Defensive machine learning.
We’re in the midst of a huge shift in power where corporations use machine learning to exploit more and more people.
There are countless examples: harmful social media algorithms, gig economy workers managed by arbitrary, shadowy machine learning driven tools, the disappearance of uniform pricing and the return of haggling (Amazon, Zalando etc set price based on machine learning).
We need defensive machine learning that gives people a chance to shift power back.
It does not have to be perfect, it just has to be more efficient than a human trying to haggle with an algorithm when it comes to a shoe.
A good early example is: ad-blockers. While most ad-blockers do not use machine learning, it’s unequivocally one of the best personal agents anyone has.
No wonder Google has been trying to wage a war on ad-blockers not just from a business but from also a trust perspective.
A large corporation trying to gaslight people into believing “they care more”
So, we need defensive machine learning, a couple of ideas:
- machine learning to haggle with online marketplaces about the best price AND/OR the most environmentally and labour friendly source
- machine learning to break news consumption into it’s own separate thing outside social networks, where truthfulness, trust, and following-up topics is prioritized, this could be as simple as a glorified feed reader
- machine learning to manage email on the client side based on priorities set by the user
@szbalint Defensive ML, as every ML need lots of data. At the same time it should be privacy oriented that mean local processing using local only data. Seems to be two opposing conditions.
@miklo we can do data decentralized - training on shared pools of stuff and use the results, this isn't rocket science
@kakure @miklo @szbalint At first it's more the case of making use of info they've already acquired through privacy abuse. Then the data they collect on the sales will further exploit people, causing cyclic abuse of privacy. #Amazon has been caught doing #personalizedPricing before, on DVDs: https://www.consumerreports.org/car-insurance/why-you-may-be-paying-too-much-for-your-car-insurance
@szbalint I thought about the "glorified feed reader" thing before - I think it doesn't even need any machine learning. It might be nice to have in small doses though, e.g. for classifying article topics or similar
@szbalint I'm just not sure about the legal implications of scraping or collecting news articles etc. in one place :/
@szbalint the last bit, the e-mail thing? You're kinda describing Mailpile: https://www.mailpile.is/