A quick snapshot of the history of AI (wait until the end to get the full picture)
We need to pause even as Generative Pre-trained Transformer (GPT) technology seems to be moving on relentlessly. Why is GPT so powerful? And what are its roots? AI first appeared–as new things most often do — in the movies. About a century ago, Fritz Lang’s 1927 dystopic film Metropolis had a humanoid robot spreading chaos. The film was hailed for its special effects, but critics thought the story was “naïve.” Today, the film hasn’t changed, but the same critics will have a different opinion.
Outside of the movies, it was a 1950 paper called Computing Machinery and Intelligence, by philosopher, mathematician, and computer scientist Alan Turing that set off today’s Age of AI by asking a simple question, “Can machines think?” The jury, for the moment, has said a firm “No,” but the answer could change over the next decade or even the following year. Currently, machines are trying hard to make us believe they can think.
From 1950 to 2023, in less than 75 years, we have seen tremendous change in the world of AI. Machines have gone from image identification to image creation. Machines may not be thinking yet, but they are getting close to making us believe they are.
The widely-recognized founder of machine vision was Laurence Roberts. His 1963 thesis, Machine perception of three-dimensional solids, described how to display a three-dimensional array of objects from a two-dimensional picture. For the last decade, AI has been used to identify objects and attributes in images. For example, AI was being used to identify the number of people at an airport boarding gate and automatically determine if it was necessary to open more gates to ease congestion.
Then machine vision began identifying specific objects and attributes, such as buildings, fish, grass, colors, and movement. This has steadily led to the creation of autonomous vehicles.
Over the years, collective computational capabilities, also known as the Deep Learning technique created by researcher JJ Hopfield enhanced these capabilities. AI began to be used to screen and detect diabetic retinopathy at scale without intervention from qualified doctors, predict the spread of pandemics without deploying the scarce resource of health experts, and replace visual inspection of anything — from pencils and oil pipelines to wind turbines and warehouses — to help enhance product quality, supply chain intelligence, and material security.
Now, AI has taken the Big Leap Forward. It doesn’t stop at identifying and labeling things. It creates them as well. Last April, the New York Times wryly observed, “Seeing has not been believing for a very long time.” It then explained how even experts could struggle to identify fake pictures conjured by AI. I created the image of a cat in a Mumbai apartment using Bing’s image creator and a simple text prompt: Cat in a Mumbai apartment, table, drinks, mellow afternoon 1:1 aspect ratio Nendo flavor with a bird on the window sill (although I can’t see the bird on the window sill!). The image is as good as — if not better — than one many professional photographers can produce. It is realistic, and I can keep adding or deleting ideas until I get exactly what I want.
The same can be done with video, audio, music, applications, and digital games. AI is reaching an inflection point where it can create a compelling version of reality or a wholly new and parallel version of reality. It has gone from doing simple things like personalization, suggesting email responses, and playing a challenging game of chess to identifying and curing disease, delivering food using autonomous drones, helping the visually impaired, and turning all of us into the fount of creativity.
Businesses will continue to use AI for tasks such as delivering “next-gen” customer experience and dynamically determining insurance premiums. But the real power of AI will be seen in the coming years — perhaps even in the next few months: AI will be used to determine foreign policy, drive social justice (which humans are poorly qualified to do), ensure that no movie on Netflix flops because it could dynamically change the storyline depending on the viewer’s current state of mind, and give us cameras that use AI, GPS, environmental data, and prompts to create images. These ideas are not quite as outrageous as they may have seemed a decade ago.
Today, some of them are already available. Go ahead, and try the Paragraphica, a context-to-image camera created by Danish artist Bjørn Karmann that uses prompts, AI, and locational data. The viewfinder will tell you how different the world to come is going to be: You will see a description of your current location in the viewfinder. The image may or may not be exactly as you expected today. But wait a bit, maybe for tomorrow or the week after. You never know how far AI will have advanced by then — you get the picture!
