Assertion Supported AI assessment confidence: 98% certainty 4/5 debate potential 2/5

Carlson-Wee: Homomorphic encryption enables machine learning on private, encrypted user data

Olaf Carlson-Wee · Seven Trends in Blockchain Computing (Spring 2019) · May 8, 2019 · at 45:34

Olaf Carlson-Wee, founder of Polychain Capital, explains how homomorphic encryption allows corporations to train AI models without accessing raw user data.

0:00 / 0:24exact quote · 25.0s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“One interesting intersection is homomorphic encryption, which allows you to train a machine learning system based on data that you actually don't know the plain text, so you only see the encrypted version. It allows people to say, okay, I'm going to share the data from my Tesla or my smartphone with a major corporation and get paid for that data. And that corporation will actually never learn the data, but can still train the machine learning algorithm.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Homomorphic encryption allows machine learning models to be trained on encrypted data from devices without revealing the underlying plaintext data to corporations.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

Academic cryptography literature explicitly confirms that homomorphic encryption permits training machine learning models directly on ciphertext without revealing the underlying plaintext data. Research demonstrates this across multiple algorithms and schemes, including logistic regression training via the CKKS scheme (Kim et al., 2018; Han et al., AAAI 2019) and neural network training via Fully Homomorphic Encryption (FHE) without requiring a decryption key during training.

Kim et al. (2018) published 'Logistic regression model training based on the approximate homomorphic encryption', demonstrating ML training on encrypted data using CKKS without decrypting input features or intermediate outputs.
Han et al. (AAAI-19) demonstrated training an encrypted logistic regression model at scale without requiring a decryption key during the training phase.
Colombo et al. (WAHC '24) demonstrated end-to-end training of non-linear neural networks directly on ciphertext using Torus FHE (TFHE).

Checked against

[1] medium.com [2] medium.com [3] nih.gov [4] arxiv.org [5] d-nb.info [6] aaai.org [7] github.io [8] homomorphicencryption.org

"logistic regression" "homomorphic encryption" training 2018 OR 2019 OR 2020 paper homomorphic encryption machine learning model training on encrypted data literature ciphertext "Cryptonets" "homomorphic encryption" "machine learning" training encrypted data training machine learning models on encrypted data fully homomorphic encryption academic paper

assessed as of 2026-07-28

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

homomorphic encryption Tesla

⚑ Spot an error? Report it

More from Olaf Carlson-Wee

Opinion

Carlson-Wee: ICOs are the most efficient capital coordination mechanism in history

“With ICOs, I think that it's probably the fastest and most efficient capital coordination we've ever seen in capitalism, like, since money was invented.”

Olaf Carlson-Wee Jan 2, 2019 ▶ 19:31 a16z Podcast | Trends in Cryptocurrencies

Assertion Partly supported

Carlson-Wee: Ethereum has 100 times more app developers than core developers

“Ethereum has this weird problem where there's probably a hundred X the number of developers building apps on top as there are building core protocol stuff for Ethereum.”

Olaf Carlson-Wee May 8, 2019 ▶ 11:27 Seven Trends in Blockchain Computing (Spring 2019)

Insight

Carlson-Wee: Mitigating 51% attacks is significantly easier in proof-of-stake networks

“The second thing is that if you attack a proof of stake network, mitigation of the attack after it happens is significantly easier. So if you, come in with, you know, 51% of the coins, and in, in most proof of stake, it's actually 34% of the coins is enough to…”

Olaf Carlson-Wee May 8, 2019 ▶ 14:48 Seven Trends in Blockchain Computing (Spring 2019)

Insight

Carlson-Wee: Decentralization is an architecture for permissionless building, not an ideology

“Decentralization is not an ideological thing. It's an architecture to support that permissionless building.”

Olaf Carlson-Wee May 8, 2019 ▶ 25:27 Seven Trends in Blockchain Computing (Spring 2019)

Prediction Not checkable as stated

Carlson-Wee predicts DAOs will use internet money for financial contracts

“If we do have these sort of decentralized, autonomous corporations or something, they're going to be using the internet money in order to communicate among each other and create financial contracts and things like that.”

Olaf Carlson-Wee May 8, 2019 ▶ 27:49 Seven Trends in Blockchain Computing (Spring 2019)

Opinion

Carlson-Wee: Locking 5% of Ether in MakerDAO creates systemic risk

“Now I do think it almost starts to create systemic risk at around, say, five percent of all ether.”

Olaf Carlson-Wee May 8, 2019 ▶ 35:16 Seven Trends in Blockchain Computing (Spring 2019)

Carlson-Wee: Homomorphic encryption enables machine learning on private, encrypted user data

More from Olaf Carlson-Wee

Turn any episode into a week of clips.