Sometimes I dream a phone call. “The blackberries are doing well, Dad. And your old bike was stolen recently. Any news with you?” “Oh, no. Going fine here.” “Fine, see you soon!” Even though I know he’s gone, for a moment I assumed my father was listening on the other end of the line. Dreaming is no longer necessary – we have artificial intelligence. It brings voices from the past to life. The latest feat in that regard is a fictional podcast interview with the late Steve Jobs. The Apple co-founder passed away in 2011, but there were enough audio recordings to train an algorithm to fabricate a synthetic voice. You can make the computer say anything with Jobs’ timbre. ‘He’ does sound like Steve Jobs addressing a roomful of audience, because the voice is based on recordings of his Apple presentations. The cloned voice of the interviewer, talk show host Joe Rogan, sounds more believable. There were hundreds of hours of Rogan’s podcasts available for the algorithms to chew on. And the more data, the better the end result. The conversation between Jobs and Rogan is also made up by artificial intelligence. The conversation is stiff: especially the laughs in between sound uncomfortable. It is only a matter of time before such irregularities are gone, says Jarno Duursma. He is a Dutch tech expert who studies artificial intelligence. Duursma’s fascination: deepfakes, manipulated versions of reality. He built a digital version of himself, complete with simulated voice. Batman eats a burger This is the era of text-to-everything , says Duursma. You can create credible texts based on a few keywords – a trick of language model GPT-3 . The latest trend in artificial intelligence is the generation of images based on a textual description. It’s an addictive activity, such a picture maker: everything you can think of is visible in two clicks. Almost unimaginable. For example, you type “Batman eating a hamburger in Munich” or “A baby with an old face in a car on the moon” and the image generator creates a unique image, in four variants. The results are surprisingly real. Some pictures have something absurd about them. Distorted faces, or people with one limb too many; somewhere between Halloween and a hallucination. But technology is advancing – more computing power, more data – and is becoming more realistic. Many images resemble cartoons or paintings. This way you can easily create ‘art’ yourself in the style of Monet, Van Gogh, Picasso or Warhol. Art becomes more accessible with such powerful tools. Everyone could already paint with Ravensburger, now everyone can have their fantasy portrayed by an algorithm. The real artists scratch their heads. Raked together There are several projects where you can try out this technology. They are called Dall-E2 , Stable Diffusion and Midjourney and have recently become accessible to everyone. The technique works something like this: you can train algorithms to remove noise from an image. If you reverse that process, you can teach the computer from noise to create a completely new image step-by-step, for example using the terms Batman and hamburger. The networks are trained on databases with billions of (human) labeled pictures, raked together on the internet. You can add your own images to it. Microsoft, one of the investors in Dall-E, is adding the picture maker to its design software so that you can “create unique postcards and invitations yourself.” That’s the friendly version. The image generators beg to be abused. To spread fake events, to put words in politicians’ mouths. To extort or scam people, through the screen. Google has built in filters to prevent you from creating images or videos that are violent or pornographic in nature. But not all abuse can be prevented. At my request, Dall-E2 generated an image that looked like a demonstration was taking place at the Peace Palace in The Hague. Fake. Through Stable Diffusion, I had Vladimir Putin and Xi Jinping kiss deeply – also fake. It is easy to fool the average media consumer when technology develops so quickly, thinks Jarno Duursma. “Many pictures still look like a drawing or painting. But in a year you will be able to create photo-realistic images with a few words of text.” There’s only one thing to do: zero trust . Don’t trust anything you see on the screen. Three assignments to the Dall-E 2 image generator: Batman eats a hamburger in Munich, demonstration at the Peace Palace and a Russian soldier kissing a frog. With the first application of deepfakes, in 2018, algorithms pasted movie stars ‘ faces onto pornstars’ bodies. Four years later, the bar to imitate someone is much lower. Amazon says that you can reproduce someone’s voice based on one minute of audio material. Grandma can continue to read stories through Amazon’s smart speaker for eternity . Videos also don’t require a lot of data: you can bring an old photo to life via an algorithm called Deep Nostalgia. This produces a moving picture like in a Harry Potter newspaper. Duursma: „I tried it with a photo of my mother, who died three years ago. Her photo moved, but I thought: my mother definitely doesn’t look like that. This is surrogate.” A new market for fakes is emerging. That of the grieving process. The idea is that you can still talk to the dead via artificial intelligence. Wired magazine called that phenomenon an example of “our changing relationship with death, not necessarily creepy .” The question is whether you want to continue living like this after your death, as a digital museum piece . Jarno Duursma is not interested in it: „I made an avatar of myself and a voice clone. So if I die this afternoon, my kids and my wife can keep making videos of me indefinitely. But I hope they don’t.” Losing is part of life, says Duursma. And if you want to be sure of this, you will have to record in your will whether relatives are not allowed to reuse your voice and image. Think of it as an extension of the current donor register : use my organs, but delete my data. Dreaming is allowed. Marc Hijink is a technology editor and writes here every week about the back of tech. React? Contact Marc Hijink via LinkedIn .