplete data. This flexible, energy-efficient approach inspires thermodynamic computing, where randomness becomes a resource rather than a challenge.

Breaking Down Randomness and Probability: Game-Changers for AI

Here’s why randomness and probability are so powerful for AI:

1. Exploring Many Solutions Quickly: Randomness allows systems to try multiple approaches at once, saving time and energy. Imagine solving a maze by testing random paths until the best one is found. Faster, more efficient, and ideal for problem-solving.
2. Handling Uncertainty: Probability enables systems to make informed guesses with incomplete data. For example, Google Maps uses probability models to predict traffic based on patterns and live data, choosing the most likely “best route” based on real-time conditions. A thermodynamic AI could take this further by processing even more variables and dynamically predicting congestion before it happens.
3. Energy Efficiency: Embracing randomness reduces the need for tight control, which is energy-intensive. A thermodynamic approach allows systems to adapt naturally, using less power for similar or even better outcomes.

“Nature is the best problem-solver. It doesn’t aim for perfection; it aims for efficiency.” — Gill Verdon

Thermodynamic Computing: Redefining AI from the Ground Up

Thermodynamic computing turns traditional logic on its head. Instead of precise 0s and 1s, thermodynamic systems use stochastic units—values that change based on natural energy levels. These systems are unpredictable, yet that unpredictability is powerful. Just as water finds the lowest point to flow downhill, stochastic systems find optimal solutions without strict control. This design allows thermodynamic computing to tackle complex tasks more naturally, using randomness to explore a broad range of solutions while requiring less energy.

Real-World Applications: Smarter and More Efficient AI

What could the future look like if AI systems leaned into thermodynamics and probability? Here are a few examples:

1. Google Maps and Traffic Prediction: Google Maps uses probabilistic models to predict traffic by analysing patterns and live data. Thermodynamic computing could enable it to process more factors at once, potentially predicting traffic congestion before it even starts.
2. Streaming Recommendations (Netflix, Spotify): Today’s recommendation engines make suggestions based on your history. With probability-driven models, thermodynamic AI could predict your preferences even without direct data, making creative, spot-on suggestions from minimal input.
3. Healthcare Diagnostics: Diagnosing diseases often requires spotting patterns in complex data. Thermodynamic AI could help “fill in the blanks” with probabilistic models, improving diagnostic accuracy and detecting subtle signs of illness.
4. Weather Forecasting: Weather is inherently chaotic, with countless variables. Thermodynamic AI, using random sampling and probabilistic predictions, could handle the complexity better, leading to more reliable long-term forecasts.
5. Financial Market Predictions: Stock markets are unpredictable and influenced by countless factors. Thermodynamic AI could use random fluctuations to predict trends in real time, leading to smarter investment strategies.

Quantitative AI Models: The Next Frontier in Intelligent Systems

So far, most AI models rely heavily on large datasets sourced from the internet, which means they are prone to biases and limits inherent in human language and online data. But an emerging approach, Quantitative AI Models (LQMs), offers a different path. Instead of being trained on internet data, LQMs use datasets drawn from real-world equations and scientific laws in biology, physics, and chemistry. Imagine an AI model that doesn’t just “know” language but understands the fundamental principles governing electrons, molecules, and natural forces.

LQMs could complement current large language models (LLMs) by bringing a depth of understanding rooted in the physical world. For example, by combining LQMs with thermodynamic computing, AI could run more efficiently and with greater accuracy, using GPUs from Nvidia and TPUs from Alphabet to process both types of models on the same infrastructure. In a future issue, we’ll explore how LQMs and generative data—data created by simulations of real-world equations—could revolutionise AI, bringing a level of understanding far deeper than current LLMs can achieve.

