The pace of improvement in AI techniques still surprises. It’s easy to forget that the path breaking technique, Generative Adversarial Network (GAN), was invented as late as 2014 – just over 4 years back. It was only in 2009 that Google started research into self driving cars – a concept that must have seen ridiculous at that time. A mere 9 years later (and it is ‘mere’ given that it was basically stuff of science fiction!) in October 2018, Alphabet’s self-driving car had driven more than 10 million miles and in that same month, California Department of Motor Vehicles issued a permit to operate a fully autonomous car.
Similarly, just a couple of years back, it was really easy to tell the difference between machine generated text and human written text. The machine generated text was often stilted, colloquially (or even grammatically) incorrect and so on. Last week however, Open AI, an non profit backed by Elon Musk, Sam Altman and others decided NOT to make available (as is their standard practice) their new AI Text Generator to public. The reason – it was so good that it could be misused.
This AI was fed just the first (yes, just the first) line from 1984: “It was a bright cold day in April, and the clocks were striking thirteen”. The AI wrote “I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science.” Wow! Right?
Guardian did a story on this AI. As part of the story, they fed the first two paragraphs of their story to the Open AI generator. Based on just the two paragraphs, it wrote a full story with quotes, statistics, background, etc. – all of it fake of course!
That’s really the risk Open AI saw – a text generator so good, it can feel like that text was written by someone credible and was backed by facts. Now combine a text generator with fake images (check this out) or with video deepfakes, and you can, literally, create your own reality – a reality backed by videos, images, news articles, blog posts and so on.
I think in less than a decade, definition of Truth will be “Something you have seen with your own eyes or physically experienced” because everything else will be “fakeable!”