AI Poses a Threat to Privacy

When privacy-aware people hear about AI threatening privacy, they typically imagine AI-powered facial recognition, gait analysis, and the like. While these AI-enabled technologies are concerning and I’ve lamented about the use of facial recognition in the past, today I’ll come at the subject of AI and privacy from a different angle which I’ll introduce by first talking about smartphones.

If you live in a technologically advanced society, it’s somewhere between inconvenient to impossible not to own a privacy-invading smartphone. Just two decades ago, no one needed one. Since then, society has become so acclimated to everyone having them that those who refuse to use one for privacy reasons get left out. They’re partially locked out of the workforce, excluded from social circles, blocked from vital resources, and generally live an inconvenient life.

I fear that the same sort of thing will happen with AI. I’ll explain.

There’s Github Copilot, an AI productivity tool which helps programmers be more efficient. Just a few weeks ago, Microsoft announced an AI assistant for Microsoft office. School children use OpenAI’s ChatGPT to do their homework. These AI productivity tools may not be essential just yet, but they’ll get better. And as they get better, it may eventually become infeasible to even compete in the workforce without being AI assisted since those willing to use it AI will have a huge advantage over you.

The same thing could happen in people’s personal lives. We may eventually reach a point where nearly everything people do is AI-assisted. Want to learn how to cook? Your smart assistant will teach you with a teaching style that’s personalized to you. Need to improve your diet and exercise habits? Your AI smart home will create a personalized healthy diet and exercise regime that works for you. Your AI therapist will listen to all your problems free of charge and offer scientifically supported advice for your unique problem. AI may even augment your sex life. It will reach a point where people who refuse to use AI assistants are at a distinct disadvantage compared to those who use AI.

What possible reason would there be to object to using AI when it comes with such great benefits?

The only way to benefit from today’s most powerful AI and possibly the AI of the future is by sending your data to the untrusted remote server on which the AI runs. This server may be collecting the data you give the AI. You can’t get around this by running the AI software on your own computer. That requires more computing power and electricity costs than most people can afford. Also, companies are disincentivized from revealing how the AI works to competitors, so the program may not be made available to you even if you can afford to run it.

So currently, the only way to protect your privacy from the most powerful AIs is to sacrifice all the benefits of using them. As is happening with smartphones, making this sacrifice for your privacy may become prohibitively inconvenient. On the other hand, regulating AI so that, by law, it must be democratized, giving users full personal control over powerful AI systems probably isn’t a good idea. Safeguards will be removed. It’ll be misused for malicious ends. And as AI gets more powerful, giving everyone access just becomes reckless.

So how do we proceed? I don’t think I have the required expertise to offer any concrete advice except to say that homomorphic encryption may be part of the solution. It wouldn’t solve all the problems with AI, but at least it would allow powerful AI to benefit users without compromising their privacy or giving them full unrestricted access.