Elena' s AI Blog

AI Evolution to 2025

01 Jun 2022 / 12 minutes to read

Elena Daehnhardt


Midjourney, May 2023


I keep getting asked the same question lately: “How did we end up with AI that can write emails, create art, and have conversations?” It happened while I was explaining ChatGPT to my friend last week. She was amazed but also confused about how we got here so fast.

The truth is, it wasn’t fast at all. This journey started decades ago with some pretty wild ideas that seemed impossible at the time.

Since AI is what I write about most on this blog, I maintain a timeline of the major breakthroughs. I covered this in my 2023 post, but things have moved so quickly that it’s time for an update.

1943
First conceptualization of artificial neural networks
McCulloch & Pitts Publish the First Mathematical Model of a Neural Network
1950
Alan Turing proposes the "Turing Test"
The Turing Test
1956
Dartmouth Workshop marks the birth of AI as a field, with John McCarthy coining the term "artificial intelligence."
Artificial Intelligence Coined at Dartmouth
1958
Perceptron, the first working neural network, is developed by Frank Rosenblatt
Professor’s perceptron paved the way for AI – 60 years too soon
1969
Shakey, the first mobile robot capable of reasoning, navigation, and manipulation, is developed
Shakey the Robot
1972
The first expert system, MYCIN, is developed for diagnosing infectious diseases.
MYCIN
1997
IBM's Deep Blue defeats chess champion Garry Kasparov, signalling a significant milestone in machine learning and AI
Deep Blue
2006
Geoffrey Hinton and colleagues wrote their paper on backpropagation and further developed deep learning concepts, reviving interest in neural networks.
Heroes of Deep Learning: Geoffrey Hinton
2011
IBM's Watson wins Jeopardy! against human champions, demonstrating advancements in natural language processing.
A Computer Called Watson
2012
AlexNet, a deep convolutional neural network, achieves a breakthrough in image classification accuracy.
ImageNet Classification with Deep Convolutional Neural Networks
2016
DeepMind's AlphaGo defeats Go world champion Lee Sedol, showcasing the power of AI in complex strategy games.
Human Go champion loses to Google DeepMind AlphaGo computer in 1st game
2017
AlphaZero, developed by DeepMind, achieves superhuman performance in chess, shogi, and Go without prior human knowledge.
AlphaZero: Shedding new light on chess, shogi, and Go
2018
OpenAI's GPT-1 (Generative Pre-trained Transformer) 117 million parameters demonstrates large-scale language modeling and text generation capabilities.
GPT-1 to GPT-4: Each of OpenAI's GPT Models Explained and Compared
2019
GPT-2 (1.5 billion parameters), a more powerful language model than GPT-1, demonstrates the ability to generate coherent and contextually relevant text.
GPT-1 to GPT-4: Each of OpenAI's GPT Models Explained and Compared
Rise of transformer architectures becoming the dominant approach in natural language processing.
Transformer: A Novel Neural Network Architecture for Language Understanding and Beyond
2020
GPT-3, an even more advanced language model (175 billion parameters), sets new benchmarks in natural language processing and generates highly realistic text.
GPT-1 to GPT-4: Each of OpenAI's GPT Models Explained and Compared
Increased research and development in self-supervised learning techniques, reducing the reliance on large labeled datasets.
Self-supervised learning: The dark matter of intelligence
2021
Continued scaling of large language models (LLMs) with models featuring trillions of parameters being explored.
Beyond GPT-3: Google Brain shows off 1.6 trillion parameter language model
Growing focus on AI ethics and responsible AI development, including fairness, transparency, and accountability.
AI Ethics Guidelines Global Inventory
2022
AI-powered virtual assistants have become increasingly integrated into everyday life, assisting with tasks and providing personalized recommendations.
The Rise Of Virtual Personal Assistants: How They’re Changing The Way We Work And Live
Advancements in diffusion models lead to significant progress in generating high-quality images and other multimedia content.
What are Diffusion Models?
2023
GPT-4, the most advanced model to date, demonstrates multimodal capabilities, accepting both images and text as input (number of parameters is in the trillions).
GPT-1 to GPT-4: Each of OpenAI's GPT Models Explained and Compared
Increased adoption of AI in scientific discovery, accelerating research in fields like medicine and materials science.
Artificial intelligence for science
2024
Widespread adoption of generative AI models for content creation, including text, images, audio, and video, impacting various industries.
The rise of generative AI and its implications for the future of work
Emergence of more sophisticated AI agents capable of planning and executing complex tasks autonomously.
What are AI Agents? The Next Evolution of AI
2024
Continued advancements in AI ethics and governance frameworks to address challenges related to bias, transparency, and accountability.
AI Governance
Growing focus on developing more efficient and sustainable AI models and hardware.
Sustainable AI: The Quest for Energy-Efficient Artificial Intelligence
2025
Increased focus on energy efficiency and sustainability in AI development and deployment due to the growing computational demands of large models.
The growing energy footprint of artificial intelligence
Further development and deployment of AI in robotics, leading to more autonomous and intelligent robots in various sectors.
Key trends driving robotics adoption in 2025

The Beginning: Math Meets Biology

Back in 1943, Warren McCulloch and Walter Pitts proposed the first mathematical model of a neuron in McCulloch & Pitts Publish the First Mathematical Model of a Neural Network. They were doing this while most people were focused on World War II.

