Chapter 22 — AI & Robotics, 1926–2026: Computing to Machine Creativity

The Thinking Machine

In 1926, the word "robot" was only six years old—coined by Karel Čapek in his 1920 play R.U.R. (Rossum's Universal Robots). Artificial intelligence as a field didn't exist; the term wouldn't be invented for another thirty years. Computers were human beings—women, mostly—who performed calculations by hand. The machines that would eventually think were unimaginable.

A century later, AI systems write code, compose music, generate images, diagnose diseases, and engage in conversations indistinguishable from human speech. Robots build cars, sort packages, vacuum floors, and perform surgery. The dream of thinking machines—once the province of science fiction—has become daily reality for billions of people.

The transformation happened in stages: first mechanical calculation, then electronic computation, then programmed intelligence, then machine learning, and finally—in just the past few years—foundation models that exhibit general-purpose capabilities no one quite predicted. Each stage seemed revolutionary; each was surpassed by what came next.

This chapter traces that century of development, from vacuum tubes to transformers, from chess-playing programs to systems that pass bar exams. Understanding this history provides essential context for understanding what comes next—and why the next decade may see changes as dramatic as any in this remarkable century.

2026 Snapshot — The AI and Robotics Landscape

Artificial Intelligence

Foundation models dominate the landscape:

Large language models (GPT-4, Claude, Gemini, Llama, Mistral) demonstrate broad capabilities: writing, coding, analysis, reasoning, multilingual translation
Multimodal models process text, images, audio, and video in unified systems
Agent systems execute multi-step tasks: browsing, coding, operating software
Capabilities continue advancing: Each generation shows meaningful improvements; the trajectory is steep

Deployment is widespread:

Millions use AI assistants daily for writing, research, and creative work
Enterprises deploy AI for customer service, coding assistance, document processing
AI aids medical diagnosis, legal research, scientific literature review
Code generation tools (GitHub Copilot, Claude) transform software development

Limitations persist:

Hallucination: Models confidently state falsehoods
Reliability: Performance varies unpredictably
Reasoning: Complex multi-step reasoning remains inconsistent
Alignment: Ensuring AI does what users intend is an ongoing challenge

Robotics

Industrial robotics is mature:

3.5+ million industrial robots operating worldwide
Automotive, electronics, and logistics are the largest sectors
Collaborative robots ("cobots") work alongside humans
China is now the largest market and manufacturer

Logistics automation has accelerated:

Amazon operates 750,000+ robots across fulfillment centers
Autonomous mobile robots (AMRs) navigate warehouses
Picking and sorting are increasingly automated

Consumer and service robotics remains limited:

Vacuum robots are widespread; other home robots less successful
Commercial cleaning and security robots deploy at modest scale
Hospitality and retail robots have had mixed reception

Humanoid robotics is emerging:

Tesla's Optimus, Figure 01, Agility's Digit, 1X's Neo in development/early testing
Boston Dynamics' Atlas demonstrates remarkable mobility
Commercial deployment is nascent but accelerating

The Convergence

AI + Robotics integration is the frontier:

AI provides perception, planning, and decision-making for robots
Foundation models could enable robots to generalize across tasks
The promise: robots that can learn from demonstration and adapt to novel situations
The reality: Still early; most deployed robots remain specialized

Notable Players

AI Labs

OpenAI: Founded 2015 as nonprofit; became for-profit 2019; partnership with Microsoft. GPT series; ChatGPT reached 100 million users in two months. Industry leader in large language models.

Anthropic: Founded 2021 by former OpenAI researchers focusing on AI safety. Claude models; Constitutional AI approach. Significant investment from Google and others.

Google DeepMind: Merger of Google Brain and DeepMind (2023). AlphaFold, Gemini models, broad research portfolio. Integrated into Google products.

Meta AI: Open-source strategy with Llama models. Significant research in computer vision, embodied AI, and language. Powers Meta products.

Mistral: French startup; raised significant funding quickly. Open and commercial models; European champion.

xAI: Elon Musk's AI company; developing Grok. Integrated with X (Twitter) platform.

Chinese labs: Baidu (ERNIE), Alibaba (Qwen), ByteDance, Moonshot AI, 01.AI (Yi). Rapid development; focus on Chinese language and domestic deployment.

Compute and Hardware

NVIDIA: Dominant in AI training and inference hardware. GPUs (H100, Blackwell) are the standard for AI training. Market capitalization exceeded $1 trillion.

AMD: Competing with MI300 series; gaining market share.

Google (TPUs): Custom tensor processing units for Google's AI workloads.

Amazon (Trainium): Custom chips for AWS AI services.

Microsoft (Maia): Developing custom AI accelerators for Azure.

AI chip startups: Cerebras, Groq, SambaNova, Graphcore—various architectural approaches.

Cloud Providers

Microsoft Azure: Exclusive OpenAI partnership; Copilot integration across products.

