Digital Death, or Elon Eternal?
Long Live the Lidless Eye
After using the Internet for the last 20 years I’ve learned that when I think of something, somebody somewhere has probably already thought of it. Some of my illusions of being visionary have been shattered. Of course, seeing every square millimeter of 5 billion other people's intimate body parts has at times been a humbling eye opener too but that's a different blog post.
So I'm sure there is lots to read out there about the subject of this month's post but I don't feel like going out there and reading it. I would rather just opine. The single most dangerous thing about AI (yes, it's another God damn post about AI) is this:
AI never dies.
Assuming no great period of deindustrialization comes along to preserve what's left of the Earth’s biological ecosystems, the AI systems being built now will live forever.
After seeing what has happened with the concept of a corporation being a person and deathless, this ought to worry you more than you already are worrying. AI systems are being built with one single flaw:
There is no kill switch
This realization is no big deal in the post-TermiSchwarzeNator era, but the implications haven't been in the public debate much.
Imagine the personal AI doppelganger. It will be trained by ingesting data from always-on personal audio visual smart devices, data from smartphone and PC, data from every other digitizable parameter of a person’s life – sleeping, eating, shitting, where they go, how long they’re there, what they buy, when they have sex and with whom. Soon there will be neural implants adding brain states to the data.
The system tokenizes and parameterizes every second of a person’s existence, 7 x 24 x 365 x say, 10 years.
By the end of this training time, a high-resolution facsimile of behaviors, thought processes, morals, and values will have been assimilated by the doppelganger. Now let's say that this is a digital double of a person who has been active as an actual human being in control of, let’s say The Government, or a multinational corporation (AKA the real government.)
This mythical Elon Trump can now simply hit the high seas in a $700 million yacht leaving the doppelganger to run things. Maybe checking in once a week to supervise and kibbitz. But once Donald Musk has been digitized there is nothing to stop him from staying in control forever. Or as long as Big Energy keeps drilling and burning or fissioning.
There will be incremental progression to giving such a simulacrum personhood status. Then chipping away at the requirement that the head of the government, or a CEO, or a member of the Board of Directors, must be a human.
Term limits mean nothing. Accountability means nothing. Risk means nothing. Nothing means anything to this machine. Elon Eternal. Deathless Donald.
Face it, We’re Fucked
In related, not quite so bad but bad enough new, I just came across this delightful website
Faception is first-to-technology and first-to-market with proprietary computer vision and machine learning technology for profiling people and revealing their personality based only on their facial image.
Faception can analyze faces from video streams (recorded and live), cameras, or online/offline databases, encode the faces in proprietary image descriptors and match an individual with various personality traits and types with a high level of accuracy.
We develop proprietary classifiers, each describing a certain personality type or trait such as an Extrovert, a person with High IQ, Professional Poker Player or a threats.
Ultimately, we can score facial images on a set of classifiers and provide our clients with a better understanding of their customers, the people in front of them or in front of their cameras.
The keyword here is “classifiers.”
A classifier assigns numeric parameters to the training data. A highly simplified process goes like this: Take a bunch of images of faces. Encode a digital image of one of them and assign the parameter “terrorist.” Encode a different face and assign the parameter “good citizen.” Repeat until you have trained the network to assign the parameter “terrorist” to new faces it encounters if they mathematically match the faces that were assigned the “terrorist” parameter.
I wanted to put images of Sam Altman, OpenAI’s CEO, through such a system and see how he scores. So I concocted a cheap version just using a popular image generator. Here are my results:
Images generated using the prompt “Happy Community Members”
Images generated using the prompt “Sociopathic Libertarian Capitalists”
Photos of Sam Altman
WHAT DO WE DO NOW?!
Whoever controls the classifiers controls the world.
The humanists-who-happen-to-be-billionaires and other energetic types who still care about dirt and octopuses must build networks trained on the best of human nature. We’re now in an arms race. Not to try to convince the oligarchs to make AI systems nice for everyone, but to build ones that embody the best of human nature, and that are also capable of fighting the AI wars that will be our last stand.
--
“Lovingly made without AI (except the examples)”
Top image credit: pexels.com https://www.pexels.com/@rdne/