Ethical Considerations: The Power—and Responsibility—of Probabilistic AI

As with any powerful technology, thermodynamic AI raises ethical questions. Here are a few considerations:

1. Reliability in Decision-Making: Probabilistic AI inherently operates in uncertainties. If it’s used in critical fields like healthcare or law, how do we ensure its “guesses” are fair and reliable? Decision-making processes must be transparent and rigorously tested to avoid unintended consequences.
2. Data Privacy and Bias: Models like LQMs that pull from “clean” data—data free from human biases in language or opinion—may offer a more objective perspective. However, even physical models can reflect biases if improperly applied. Careful oversight will be needed to prevent AI from reinforcing inequalities.
3. Environmental Impact: The energy demands of high-powered AI raise serious sustainability questions. Thermodynamic computing offers an energy-efficient alternative, but ongoing innovation will be required to keep up with the world’s growing computing needs.
4. Unpredictable Behaviour: Embracing randomness can lead to unpredictable outcomes. Probabilistic AI may produce solutions that seem “out of left field”—great for creativity but potentially problematic in fields that demand accountability.
5. Human-AI Collaboration: As AI takes on more probabilistic reasoning, it will be critical to define how humans and machines work together in decision-making, especially in fields where trust and empathy are key.

The Future of AI: Rethinking What’s Possible

By drawing inspiration from the human brain’s efficiency, randomness, and probabilistic thinking, AI could become more than a powerful tool—it could become a flexible, sustainable partner in addressing humanity’s greatest challenges. This isn’t just about making machines faster. It’s about making machines that are in harmony with the laws of nature, efficient enough to reduce their environmental impact, and adaptive enough to handle the unpredictability of real life.

In future issues, we’ll dive deeper into Quantitative AI Models (LQMs), generative data, and the impact of using physics, biology, and chemistry to teach AI how the world works at a fundamental level. With LQMs and thermodynamic computing, we may reach a point where AI doesn’t just generate text or images but also generates insights, predictions, and solutions based on nature’s own equations. Think Heisenberg, Schr?dinger, and beyond. Please read my newsletter “Quantum Computing: Your Brain's Next Big Workout” to learn more.

The next era of AI won’t be about overpowering nature; it will be about learning from it. By redefining how AI thinks, we’re expanding not only the boundaries of technology but also our understanding of intelligence itself.

-Kevin

Thanks for reading Ethics and Algorithms! This post is public so feel free to share it.

References

1. Taleb, N. N. (2004). Fooled by Randomness: The Hidden Role of Chance in Life and in the Markets. Random House.
2. Verdon, G. (2023). [Quote on nature’s efficiency].
3. Hypertextbook.com. (2001). The Human Brain Energy Consumption. Retrieved from https://hypertextbook.com/facts/2001/JacquelineLing.shtml
4. NIST.gov. (2021). Brain-Inspired Computing: Creating Faster, More Energy-Efficient AI. Retrieved from https://www.nist.gov/blogs/taking-measure/brain-inspired-computing-can-help-us-create-faster-more-energy-efficient

Empowering You to Thrive in the New Technology Revolution

I’m on a mission to equip you with the insights and strategies to harness the transformative power of technology—a revolution I believe will have an even greater impact on the world than the internet, the personal computer, and the smartphone combined. Through my newsletters "Ethics and Algorithms" and "Baker on Business," I deliver forward-thinking perspectives that help you unlock unprecedented opportunities, navigate complex ethical landscapes, and position yourself for wild success in a rapidly changing world.

In the next issue, I will ask you to consider how to help with my research, tools, and time for 2025.

By supporting this mission, you’re not only investing in your own potential but joining a movement that empowers leaders to shape a responsible, impactful future. Together, let’s turn challenges into breakthroughs and opportunities into lasting success.

Be part of this revolution—your support makes it possible. Please link, share, and comment so more people will receive this in their feed.

I am a business and technology writer, consultant, C-level business executive, adjunct academic, former social entrepreneur, and long time self-taught tech geek.

-I am a professional board member with a Certificate in Governance Practice, Governance Institute of Australia; Issued Feb 2024 Credential ID 158584

-I founded Kevin Baker Consulting in 2012. With a rich background spanning philosophy, technology, and global business, I bring a unique international perspective to the evolving dialogue on the business of ethics and technology.

-You can view links to my website, newsletters, podcast, and social media by clicking here. (Link Tree).

Kevin Baker MASTERMIND ADVISORY GROUPS forming in January 2025. Learn more here.

Stay Connected

If you found this article thought-provoking please like, share, or comment so more people will see this. Please like my social media pages, and consider subscribing to "Ethics and Algorithms" for more insights at the intersection of ethics, technology, and personal growth.

Thanks for reading Ethics and Algorithms! Subscribe for free in the comments to receive this by email, receive new posts, and support my work.