Many of us Baby Boomers remember those old, beloved science fiction shows where people were interacting with computers. The few that come to mind are the Jetsons, Star Trek, and Lost in Space. In each of these shows, a computer played a prominent role in the series. George Jetson relied on his handy Referential Universal Digital Indexer; a computer called RUDI. The Star Trek crew on the U.S.S. Enterprise frequently used the ship’s computer by giving it voice commands. The Lost in Space folk relied on a walking, talking computer simply called Robot. At that time most of us saw this strictly as science fiction; something made up in the minds of the script writers. We never realized that behind the script of these shows, there actually were government, research and other institutions working on the precursors of what we now call Artificial Intelligence or AI.
The origins of AI began with much simpler systems. In the 1950s and 60s, researchers created basic programs that could manipulate language, but they were rigid and limited in their abilities to learn from data. The 1980s and 90s saw improvements with systems that could learn from data, but they still struggled with the complexity of human language. These early computer programs could only respond in limited, scripted ways, so were limited in their level of interactivity. Examples include those early, clunky chatbots of the 1990s that gave short, canned answers to pre-programmed scenarios.
The real breakthrough came in the late 2010s with the invention of the transformer, a new type of computer model developed by researchers at Google. Transformers allowed machines to read and understand language far more effectively than older systems. This new approach enabled computers to process language by allowing them to pay attention to all parts of a sentence simultaneously, much like how we understand context when reading. It then became possible to train computers on vast amounts of text from books, articles, and the Internet, giving them the ability to “predict” words in a way that made their responses sound more natural and thoughtful.
With these advancements, OpenAI released GPT-1 in 2018, followed by increasingly powerful iterations. When GPT-2 was released in 2019, it was so impressive that, fearing misuse, OpenAI initially hesitated to release it publicly. The advent of the release of GPT-3 in 2020 marked a turning point in that suddenly, AI could write coherently about almost any topic.
The “engines” behind these advancements are widely known as Large Language Models (LLMs.) At their core, LLMs are pattern-recognition machines. They don’t “think” like us, but they are extremely good at spotting relationships in language. If you type a sentence into one of these systems, it uses what it has learned from millions (billions?) of examples to generate the most likely and relevant response. Because they’ve been trained on so much information, LLMs can summarize reports, explain scientific ideas, translate languages, or generate creative content.
What makes them exciting is that they allow us to interact with technology using everyday language instead of computer code. What makes them risky is that we can develop an over-reliance on the information they produce. These LLMs have been known to “hallucinate” which is a way to describe the occasional output that is purely made up! Additionally, they require enormous amounts of computing power, which leads to high energy consumption and environmental impact, such as the huge amount of water required to keep them cool. There are also worries about job displacement as AI becomes capable of handling more human tasks.
Privacy and security present additional challenges. Since these systems learn from vast amounts of internet data, they may inadvertently reflect the biases or misinformation present in that data. Ensuring AI systems remain helpful rather than harmful requires constant vigilance and improvement. The advancement in the generation of images, audio, and video also provides opportunities for “bad actors” to misuse these capabilities to produce intentionally misleading information.
As a result, researchers and companies are working hard to make them more accurate, transparent, and efficient. At the same time, we are grappling with larger questions such as how should AI be regulated? How do we ensure it’s used responsibly? What role should it play in workplaces, schools, and everyday life?
Efforts are underway to make these systems more efficient, requiring less computing power while becoming more capable. There’s also growing focus on making AI systems more reliable and truthful, addressing current limitations around accuracy and consistency. Ultimately, we could envision future versions tailored for specific fields such as medicine, law, education, etc. all while becoming more energy-efficient, affordable and integrated into our daily lives.
