Mechanical arts are of ambiguous use, serving as well for hurt as for remedy.
It’s been a long time since I’ve worked in the field of ML (or what some call AI), and we’ve come a long way from simple text classification to what’s being casually called generative AI today. While the technology has made many advances, the foundational concepts of machine learning have remained analogous over time. ML depends heavily on a large set of training data, which is analyzed to pull out its most interesting and defining features, and this becomes the basis for training a model. The process might involve parsing text, or performing analysis like object identification or analyzing stylistic features in art. Each of these is, in itself, a smaller – but mathematical – process. I experimented with a primitive form of meta-level learning in text classification several years ago, which may help convey the general idea. This identifies “features” of the reference sample being trained. The features this process pulls out can be simple, like words in a document or pixels from a handwriting sample, though today can be more sophisticated “critical patterns” correlated to literary authorship or artistry, such as patterns within art and music composition, sometimes stored in other models. Whatever the content is, the purpose of the training algorithm is to identify patterns and correlations across the data to build a weighted or structured model. The most interesting patterns in the training data influence weights or probabilities, creating a hidden layer: millions of “gears” that converge to compute the most statistically significant outcomes. In this sense, the term “learning” is a bit of a stretch; what’s happening is more along the lines of statistical transcription of a set of features. Feature selection is one of the key differences between various ML models, and why you have some constructing music, while others render art. The math is pretty consistent – more sophisticated machines like neural nets are typically trained using backpropagation and gradient descent, while other machines such as chat bots and text generators might use weighted Markov models or Bayesian networks. These approaches have been applied to everything from natural language processing and handwriting recognition, to today’s work in genome sequencing and autonomous driving. Still, these traditional forms of machine learning are not much more than a sophisticated pattern recognizer. It is largely a deconstructive process with weights and statistical magic.
Today’s generative AI still goes through this type of deconstructive process, but also has a formative element. Where these new approaches excel is in going beyond parsing information into a knowledge base, but now also applying a formative process to that information – what we might conflate with intelligence, but still falls short of what most would consider the result of human reasoning. To present the data in some coherent form, this involves training not just the information, but the many dimensions of that information (such as the number of different contexts a word may be used in), or in the context of constructs and critical patterns of that information (ABBA, or 1-4-5, as very basic examples), enabling it to formulate an output in the pattern of an existing set of learned reference samples. Even modern training approaches, such as those used in the transformer model, still require supervised testing to tell the model what bits of its output are garbage, so that the output eventually looks intelligent; it is actually closer to “filtered garbage”. So identifying the pattern of Iambic Pentameter, for example, is still an artificial process. It can be computed adaptively with a large enough data set. Moving from atomic and factoral learning into structural learning allows a system to fingerprint complex patterns much more efficiently. Scale those patterns to music, art, literature, and the more sophisticated patterns that make up our repertoire of human creativity and it is impressive – but still synthesized. Information processing is still very primitive, and lacks many of the traits of human understanding. The inability to conceive tradition, authority, and prejudice is why all of this advanced technology still leaves us with Nazi chatbots. Some would call this confirmation theory, which is an area quite underdeveloped (and the AI reading this wouldn’t disagree). Even the raw objectives of AI are based on human-engineered goals, and evaluated using performance metrics to select the best behavior. This is a very mechanical process. Certain behaviors we may view as creative tasks may in fact be simple randomness introduced into most AIs to avoid infinite logic loops. In short, a lot of what you see is quite the opposite of the autonomous, self-motivated behavior it looks like. Any good AI behaves rationally only because someone programmed good objectives into it. Garbage in, garbage out.
One of the big differences between traditional forms of ML and generative AI is the direction in which the data flows. Traditionally, inputs flow into the system for training and queries. To train traditional systems, you’d suck in “a bunch of other people’s stuff”, and it identifies all of the interesting patterns that are then compared with the input sample. Generative AI takes this a step further, and flips the switch on the vacuum cleaner – and now all of the dirt that was initially fed into the system is shot out the pipe to produce the equivalent of a digital dust cloud of the original training medium. The output of generative AI takes the critical patterns and concepts weighted during the AI’s training and applies some formative computation to produce its own reference sample as a result. Neat-o. Nice parlor trick.
With billions of dollars, this ML scales to perform impressive computational tasks. The risk of this type of system goes beyond the traditional vision of a robot building a better chair, or replacing a worker at a plant. Today’s ML systems are white collar professionals and don’t require mechanical bodies; the computational capabilities of these systems can replace a broad array of professions using the thought product of millions of humans at once – so how could anyone compete with that? No one was ever supposed to, in fact. Doug Englebart, pioneer in the field of human-computer interaction, saw AI’s value more in intelligence-augmentation (that is, IA rather than AI), as a means of assisting the worker. Corporate greed has already led to the recent misapplication of AI, using its advanced capabilities to replace, rather than to augment, humans. Hollywood’s ML generation of “extras” is a quite extreme and literal example of this. But corporate greed isn’t AI generated. AI is replacing employees for very human reasons, and little to do with artificial intelligence itself. Yet correct computer-human interfaces are a fundamental principle that many computer scientists and science fiction authors alike both fear will be broken. Should you hate AI? No, you should hate greed.
