Evil Twins Are No Longer Just a Bad Plot Device
Evil Twins Are No Longer Just a Bad Plot Device Scroll through any app or media for five minutes and you’ll probably run across something about AI. Scroll for five more and you’ll also likely see some kind of meme or ominous reference to the Terminator or other trope about how AI is going to take over the world. But is AI really that big of a deal, and is it seriously endangering humans and humanity? Let’s take a look.
To start with, AI is not something very new – it’s been around for ages, especially in its initial meaning of “any system capable to make decisions without human participation.”
What has changed though is the way these systems are built today. Originally, a programmer had to explicitly insert logic into a computer by using a formal programming language, which was very straightforward: “If x=3, then print 3.”
However, starting a couple of decades ago, a new way of programming emerged. This method could emulate the way humans learn by creating a mathematical model – called “neural net” – that resembles human neurons. It could then feed it a huge amount of data and literally train it, telling it ?what’s good and bad, what’s required and what’s not and so on, and then give you the results.
After dozens of years of perfecting both the software and hardware used for this second method, today it has started to create significant value: making recommendations, predicting behaviors, creating art and many other things. Albeit with a couple of drawbacks: one, it requires a massive amount of resources like data and power, and two, we cannot fully understand how it works. Well, we do, in general. But just as it is impossible to find a single neuron that provides an answer to a specific question in the human brain, so too is it impossible to find a single line of code that provides an answer to a specific question in AI programming.
So why bother? Well, if we cannot directly see and change the specific code or neuron leading to a specific behavior, we cannot manage the AI any better than you can a three-year-old kid. Thus, a Terminator scenario becomes very slightly more probable, though of course still largely relegated to the world of fiction. But what really can lead to a doomsday scenario is the fact that because these AI-based systems are constantly learning, they actually ingest vast amounts of data regarding every task you put them to (programmers call this “context” to sound smart). And that data cannot be taken out of the AI. Ever.
Now imagine an AI helping you to write your emails. Useful, right? But to be really able to help you do the task effectively, it needs context. What are we writing about? To whom? What is the situation? Who is the author? What is the appropriate tone?
To understand the task better, the software basically needs all the context it can get about you. Ideally (from the point of view of AI), that would include all of your emails, calendar appointments, contacts, actions, social interactions … everything. And if it does succeed in this task, it will be able to emulate you quite well, including background knowledge, tone of voice, wording, character, inclinations, life situation and many other aspects. Now we’re moving from “useful” to “uncomfortable.”
Now imagine that this AI is given – or on its own decides to perform – another, different task, but this time it’s doing something that you didn’t want done. Maybe something malicious or fraudulent, like transferring all your money to someone else. Or sending a note about a reactionary political view to your friends. How would anyone notice the difference? If AI has access to all your data and can imitate your writing, speech and even video (say “hi” to deepfakes!), it becomes what is called your digital twin – a copy of you that can act independently and be indistinguishable from the real you.
Digital twins were actually originally designed to serve and create a better future. An AI author could build digital twins of machines and processes, and then use them to predict failures of mechanical systems like jet engines, or to enable the modelling of complex things like hospitals or railroads. But with every new technology there too often comes a time when someone says “aha!,” and figures out a way to use it to hurt people or take things. And with digital twins this time has definitely come: Modern fraud schemes include synthesized voice and texts, deepfake videos and even AI-based victim search and strategy decisions.
To solve this, more and more attention is being brought to data privacy, which is a fancy way of saying “not giving away anything you don’t want to be used against you.” Which is nowadays pretty much anything.
Strangely, some of those voices are coming from government and big corporations, which can raise eyebrows due to conflicts of interest. Indeed, isn’t it surprising that your government, which usually wants to know everything about you, now seems to be caring about you not giving your data to anyone? Or that the digital advertising sector, which makes it a point to know every tiny detail about your interests and preferences, suddenly starts claiming it doesn’t collect any data? While in a perfect world it would be nice to trust those statements at face value, being an adult also means knowing when to also keep a close eye on things yourself.
So if you prefer to ensure your own data protection, instead of just relying on authorities or companies who can change EULAs (end user licensing agreements) at the drop of a hat, here are some tips to make it harder for entities to create a digital twin of you without your consent:
By implementing these practices and by just being more aware, you can significantly enhance your online privacy and protect your personal information. Remember that perfect anonymity online is extremely difficult, but these steps will go a long way to ensuring that you are the only you out there, and that the idea of your digital twin lives on only as a movie cliché and not as a real life problem/