The Future is Private Compute Through Homomorphic Encryption

We are already in a world where computers can compute without knowing what they're computing. Its a crazy, almost magical concept called Homomorphic Encryption (HE). What is even more surprising is the fact that this is already being used by regular users if you use Apple devices (more on that later in the post).

What is Homomorphic Encryption?

Before we get into HE, it's important to understand where we are today with encryption and privacy. End-to-end encryption (E2EE) is already widely used in private messaging apps like WhatsApp, ensuring that only the sender and recipient can read the messages. No one, not even the company providing the service can ever read this data. That is how your private messages stay private.

While E2EE keeps data private during transmission, any computation on it (like editing it, changing it, analyzing it etc) is not possible do on the server. This means anything that needs to interact with the data is limited by the technology on the end device where the data is decrypted (in messaging - it is your phone).

Homomorphic Encryption (HE) is the next step in this evolution. Homomorphic Encryption takes it further by enabling computations on encrypted data without ever decrypting it. This means that one can do things with encrypted data on the server. It unlocks a whole host of services that were not possible or very hard to do just on the end device.

At its core, Homomorphic Encryption is a type of encryption that allows a computer to perform calculations on encrypted data without ever needing to decrypt it. Think of it like this: if normal encryption is locking data in a safe, homomorphic encryption lets you perform calculations on the locked safe without ever opening it. The result is that you can generate meaningful output without anyone ever seeing what’s inside. See the video below.

This means that sensitive data—like medical records, financial information, or even private conversations—can be processed securely, with no one (not even the computer doing the processing) ever gaining direct access to it.

If you use an Apple device right now. You are using HE.

Apple is already using it in their services that are focused on privacy. Like their Caller ID service. They can show you the ID of the caller by retrieving data from Apple servers without knowing which number the user queried for. Perfect privacy preserved while doing the required job.

They recently extended it from Caller ID Lookup to two more things - 

• Enforcing adult website filtering for kids - they can tag which urls are adult vs not, without knowing which URL the device is requesting
• Photos keyword search and filtering using the same techniques - when you search for "Eiffel Tower" in your photos, apple has tagged your photos with "Eiffel Tower" via its servers without actually knowing which photos or which users have Eiffel Tower in their photos

Both of these applications handle info which is private and sensitive. And it works without compromising the privacy of the user.

HE enables Distributed Private Compute

Now, let’s extend the availability of this technology from Apple to everyone. Imagine you have a huge dataset that you want processed, but you don’t have the hardware to do it yourself. Today, you might rent cloud servers and trust them with your data, but with HE, you wouldn’t need to trust anyone. You could send encrypted data to a distributed network of computers, and they could process it—all without ever having a clue what they’re handling. This could be anyone’s computer: mine, yours, even a whole network of idle machines.

This kind of “private compute” could make it much easier for people to rent out their unused processing power without worrying about the data they’re handling. Think of it like Airbnb for compute—but you don’t need to worry about what’s happening in the “guest room” because it’s all encrypted and secure.

HE, AI and Blockchain - Buzzword Central

Where it gets even more interesting is when we combine HE with AI and blockchain into the mix. These AIs could handle anyone's data securely while still doing arbitrary computations with it, and use blockchain for payment handling for any economic activity. This unlocks whole new way to do things automatically. My mind cannot even come up with the things that are possible with combination of HE, blockchains and AIs.

But I am gonna give it a shot anyways, let me try to paint an exaggerated picture.

Lets say there is an entity (call it ACME) that wants to run a particular query (Q1) - "What % of males aged between 30-50 are diabetic in the state of California but only the urban regions?". ACME is willing to pay for this information. Doing it manually today is extremely cumbersome, extremely expensive and host of privacy issues that need to be handled to do it properly.

Enter our health marketplace (HealthMart) that gives fast and accurate answers, cheaply. ACME is willing to pay upto $0.5 / population of 100 people. This is fed into ACME's AI data agent that interacts with HealthMart along with other similar marketplaces. ACME's AI posts the query on Healthmart public query board. The board has a bunch of healthcare providers who have other AI agents that look for such queries, negotiate a price (perhaps an auction?) and then provide access to encrypted data (ED1). Once ACME gets access this data, they don't know what is raw inside but they can run queries against all the information they need to answer the question. ACME does not want to run the HE server infrastructure 24/7 just to do this. They just want an answer to this question. So ACME encrypts their query Q1 into EQ1.

Now both ED1 and EQ1 is given to compute providers who run computing infrastructure to answer such queries and charge a few cents to run a query.

This entire transaction can happen within a few seconds of someone in ACME entering their query in their dashboard and they have the answer back in a few seconds.

No one in this entire transaction - ACME, HealthMart, the AI agents or its compute providers have access to the private data of individuals. But ACME got its answers and others made money. Even if ED1 or EQ1 is leaked, no one will know anything specific about what was queried or what the data result was.

This can even lead to a publicly available infrastructure for access to data in markets, a "google for data" where privacy is preserved. Google / Meta will probably end up using or providing such a thing.

Extending it further, it means: distributed AI models that learn from private data without ever compromising confidentiality, financial models that are computed without leaking a cent of sensitive data, and a world where compute is as decentralized as the internet itself, without the risks of data exposure.

It is almost embarrassing to cram Homomorphic Encryption, AI, and blockchain into a single post, but it is clear that they are the fundamental building blocks of the future.

HE is "safe", but for the bad guys too

Every technology is a double-edged sword. It's what the humans do with it that matters.

HE will allow governments to put censorship filters in place without the public explicitly knowing what the filters are. I can totally see the dictators of the world misuse "private" to now exert even more control over flow of information. Because they can now censor "private" too. For example, encrypted communications or content could be scanned for certain encrypted markers that match predefined criteria—without revealing the specific data or keywords being targeted.

Just like the internet, AI, and blockchain, applications of HE (not HE itself) will come under regulation. It has to be. There are too many ways it can be used for bad.

I am really excited to see how the fundamental technology breakthroughs of late 2010s and early 2020s play out over the next couple of decades as they scale and proliferate through our daily lives.

Special thanks to Paulami Sen, Mansi Gandhi and Neehar Cherabuddi for reading drafts of this post.