What is AI? - Part 1
What is AI?
"A machine-based system designed to operate with varying levels of autonomy that may exhibit adaptiveness after deployment, and that infers from the input it receives how to generate outputs such as predictions, content, decisions, or recommendations." -- The EU AI Act definition
The concept of artificial intelligence has been around for a while. Alan Turing proposed a test to check the sentience of artificial systems back in the 1950’s. So, I’ve often said that AI has been around forever and then talk about autocorrect, spam filters, and fraud detection. That is technically true. But when it comes to the current concerns about resources, copyright, and the potential SKYNET situation, we’re talking about the third generation of generative AI (image and video) and large language models (LLM) that power the AI assistants.
These are (arguably) the three generations of things all called "AI":
Rules-based systems – 1950s — If/then logic, lookup tables. Autocorrect, early spam filters, early chatbots like ELIZA
Classical machine learning – 1980s — Statistical models that learn from data but aren't neural networks. Fraud detection, recommendation engines, search ranking
Deep learning / neural networks – 2012 — The modern era, image recognition, LLMs, generative AI
The LLMs started off as just a research question: can we build a system that understands the statistical relationships between words, sentences, and concepts well enough to predict what comes next? A training model to simply predict the next word at massive scale turned out to produce something that appeared to understand context, reason, summarize, and generate coherent long-form text. It felt more like you were having a conversation rather than using a search engine. That was somewhat unexpected even to the researchers.
Around 2015-2017 researchers noticed that neural networks were exceeding human performance on specific tasks - image recognition, game playing, protein folding. DeepMind's AlphaFold solving protein folding in 2020 was arguably the moment the scientific community collectively said, “Wait, this is faster at certain research tasks than humans will ever be”. That's what led to the scientific applications: drug discovery, genomics, climate modeling.
From the perspective of the general user, DALL-E, MidJourney, and Stable Diffusion image generators, and the ChatGPT AI Assistant, suddenly exploded on to the scene in 2022. AI didn't accelerate because researchers suddenly got smarter. It accelerated because the industrial infrastructure - chips, data, cloud computing - finally caught up and everything researchers had been theorizing about for the past 75 years could suddenly be built. That’s why it feels like things are moving so quickly. All the research and ideas were caught behind a floodgate that’s just been ripped open.
AI is now growing and learning at an incredible rate. The ChatGPT you used a year ago is a fossil in comparison to what’s available now. According to METR (Model Evaluation and Threat Research), a nonprofit that independently evaluates AI capabilities, the length of tasks that AI can successfully complete has been doubling approximately every seven months since 2019. To put that in concrete terms: around the launch of ChatGPT in late 2022, top AI models could handle tasks that took a human expert about 30 seconds. By early 2026, frontier models were completing tasks that would take a human expert more than 14 hours. And the pace is accelerating - METR's January 2026 updated estimates put the current doubling time at around 4.3 months, roughly 20% faster than before.
The impact on math, science, and medicine will be stunning. I suspect there will be cures for cancer and MS within our lifetimes - as long as AI remains under human control. Something that appears to be questionable given recent developments.
I wish I could say that the majority of AI resource usage is due to scientific advancements but… it’s not. Only about a small percent of the API calls being made are by researchers. The vast majority are being made by corporations automating away human labor to cut costs. Individual users are a rounding error.
References
European Parliament. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council — Artificial Intelligence Act. EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689
Turing, A.M. (1950). Computing machinery and intelligence. Mind, LIX(236), 433–460. https://doi.org/10.1093/mind/LIX.236.433
McCarthy, J., Minsky, M., Rochester, N., & Shannon, C. (1955). A proposal for the Dartmouth summer research project on artificial intelligence. Dartmouth College. https://home.dartmouth.edu/about/artificial-intelligence-ai-coined-dartmouth
IBM. (n.d.). Deep Blue. IBM History. https://www.ibm.com/history/deep-blue
DeepMind. (2020, November 30). AlphaFold: A solution to a 50-year-old grand challenge in biology. https://deepmind.google/science/alphafold/
Jumper, J., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583–589. https://www.nature.com/articles/s41586-021-03819-2
METR. (2026, January). Measuring AI ability to complete long tasks. https://metr.org/blog/2025-01-13-measuring-ai-ability-to-complete-long-tasks/
ABC7 News. (2026, June). Anthropic calls for global freeze on AI development. https://abc7news.com/post/san-francisco-based-anthropic-calls-global-freeze-ai-development-warns-could-soon-escape-human-control/19240090/