AI & Privacy: Do they Mix?

Is privacy something that can survive the age of AI?

Introduction: The personal price of privacy

I hate social media. It’s a time-sucking vortex of distraction. It’s a soul-sucking hell of comparison to others. It creates echo chambers that stifle intellectual curiosity and nuance. I try not to hate it. I can (just about) stomach LinkedIn - professional, clear purpose, not 100% vacuous all of the time. On the pile of unmet New Year’s resolutions are promises I will post more, engage more, and commit to consistent effort in embracing the platforms… but my privacy gets the better of me each and every time.

That privacy comes at a cost. A version of me willing to post consistently, share enthusiastically, and be an open digital book - they would reap the benefits of a digital media which allows a huge, never before accessible scale of reach to everyone. That means new connections, new opportunities, and the fly-wheel of content turning on its own steam - your content be seen by people while you sleep. Even with a modest audience and reach, each post might be seen by 2000 people - 10 posts a week means 20000 views. Social media influencers get it - it’s more important to be seen, be in the conversation, be front of mind, than it is to be private.

I don’t get it. The price for that is missed connections, missed opportunities, and more manual effort to achieve the same results as those social media butterflies. I’m a social media luddite - I like the idea of being in the room with people. I still have options and choices - I could maybe even hire a room, fill the event with 100 people, and tell them my message. A lot of work, but it’s still in my power to try that. However, in order to compete with just that 1 week of social media posts I’d have to plan, organise, and run that event 200 times… that’d be a busy year, nevermind week!

As much as I resist, I can’t deny the reality: in the age of the internet, privacy has a price. And as we’ve become accustomed to with our recent inflation experience, that price is going to shoot up!

The AI revolution: Context is king

The trade-off for privacy in the social media world is reach and opportunity. That’s acceptable to some of us - we don’t all need to be influencers, or all need to sell a product or message or brand.

But the trade-off for privacy with AI will be value creation - a much higher stakes game. AI’s transformative potential is dependent on access to personal data and context.

Interacting with AI, although it’s already incredibly powerful, still feels hollow after you get over the initial buzz of having a conversation with a machine. These systems are reliant on huge amounts of data. ChatGPT was reportedly trained on 570 GB of data, from web pages to books. That’s huge, but given that each day 2.5 quintillion bytes of data are produced by internet users, it’s barely scratching the surface. But I’m not sure it’s a lack of this world data that is stopping AI being useful. It’s personal context.

The reason programming is so annoying is because you’re having to tell a computer, essentially a rock, how to do something in the real world. A rock has no ability to understand the real world, it has no context, and so programmers have to painstakingly input this context into the computer through the frustrating process known as coding.

A similar pain can be felt when you use Large Language Models (or LLMs). If you ask the AI to perform a simple task, you’ll often be underwhelmed with a generic, passable response. That’s because the system has no context. There’s plenty of tips and tricks that help improve performance when writing prompts, but the biggest win by far is giving as much context as possible. That’s the job of Prompt Engineers; what programmers are to software, prompt engineers are to LLMs. They create prompts that are full of clear, well-defined, well-structured context so the AI knows what we want, why we want it, and how we want it. Creating these prompts is still a painful process, but I believe it’s a temporary pain.

The Faustian bargain: Data for superpowers

If context improves AI, and it can handle incredibly large data sets, then I can see only one outcome - you give AI access to your whole life. Everything. It means that when you say “give me some recipes for a dinner with Sonia and Jason” it will already know what’s in your fridge from analysing all your shopping habits and amazon baskets. It will already know your allergies, preferences, current fad diet requirements, and will even have access to your conversation history with Sonia and Jason to piece together their requirements too. It will even remember that the first time you all ate together was in Italy, so it would recommend some Italian dishes to bring nostalgia to the party. That results in the best possible recommendation for you rather than a completely generic set of recipes, for no extra effort or monetary cost - all that was required was you to share your life with AI.

That’s just something trivial, like picking a meal to entertain whoever Sonia and Jason are. Think about the impact on life in general, financial advice, health recommendations, work decisions, productivity in general. AI will be able to know you better than you know yourself, and take actions on your behalf in a fraction of the time and probably with better outcomes.

The benefits of AI will create an irresistible incentive for people to trade privacy for performance. But there’s hurdles to get over yet - like with me and my approach to social media, you might be feeling like it’s not quite for you. Despite seeing the benefits, you might wrestle with your privacy. In doing that, you’ll no doubt be left behind at warp speed by everyone able to take full advantage of AI rather than having to painstakingly write complex prompts just to get something useful out of the system.

I think most people are going to move towards willingly inviting AI into their lives, in order to reap the benefits and not be left behind by the coming wave of value. That’s really going to put a spotlight on how secret, safe, and secure your data is.

The path forward: Federated data and privacy innovation

Innovations in data privacy and security will be critical to realising AI’s potential while protecting individual rights. Rather than all our data going to live in a big cloud in the sky, we might be on the verge of federated data being the mainstream. Sir Tim Berners-Lee is a champion of Solid, an open-source private data store for people. Apple is the £2 trillion behemoth who might be best placed to leap out of the AI shadows and take a lead on AI with federated data privacy - the data might live on your device alongside your local AI, you can be fully in control of who sees your data, and if it’s sent anywhere it can be anonymised.

“Privacy is built into all of Apple’s products and services from the ground up, from the moment users open their devices to every time they use an app. Apple products and features include innovative privacy technologies designed to minimise how much user data anyone can access.”

As individuals, we will surely demand robust data protections as a prerequisite for sharing our lives with AI. Companies and developers must prioritise privacy and security at every stage of the AI pipeline. And policymakers must update regulations to keep pace with AI’s rapid evolution, ensuring that our laws protect privacy without stifling innovation.

With AI, it’s not a question of if, but when. Will we rise to the challenge and build a future where the benefits of AI are available to all, while our most intimate data remains secure? The choices we make now will determine the answer.