AWS: Broad AI services; Bedrock platform for foundation models.

Google Cloud: Vertex AI platform; Gemini integration.

Industrial Robotics

Fanuc, ABB, KUKA, Yaskawa: Traditional industrial robot leaders (the "Big Four").

Universal Robots: Pioneer of collaborative robots; now part of Teradyne.

Chinese manufacturers: Siasun, EFORT, others growing rapidly.

Logistics and Warehouse

Amazon Robotics: Largest deployment; acquired Kiva Systems 2012.

Locus Robotics, 6 River Systems (Shopify), Fetch (Zebra): AMR providers.

Symbotic: Walmart-backed warehouse automation.

Humanoid Robotics

Tesla (Optimus): Humanoid robot leveraging Tesla's AI and manufacturing.

Figure: Backed by OpenAI, Microsoft, NVIDIA; developing general-purpose humanoid.

Agility Robotics (Digit): Bipedal robot for logistics; deploying with Amazon.

1X Technologies: Norwegian company; NEO humanoid; OpenAI-backed.

Boston Dynamics: Research leader; Atlas humanoid; Spot quadruped commercial.

Sanctuary AI: Canadian company; Phoenix humanoid with focus on hands.

The Century in AI and Robotics: A Brief History

The Mechanical Age: Before Computers

Calculation aids preceded computers:

Slide rules, mechanical calculators, adding machines
Charles Babbage's Analytical Engine (1837 design)—programmable but never completed
Punch card tabulation (Herman Hollerith, 1890 census)

Automata preceded robots:

Clockwork devices mimicking animals and humans
Industrial automation: looms, assembly lines

The conceptual foundation was laid:

Boolean logic (1854)
Mathematical logic (Frege, Russell, Whitehead)
Turing's computational theory (1936)—defining what machines can compute

Electronic Computers: 1940s–1950s

World War II accelerated computing:

Colossus (1943): British codebreaking
ENIAC (1945): American general-purpose computer
Vacuum tubes, room-sized, thousands of components

Early AI dreams emerged:

Turing's "Computing Machinery and Intelligence" (1950): Can machines think?
Turing Test proposed as measure of machine intelligence

The Dartmouth Conference (1956): Term "artificial intelligence" coined. Optimistic predictions: "Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it."¹

The First AI Era: 1956–1974

Symbolic AI dominated:

Programs that manipulated symbols according to rules
Logic theorem provers, game playing, early natural language
General Problem Solver (Newell and Simon)

Early success and hype:

Chess programs improved
ELIZA (1966): Simple chatbot that fooled some users
Predictions of human-level AI within 20 years

Limitations became clear:

Programs were brittle—working only in narrow domains
Common sense knowledge proved impossible to encode
Progress slowed; funding dried up

First AI Winter (1974–1980): Reduced funding, skepticism about the field.

Expert Systems and Second Wave: 1980s

Expert systems: Programs encoding human expert knowledge:

MYCIN (medical diagnosis)
XCON (computer configuration)
Commercial deployments in industry

Renewed investment: DARPA, corporations, Japan's Fifth Generation project.

Hardware advances: Personal computers; graphical interfaces; networking.

Robotics advances: Industrial robots became standard in manufacturing.

Limitations again:

Knowledge engineering expensive and brittle
Systems couldn't learn or adapt
Second AI Winter (late 1980s–1990s)

Machine Learning Emerges: 1990s–2000s

The shift: Rather than programming rules, let machines learn from data.

Key techniques:

Neural networks revived (backpropagation)
Support vector machines
Statistical approaches to NLP
Reinforcement learning

Practical applications:

Speech recognition improved
Spam filtering
Recommendation systems
Search engines (Google)

Computing power grew:

Moore's Law continued
Internet generated vast data
The web became a training ground

Deep Learning Revolution: 2010s

The breakthrough: Deep neural networks with many layers, trained on massive data with GPUs.

AlexNet (2012): Deep convolutional network crushed ImageNet competition, igniting the deep learning revolution.²

Rapid advances:

Image recognition surpassed human performance
Speech recognition became practical (Siri, Alexa)
Machine translation transformed (Google Translate)
Game playing: AlphaGo defeated world champion (2016)

Investment exploded: Tech giants poured billions into AI research.

Foundation Models: 2017–Present

The Transformer architecture (2017): Attention mechanisms enabling parallel processing of sequences; basis for all major language models.³

GPT series:

GPT-1 (2018): Demonstrated generative pre-training
GPT-2 (2019): Surprisingly coherent text generation
GPT-3 (2020): 175 billion parameters; few-shot learning; "unreasonable effectiveness"
GPT-4 (2023): Multimodal; passes professional exams; remarkable breadth

ChatGPT (November 2022): Conversational interface reached 100 million users in two months—fastest adoption of any technology ever.