Here’s what blows my mind about this: they didn’t even have real computers yet. The most advanced machine was the Colossus Mark 1 in Britain—a room-sized code-breaker full of vacuum tubes.

Alan Turing came along in 1950 with The Turing Test. His simple question—if a machine can fool you into thinking it’s human, is it intelligent?—still sparks debate.

The 1956 Dartmouth Workshop officially named the field during Artificial Intelligence Coined at Dartmouth.

1957 brought Frank Rosenblatt’s perceptron, chronicled in Professor’s Perceptron Paved the Way for AI – 60 Years Too Soon. For the first time, a neural net could learn from examples.

Early Experiments: The Art of Fooling Humans

Joseph Weizenbaum’s therapist parody ELIZA (1966) hinted at our tendency to anthropomorphize software.

1969 gave us Shakey the Robot, the first mobile bot that could reason about its environment.

Medical AI started in 1972 with MYCIN, which diagnosed blood infections better than many doctors—but no one trusted a black-box system.

When AI Started Beating Us

IBM’s Deep Blue defeated Garry Kasparov in 1997, proving brute-force search could master chess.

The deep-learning renaissance began in 2006 (see Heroes of Deep Learning: Geoffrey Hinton), when new training tricks made very deep neural nets practical.

2011 was Watson’s year—Watson, ‘Jeopardy!’ Champion showed NLP could handle puns and pop culture.

ImageNet Classification with Deep Convolutional Neural Networks (aka AlexNet, 2012) kicked off today’s computer-vision boom.

The Moment Everything Changed

2016: Artificial Intelligence: Google’s AlphaGo Beats Go Master Lee Se-dol. Creativity and intuition were no longer uniquely human.

2017: AlphaZero: Shedding New Light on Chess, Shogi, and Go trained itself from scratch and rewrote centuries of strategy.

Language Models Change Everything

The transformer exploded onto the scene with Transformer: A Novel Neural Network Architecture for Language Understanding.

GPT-3’s debut paper, Language Models Are Few-Shot Learners, in 2020 showed 175 billion parameters can generate eerily human-like prose.

Around the same time, Self-Supervised Learning let models learn from raw data without labels.

Where We Are Now

Diffusion models hit prime time in 2022 thanks to What Are Diffusion Models?. 2023 brought multimodal GPT-4; 2024 – 2025 made gen-AI mainstream. For labor impacts, see Generative AI, the American Worker, and the Future of Work.

At the same time we’re debating ethics (AI Ethics Guidelines Global Inventory) and carbon footprints (Generative AI’s Energy Problem Today Is Foundational).

Robotics is racing ahead too—see the International Federation of Robotics’ TOP 5 Global Robotics Trends 2025.

What Comes Next

Integrating these brains with physical bodies excites me most. Every milestone shows human creativity turning science fiction into Tuesday morning routine. What’s the next “impossible” thing to become ordinary? Stay tuned and subscribe if not subscribed yet :)

What do you think was the most significant milestone? Let me know!

References

  1. McCulloch & Pitts Publish the First Mathematical Model of a Neural Network
  2. The Turing Test
  3. Artificial Intelligence Coined at Dartmouth
  4. Professor’s Perceptron Paved the Way for AI – 60 Years Too Soon
  5. Shakey the Robot
  6. MYCIN – Clinfowiki
  7. [Deep Blue IBM](https://www.ibm.com/history/deep-blue)
  8. Heroes of Deep Learning: Geoffrey Hinton
  9. Watson, ‘Jeopardy!’ Champion
  10. ImageNet Classification with Deep Convolutional Neural Networks
  11. Artificial Intelligence: Google’s AlphaGo Beats Go Master Lee Se-dol
  12. AlphaZero: Shedding New Light on Chess, Shogi, and Go
  13. Transformer: A Novel Neural Network Architecture for Language Understanding
  14. Language Models Are Few-Shot Learners
  15. Self-Supervised Learning
  16. AI Ethics Guidelines Global Inventory
  17. What Are Diffusion Models?
  18. Generative AI, the American Worker, and the Future of Work
  19. [AI Policy and Governance The 2022 AI Index Report](https://hai.stanford.edu/ai-index/2022-ai-index-report/ai-policy-and-governance)
  20. Generative AI’s Energy Problem Today Is Foundational
  21. TOP 5 Global Robotics Trends 2025
desktop bg dark

About Elena

Elena, a PhD in Computer Science, simplifies AI concepts and helps you use machine learning.

Citation
Elena Daehnhardt. (2022) 'AI Evolution to 2025', daehnhardt.com, 01 June 2022. Available at: https://daehnhardt.com/newest
All Posts