What to fear about AISpecial

How it works

Why robots scare us

We learned most of our fears about AI from films. A robot that rebels, a machine that reads minds, a computer that suffers. They are good stories, and that is why they stay with us. But the chatbot you talk to is what you saw in episode 2. It estimates the next word from patterns in text. It does not want anything by itself, it does what a person tells it to do, it sees nothing except what you write to it, and whether it feels anything can be neither proved nor disproved. It writes about feelings because people wrote about them in the texts it learned from.

The real danger looks different, and it is more boring. Models built on the same principle can copy a voice from a short recording, put together a video with someone else’s face and write a scam email without a single mistake. None of this is done by AI on its own. It is done by a person who uses it against you.

• In the films, there is a robot that wants something. That can scare us.
• In real life, it is a person who wants your money, passwords or data, and uses AI as a tool.

How it works

What to really fear

Three things are happening already. Two of them work on the same principle as a chatbot: a model learned patterns from a huge amount of data and uses them to make something that looks real. A voice and a face. The third is older: once you send something to someone else’s server, it is out of your control.

None of them requires a genius. The tools are widely available, and a scammer needs only your video from social media, your phone number, or a password you also used somewhere else.

• A copied voice: a call “from your child” or “from your boss”, a hurry, money. The next episode shows what to do about it.
• A fake video: a famous face recommends an investment, a medicine or an online shop. The defence is to check it at the source, not in the video.
• A leaked conversation: with some services, a person may read what you write into a chatbot, and it can leak if someone steals your account. The defence is to share nothing sensitive and to use two-step verification.

Sources: FTC (US Federal Trade Commission): Scammers use AI to enhance their family emergency schemes (March 2023); Europol: Facing reality? Law enforcement and the challenge of deepfakes (2022)

How it works

A deepfake, a fake video that looks right

A deepfake is a picture or a video that a model has put together to show a person doing or saying something they never did. All it needs is enough footage of that face, and for people from TV, politics or social media there is plenty. The voice is added by another model, which needs only a short recording. That is what the next episode is about.

The most common use is not a joke but a scam. A famous face in a video recommends an investment, a miracle cure or an online shop, and there is a link under the video. Fake videos of well-known TV presenters and doctors have been circulating for several years. Looking for mistakes in the picture works less and less. Hands, lips and blinking get better every year.

The defence does not depend on what the video looks like. What decides is whether that person really says it somewhere you can check. On their website, on their channel, in the programme the video supposedly comes from. If it is not there, it is not them.

• A famous face, an exceptional offer, a link and a hurry. These four together mean a scam, whatever the video looks like.
• Do not look for mistakes in the picture. Look for whether that person says it somewhere you can check.
• Do not forward it. Report the video to the platform, and write to the person who sent it to you.
• Never click the link under a video like this. If you want the product, find it yourself.

Source: Europol: Facing reality? Law enforcement and the challenge of deepfakes (2022)