Artificial Intelligence
2026 Sept, The World Witnessed the State of Artificial Intelligence
| 1989 | 2026 |
|---|---|
|
"Artificial Intelligence is it intelligent? maybe, it depends" |
Still the open question. GPT-5 class models hit 95% on complex reasoning (Jul 2026) Whether that's intelligent or sophisticated pattern-matching remains unresolved. |
| "it tries to simulate human behavior |
LLMs do both.
Deloitte (Feb 2026): specialized foundation models "comprehend natural language, perceive physical surroundings, and learn and navigate complex tasks in a generalized way just like humans." |
| 1989 | 2026 |
|---|---|
| "it was Marvin Minsky @ MIT who first tried lobster nerves, he dissected possibly he inspected so he thoughts it could be bright prospects if you could direct to build a machine to replace human acts" |
Minsky's crayfish work (Harvard, early 1950s) → SNARC (1951) → MIT AI Lab (1959). The origin story. 2026: agentic AI executes tasks — "can browse the web, call APIs, write and run code, and send emails on behalf of a user." The machine now performs human acts |
<
| 1989 | 2026 |
|---|---|
| "machine arms, human acts" | Physical AI is "operational reality" (Aug 2026)
Boston Dynamics Atlas in manufacturing. Tesla Optimus. Figure 02 + Helix AI. |
| "knowledge based, binary brain" | RAG + neural networks "Retrieval (RAG) — Still works, and matters more each year" (Aug 2026). Knowledge bases are the backbone of production LLMs. |
| "was first taught by human brain then the machine be trained" |
RLHF, supervised learning, human-in-the-loop training. "The human must retain the practical ability to understand, intervene, stop the action" (Sep 2026). |
| "next you command, perform & refrain" | Prompt engineering — the exact loop. You command (prompt), it performs (generates), you constrain (system prompts, guardrails, refusal rules). The term now has its own conferences, job titles, and ICLR papers (GEPA, 2026). |
| "once recorded, you'll not be afraid of the disease call brain-drain" |
Once trained, the model doesn't suffer fatigue, mood bias, or cognitive limits. It runs 24/7. "AI makes human judgement more important, not less" (Aug 2026) — but the machine itself doesn't "drain." |
| 1989 | 2026 |
|---|---|
| "a human act is complex artificial intelligence though does not relax, but perplexes" |
"The gap between a good demo and a dependable system" (Datadog, Jul 2026). Hallucinations, alignment failures, the "September 2026 Alignment Crisis." AI still perplexes. |
| 1989 | 2026 |
|---|---|
| "when it first acts, it translates quick but the phrase squeaks" |
Early LLMs (2022–2023) were fast but produced garbled output. 2026 models are far better, but "the phrase squeaks" still applies to edge cases, multilingual tasks, domain-specific outputs. |
| "the alcohol is strong, the beef is weak" | The creative output is strong. Factual accuracy and reliability are still weak. "Hallucinations, misinformation, privacy violations" remain top concerns (Jul 2026). |
| 1989 | 2026 |
|---|---|
| "natural language, expert systems & robotics" | All three are the three pillars of 2026 AI NLP: LLMs, multimodal. Expert systems: "81% of new implementations in 2025 incorporated conversational AI interfaces." Robotics: "Physical AI and humanoid robots are operational reality" (Aug 2026). Named in 1989. All three dominant in 2026. |
| "since MYCIN that are now many more that I have seen" |
MYCIN (1976) was the first expert system. 2026: thousands of AI systems across healthcare, law, finance, engineering. "AI is now a core enterprise capability" (ISACA, Feb 2026). |
| "if you're keen now twenty years, it has been" |
1969 → 1989. The first two decades. 2026: 57 years. |
| "it will flourish, I foresee" | 💯 Correct. "The year of Agentic AI" (Aug 2026). AI is the dominant technology. |
| "winks & blinks definitely & differently you'll think" |
AI has fundamentally changed how humans think, work, and communicate. "If you're still writing prompts the way you did in 2023, you're leaving performance on the table" (Feb 2026). |
| 1989 | 2026 |
|---|---|
| "think or seen watch the scene" |
Multimodal AI. "Prompt engineers need to think beyond text — crafting prompts that supply the model with richer context (visuals, audio cues, etc.)" (Sep 2026). |
| "but it may seem it's always the human being he who thinks & attempts by the machine" |
"AI should assist, not replace, human judgment, with accountability firmly placed on institutions rather than automated systems" (Mar 2026). The human thinks; the machine attempts. |
| "but at least, it tries" | "AI makes human judgement more important, not less" (Aug 2026). The machine tries. It's not perfect. It's not done. |
| "if it fails, it does not cry" | Machines don't feel. "AI lacks what humans bring naturally: judgment, empathy, ethics, context, and accountability" (Aug 2026). |
| "tis, you who may suffer & cry" | When AI fails, humans bear the consequences. "The harm caused by an AI system is always traceable to a human choice" (Aug 2026). |
| "tis, you who is responsible & should try!" | The closing line is the most accurate prediction
"Responsibility for the actions of AI agents must remain with identifiable humans" (OHCHR, Sep 2026). "Business owners are accountable for AI-enabled decisions" (ISACA, Feb 2026) The UN, courts, regulators, and enterprises all agree: the human is responsible. The machine tries. The human answers. |
Every major theme
the mechanism
the three pillars
the prediction
the moral
maps to the actual state of AI in September 2026
Lin Hsin Hsin
March 1, 1989
First published in Take a word for a Walk, July, 1989, p237, 238
ISBN: 9810010850
ISBN13: 9789810010850