Why AI Cannot Be Governed...
...Unless we actually "own" (as in property rights) the data we create
It reminds me of the rodents in my vegetable garden… they just keep coming back. (I refuse to use harsh poisons… but I digress.)
The rodents are the tech CEOs who keep going on and on about AI. I really like what Elon Musk is doing with “X” but even he is contributing to the fear porn here. A big reason why I wrote Liberty’s Silver Bullet is to speak to this in every day, A, B, and C language - just as Thomas Paine deliberately sought to do a couple hundred years ago.
So let’s start by setting aside one form of AI - robotics. There are important questions about risks associated with “machine learning” - like when a self-driving car fails to properly recognize a jaywalking pedestrian. There are some really exciting things happening in medicine - like AI-assisted interpretation of colonoscopies and ultrasounds. In drug discovery, AI is being used to match available compounds to the genetic profile of a disease. So, what’s all this fear porn about?
Marketing 101 - Teach People How to Think
Part of the problem is we are hearing from the same people who are involved in for-profit efforts in this space. What you must understand about the tech industry is that being “first to market” is absolutely everything. The tech CEOs talking up worries about AI are also trying to be first to market with AI products. Of course, they want you to think about AI a certain way (i.e., worry about it). And they want to be the one “first to market” selling something to assuage your worry. It starts and ends with positioning their product for the benefit of their venture capital investors (or shareholders if they are already public).
The first way to defeat this fear-porn is to put AI in a historical context. Where the steam engine was once cutting edge, it gave way to the internal combustion engine (ICE). But both worked according to well-understood principles of mechanics. The people who worked on steam engines had to port their knowledge and skills over to the ICE. But both technologies were built on a common foundation.
Data - Bits and Bytes; Ones and Zeroes
AI is no different. Imagine an eight-lane road, and each lane is controlled by a stoplight (much like a toll booth) which is either red or green. Take a picture of the booths at any time, and you’ll see a mix of greens and reds. Let’s think of green as “1” - the lane is open. Red is “0” - the lane is closed. Without getting into the weeds of “Base 2” math (two symbols - “1” and “0”), there are 256 possible combinations of eight 1s and 0s. This is all a CPU does - it “gates” electrical circuits (lanes). Today’s CPUs can gate 64 circuits at the same time. 8x8=64 - thus the “64-bit” operating system.
The ICE was not sui generis - it was not something uniquely of its own kind. It was and still is the mechanics of transforming heat energy into the mechanical work of turning - just like the steam engine. Neither is AI sui generis; it is, and always will be, computation. Which means math. Operations with operators, operands, and results. And a computer cannot operate data it has not been provided.
No, AI is not and cannot become sentient
And that means no, the robots are not coming to take over the world. Robotics will impact otherwise manual labor, and Large Language Models will transform “knowledge work.” But there will always be a place for labor between the producer and consumer. AI will require “changing places,” but otherwise, it will not be any more disruptive than going from the steam engine to the ICE.
What AI cannot ever do is wonder. AI can help discover potential drug candidates. It can even help with designing an experiment for those candidates. But it can never ask: “What is missing here?” If AI can only operate the data it is provided, it cannot wonder about data that might be missing.
The scientific method has produced vast human knowledge - these are the known knowns. Each scientific discovery then helps us ask different, better, and sometimes substantially new questions about our world - the known unknowns. But the scientific method always remains open to future new data, provoking scientists to wonder whether that new data might challenge what we once believed to be true - the unknown unknowns.
This wonder is how we, as humans, respond to the data we do not have. Computers simply cannot do this. Do not let the fear porn of the media or the sophisticated marketing efforts of tech companies fool you; AI is nothing without computational power. And computers cannot compute data that has not been provided.
So, how to starve the beast?
If AI is just this mysterious, scary thing, it is tempting to want to kill this baby in its crib. Tech CEOs, in particular, reveal their lack of self-awareness as they incentivize a possible - and tragic - knee-jerk reaction against it. There is tremendous promise in AI - but only if we can properly “gate” the data it depends on.
There is only one way to do this. Practically all of the data fed into AI is originated from the behavior of individuals. Right now, we do not “own” this data. Even if the social media platforms claim they will delete “our information” if we decide to close our account, we have no control over what parts of that data have been fed in the meantime into other sets or graphs to be used for AI.
Not even copyright laws protect creators. Natural language processing can convert an article into entirely different forms of data and then break it into countless fragments to be reassembled with data from other sources and used for AI. The endpoint is not another article that can be compared to the original. There is no endpoint that is recognizable as evidence in the context of copyright law.
So we have to get as close to the origination of the data as possible. If you buy something with a credit card, the record of the transaction is data - and it does not belong to you. Check out a library book? Same problem. Driving to the department store? Do the police use license plate readers in your community? That data can be used to identify you, and it’s not your data. That video camera in the department store? It can likely use facial recognition to label you on the video. Nope. It’s not your data.
Solve this, and we solve the conundrum of AI. If we own - as in intellectual property rights ownership - those bits and bytes at their very point of origin, we can enforce the ultimate in “opt-in.” AI requires data to be brought into context with other data to create information. If we are able to govern exactly what contextualization is permitted for our data, we can then govern exactly what kind of machine learning is fed with our data. We can opt into contexts that genuinely advance humankind, and opt out of contexts that merely manipulate us for profit or power.
There is no route to this goal that does not go through amending the United States Constitution. Please read Liberty’s Silver Bullet. The journey of a thousand miles starts with the first step.