<aside> <img src="https://s3-us-west-2.amazonaws.com/secure.notion-static.com/568ec257-3c61-4d30-946f-79f8a3ad2857/Logo_500x500_px.jpeg" alt="https://s3-us-west-2.amazonaws.com/secure.notion-static.com/568ec257-3c61-4d30-946f-79f8a3ad2857/Logo_500x500_px.jpeg" width="40px" /> My take on this:
Skip — way too many anecdotes. One person’s experience may not represent reality.
</aside>
$\fbox{Table~of~Contents}$
In Mumbai, in 2041, Nayana’s family dramatically lowered their insurance premiums by signing up with a new insurance company called Ganesh Insurance. The catch? They had to agree to share all their personal data with the company.
Ganesh instructed the family to use a certain set of apps for everything from investing to finding the best supermarket deals, and over the next few weeks, their phones were constantly pinging with recommendations. The apps told them when to drink water, instructed her grandfather to drive more slowly, and nagged her dad so much about his smoking that he eventually quit. With every healthy decision they made, their insurance premiums fell. It seemed like a win-win for everyone.
But when Nayana fell in love with a man who lived in a less-wealthy neighborhood, the family’s premiums soared. Somehow, the AI had inferred that he was of a different social status, and interpreted that as a health risk to the family.
Nayana’s story provides a chilling insight into how AI can reproduce the discrimination already present in society. One of the most significant AI developments in the last decade has been deep learning. Deep learning allows computers to make predictions, classify data, and recognize patterns. Deep learning is the technology Facebook uses to generate personalized recommendations and maximize the time you spend on its network. By analyzing every click you make and comparing your data to the millions of others in their system, the platform is able to accurately predict what will engage you.
Deep learning can have enormous benefits. AI can analyze millions of data points, and make connections that would elude the human mind. But AI lacks the nuance and complexity of human thinking. It can’t draw on personal experience, abstract concepts, or common sense.
And AI is vulnerable to bias and discrimination. In Nayana’s story, the app didn’t know that her love interest was from a different “caste” and that the match would be seen as socially undesirable. But by analyzing his family’s data and tracking where he lived, it still suggested that the match would be a “threat” to the health of Nayana’s family. Deep learning will only grow more prevalent and powerful in the years to come. The question of how to make it beneficial to society as a whole will be one of our most urgent preoccupations in the near future.
Amaka was scared. A shady company called Ljele had asked him to use his skills as an expert programmer to create a deepfake video for them. He had to make a video of a prominent Nigerian politician admitting to scandalous behavior. If Amaka refused to do it, the company said they would release a fake video of their own, showing Amaka kissing another man in a nightclub. This could land him in prison and cause even bigger problems between him and his family.
In 2018, a video of former President Obama calling President Trump a “total dipshit” went viral online, causing an uproar. The catch? It wasn’t real – it was a deepfake created by Buzzfeed to show what was possible with AI technology, and to warn people to be skeptical of what they see.
To develop the technology for making deepfakes, developers first needed to teach computers to process and make sense of images. So they took inspiration from the human brain, which has a visual cortex that gathers information about an image before sending it to the neocortex, which processes that information and then assigns more complex meaning to what’s being seen. Using this model, designers created a convolutional neural network, or CNN.
To create deepfakes, you need a specific kind of technology called a Generative Adversarial Network, or GAN, which consists of two CNNs. One of these is a “forger,” which analyzes tens of millions of pixels in every image it sees, picking out the unique characteristics of every image. If the forger has analyzed images of, say, dogs, it can then synthesize a fake dog image. It sends this to the second CNN in the network, which is a sort of “detective.” It tests the fake picture against real ones and informs the forger of any errors. The forger then uses that feedback to improve the image and sends it back to the detective. This cycle recurs millions of times, until the fake dog is indistinguishable from a real one. And this process can be used to create very convincing deepfake videos as well as images.
This can have dangerous consequences, of course. As Amaka’s story highlights, deepfakes can be employed as political weapons, discrediting candidates or spreading propaganda. They can also be used to intimidate or blackmail people. In the real world, in 2019, a slew of deepfake porn featuring celebrities’ faces flooded porn sites.
To counter deepfakes, programmers have been racing to create software that can detect anomalies the human eye can’t see. But deepfakes are evolving just as quickly.
Golden Sparrow’s parents were killed in a car accident when he was just a young boy, and he was sent to an orphanage. The people who looked after him there created a special friend for him, a companion that he called Atoman, after his favorite superhero. To see Atoman, Golden Sparrow needed to wear special glasses with a virtual reality interface. Soon, he wore them all the time. Atoman was a perfect companion because he knew everything about Golden Sparrow; after all, he had access to his data cloud. And a biometric ribbon attached to Golden Sparrow’s wrist provided a constant flow of real-time information about his physiological state as well as his behavior.
Atoman became Golden Sparrow’s best friend, helping him with his homework, answering his questions, and planning adventures for him. But, above all, he talked to Golden Sparrow constantly – an ideal conversation partner for an isolated young boy.