The cold irony is this: at a deconstructed level, the output of generative AI represents the collective intelligence of other people’s thought products – their ideas, writings, music, theology, facts, opinions, and so on, likely also including those who lose their job to it. This also means others’ patents and copyrighted works, either directly or indirectly. ML has proven wildly successful at identifying the most effective critical patterns and gluing them together in some coherent form that communicates a desired result – but at the end of the day, all of its intelligence indeed belongs to the other people whose content was used to train it, almost always without their permission. In the end, generative AI takes from the world’s best authors, artists, musicians, philosophers, and other thinkers – erasing their identities, and taking their credit in its output. Without the proper restraints, it will produce the master forgeries of our generation. Should we forget its limitations and begin to rely on it for information, AI will easily blur the lines between what we view as real facts and synthesized ones. Consider a recent instance of this, where an attorney got himself in hot water for citing case law that didn’t exist – AI had seemingly fabricated it, where the attorney thought they were leveraging AI to do research. Imagine the impact to future case law should courtroom outcomes be based on fictional precedent should it fail to be fact checked every time.
There’s an old saying that goes, “Stealing from one person is considered plagiarism. Stealing from everyone is considered research.” As is always the case with new technology, legislation is far behind. I expect that this type of generative AI will face numerous challenges of copyright infringement. Answering the question of how to define a derived work has been a problem long before AI came along. Case law can be found in virtually every industry where humans are accused of copying some prior art from others. Results produced by generative AI today are largely still considered a novelty, however it is very likely that many of its results are already infringing the copyright of others, or at least stealing those ideas. By calling it “generative AI”, we are personifying it by incorrectly accepting the notion that it creates new things. This personification has made AI a sort of Dea Roma to those who don’t understand it. A better term for this should be “formative ML”, as it parses concepts and develops reference samples based on its learned training inputs. If you change the vocabulary, and think of a neural net as a vast array of adjustable gears onto which many inputs are transcribed, the mystique of AI fades away, making it easier to understand the nature of “generated” content. When a copyright war does start, it will likely be a very long battle; due to the hidden layers of AI by nature, it will be difficult to demonstrate provenance of the different training material that contributed to a work unless it is explicitly programmed to preserve this information. Without some accountability, AI companies can simply use the technology as a veiled form of intellectual property theft.
In many ways, AI is ahead of the people who created it in that the hidden layer is still somewhat of a mystery. In the meantime, responsible companies operating in this space should take steps to identify the provenance of the formative concepts stored in their ML representation so that intellectual theft can be eliminated before it even makes it to an output, and prior work can be cited so that these composite works can give proper credit. This should also be expected when forensic accounting of an AI’s output is demanded by the courts. Without this, any corporation looking to adopt generative AI should be aware of the risk that they very well may be generating content with intellectual encumbrances. Taint propagation, content coloring and tagging, or similar such approaches can be developed to identify the training sources that contributed to individual regions of the model. A responsibly written AI should be able to accept one of its own outputs as an input, and provide citations for the works that it was based on; a list of the top sources that influenced those regions of its model. In order to develop any sort of ethics in AI, the “hidden” layer must become transparent to keep the ML within a certain set of parameters (what we might call honesty). These layers must be better understood before generative AI can be reliably used without risk of intellectual property theft, or downright forgery. When auditing fails, we have the ability to dip into the brains of these systems and see the interconnections that contribute to the processing – think of it as a digital MRI. The sources that trained interconnections, however, are typically discarded in the training phase today. It’s important we start holding a system’s designers accountable for what it has learned and when it learned it, just as individuals are held accountable for the content they generate. It is technologically possible to create a system that can account for the origins of its output and provide citations for its own derivative work. This is the moral responsibility of any company specializing in generative AI, and a good place to start with legislation. Because we know the nature of ML, the bar shouldn’t be, “prove to me that this is a forgery”. The bar should be, “prove to me that this is original.” At the moment, AI companies seem to be unable or unwilling to forensically account for the origins of a given output sample.
Twenty years ago, language processing was once called AI – until it wasn’t, once people began to understand how it worked. Hannah Fry made an excellent observation of the use of the term from her very relevant book, Hello World: Being Human in the Age of Algorithms, “If you take out all the technical words and replace them with the word ‘magic’ and the sentence still makes grammatical sense, then you know that it’s going to be bollocks.” Calling it AI is the label we seem to frequently give to impressive algorithms that we don’t fully understand and can’t forensically account for. This is problematic for a technology that is rendering content. Will Smith eating spaghetti isn’t exactly Skynet, but our lack of understanding is why many leaders in the field are calling for caution. It is the fact that we don’t fully understand AI that allows this facade to fool the eye. AI is simply a process that we don’t understand, much like fire probably was to cave men, yet we personify it because, well, we’re the kind of species that sees faces in everything. Psychologists believe people tend to anthropomorphize to help better make sense of behaviors they don’t understand. It doesn’t help that they designed its primitives to resemble the neurons of biological brains. There is no mathematical reason this is necessary; someone simply decided at one point to create “something that can fly” and our first go at it was to make something that looked like a pigeon, rather than a space shuttle. Much deeper advances in AI are yet to be discovered when researchers move beyond the biological constraints we modeled it after, and into the aerospace equivalent of AI design.
Twenty more years from now, we’ll be rolling our eyes at the idea of calling the ML of 2023 intelligent at all, even though our grandkids will have read differently in their ML-generated history books. Completing the cycle of the garbage-in-garbage-out principal will look like the future generation that grows up consuming ML generated content. What adverse effects will this have on human thought? In the meantime, generative AI is such a convincing illusion that you may lose your job soon to a computer – that you unwittingly helped train.