The capability explosion:

Coding assistance
Medical licensing exams passed
Legal reasoning
Scientific literature synthesis
Creative writing and art generation

Open models (Llama, Mistral) and competitors (Claude, Gemini, Grok) proliferated.

Robotics Timeline

Industrial robots (1960s–present):

Unimate (1961): First industrial robot
Programmable arms for welding, painting, assembly
Japan became robotics leader in 1980s
China now largest market

Mobile robots (1990s–present):

Roomba (2002): First successful consumer robot
Military robots for bomb disposal
Warehouse automation (Kiva/Amazon)

Humanoids (2000s–present):

Honda's ASIMO (2000): Iconic humanoid
Boston Dynamics' Atlas: Athletic performance
Current wave: Commercial humanoids for work

Modern Bottlenecks

AI Limitations

Hallucination: Models generate plausible but false information. Reducing this without losing creativity is unsolved.

Reliability: Performance varies across inputs. Consistency for production systems is challenging.

Reasoning: Complex multi-step reasoning, especially with novel problems, remains inconsistent.

World models: Current models may lack deep understanding of how the world works; debate continues about whether this limits capability.

Energy and compute: Training large models requires enormous energy; inference at scale is expensive.

Robotics Limitations

Dexterity: Manipulating objects with human-like skill is difficult. Picking varied items remains a challenge (see Chapter 15 on warehouse automation).

Perception in unstructured environments: Labs are controlled; the real world is messy.

Power and battery life: Mobile robots have limited operating time.

Cost: Advanced robots are expensive; labor may be cheaper for many tasks.

Safety: Robots working near humans must be safe; certification is complex.

Integration Challenges

Hardware-software gap: AI advances faster than robot hardware can incorporate them.

Training data: Physical robots can't generate training data as easily as software can.

Simulation to reality: Models trained in simulation often fail in the real world ("sim-to-real" gap).

Deployment friction: Deploying robots in diverse real environments is hard.

The AI Acceleration Factor

This is the core thesis of this book: AI accelerates everything, including AI itself.

AI improves AI:

AI helps design better AI architectures
AI generates training data
AI finds and fixes bugs in AI systems
AI writes papers about AI

The feedback loop:

Better AI → faster research → better AI
More capable AI → more useful applications → more investment → more capable AI

Already visible:

AI-assisted coding accelerates AI development
AI literature synthesis helps researchers find relevant work
AI helps run experiments and analyze results

The question for this section: How does AI transform AI and robotics themselves, and what does this mean for the next decade?

Looking Forward

The following chapters explore the transformations ahead:

Chapter 23 examines generalist AI—agents that don't just answer questions but execute complex tasks across domains. The shift from copilots to colleagues.

Chapter 24 tackles robots doing dangerous, dirty, dull, and distant work—construction, mining, disaster response, and space. The physical frontier.

Chapter 25 explores androids—humanoid robots approaching human form and function. When machines become indistinguishable from people.

Chapter 26 confronts android warfare—autonomous weapons, drone swarms, and the transformation of conflict. The gravest risks.

Chapter 27 addresses AI security, alignment, and the race for control—ensuring AI remains beneficial as capabilities grow. The essential safeguard.

The century from Turing's paper to GPT-4 established that machine intelligence is possible. The next decade will determine what that intelligence does—whether AI becomes humanity's most powerful tool or its most dangerous creation, whether robots liberate or displace, whether the transformation is managed or chaotic.

That determination is not predestined. It depends on choices being made now.

Endnotes — Chapter 22

The Dartmouth Conference proposal (1955) by McCarthy, Minsky, Rochester, and Shannon included remarkably optimistic predictions. The two-month summer project proposed to make significant progress on machine intelligence.
AlexNet (Krizhevsky, Sutskever, Hinton) won the 2012 ImageNet Large Scale Visual Recognition Challenge, reducing error rates dramatically and sparking the deep learning revolution.
"Attention Is All You Need" (Vaswani et al., 2017) introduced the Transformer architecture, which became the foundation for GPT, BERT, and essentially all modern large language models.
ChatGPT user growth statistics from various sources; reached estimated 100 million monthly active users by January 2023, approximately two months after launch.
Industrial robot statistics from International Federation of Robotics (IFR). Global operational stock exceeded 3.5 million units as of 2023.
Amazon robotics figures from company announcements. The 750,000+ robot count includes various types of automated systems across fulfillment network.
OpenAI's history includes the 2015 founding as a nonprofit, 2019 restructuring to "capped profit," and 2023 partnership expansion with Microsoft.
NVIDIA's market capitalization exceeded various trillion-dollar milestones in 2023-2024, driven by AI chip demand.
Boston Dynamics founded 1992 as MIT spinout; acquired by Google (2013), SoftBank (2017), and Hyundai (2020). Known for athletic robots demonstrated in viral videos.
The "AI winters" of 1974-1980 and late 1980s-1990s were characterized by reduced funding and pessimism following periods of inflated expectations.

Section E — AI and Robotics