AI- The Pachyderm in the Room

In the tech world, it’s all anyone wants to talk about, and even in other spaces, it’s a frequent topic of discussion.

Here’s a brief explainer that goes into what AI actually is, for those who have successfully avoided the discussions. At its surface, it’s actually much simpler than people would believe.

An AI engine is basically a massive mathematical web of language patterns gathered from millions of documents that have been fed to it—by means legal, illegal, dubious, or otherwise, depending on the developer. As it reads, the AI doesn't store the documents like a database. Instead, it breaks the text into tiny chunks and maps them mathematically, ranking terms based on their proximity to other terms. An “affinity” score, called a vector, is produced for each.

Because of this, an AI can be considered a very advanced autocomplete engine. If one were to type “not calling attention to the elephant in,” it’s mathematically highly likely that the first autofill it would suggest is “the room.” This is an oversimplification, of course, which is why it costs millions of dollars and significant computer resources to process these models.

The extreme sophistication comes in with Very Large models and specialized engines trained on multitudes of industry-specific data. When an AI is “trained,” it goes through phases: first it learns core concepts from raw data, and then it is fine-tuned with rules and policies for satisfying output. Guardrails are built in to avoid inappropriate content. To the extent that guardrails are effective is a topic all unto itself; it’s best to say that it’s far easier to instruct a system what it can do versus what it shouldn’t, as those negative possibilities are endless. Because AI is just connecting mathematical concepts, it can generate novel combinations and processes that no one ever thought of—which makes it highly powerful, but also prone to confidently making things up. We call this ”hallucination,” a very apt term.

Now that we have that covered, let’s go on to the state of AI and it’s current place in Western technological society.

At the risk of repeating myself ad nauseum, it is a very powerful tool, but the mathematics of how an AI reasons is not fully understood. Whereas it can be a tool for astonishing insight and be implemented to creating functions, agents and processes, it can effectively replace the output of a large segment of “work” that humans have been traditionally engaged in. Company stockholders demand performance for their investments, and that’s where the layoffs come in. When a person’s $100K/year+ salary can be replaced with a few thousand dollars worth of automation, it’s easy to see where people are made redundant. Earlier pre-AI programming could replace some of this already, but the rules were hard-coded, and one-off exceptions that would occur couldn’t be sidetracked until a human with judgement as part of the process, the labor replacement was minimal.

When AI came into the mix, it had deep reasoning and goals to ensure the right outcome would occur and largely replaced the human supervisory necessity. Nearly anything that doesn’t require flesh to hold a tool or produce a product requiring an untrainable aesthetic has become a target. However, like humans, AI is imperfect, and the hallucinations and imperfectly trained guardrails can make decisions that we would abhor.

In 1979, an IBM slide on automation was famously enshrined on the internet. In simple typewriter text on an overhead projector, it stated,:

A COMPUTER CAN NEVER BE HELD ACCOUNTABLE

THEREFORE A COMPUTER MUST NEVER MAKE A MANAGEMENT DECISION

Somewhere along the line, we lost this wisdom. So who is now rendered responsible for errata with consequences? Logic would state the company providing the AI system, and possibly a joint liability for the organization using it, if it were used sloppily or without oversight. The current situation of Flock cameras having a very low positive identification rate among license plate reads and alerting law enforcement to dangerous drivers, whose only crime was to have their plate misread and receiving a felony stop for their trouble, is one of the more visible symptoms in American society (And even when Flock is working correctly on a technical level, there are many other legitimate gripes regarding this technology, but this outside of my scope).

Obviously I cannot be considered “anti-AI,” as I am near to releasing an application that has extensive artificial intelligence features for writers and authors organize, score and provide associations for their texts in writing. However, I draw the line at the actual product: The AI is forbidden from touching the prose in any way, allowing the creativity, skill, style (and yes, idiosyncrasies and flaws) of the Author to shine through. This was a very deliberate decision here at GhostTractor.com. The writer is ultimately in charge of what they produce, and is accountable for it without equivocation. We’ll discard the conversation about AI generated text being flat, un-stylized and uninspired for now, with only the positing that my ongoing belief that AI-generated prose-producing software applications are complete slop, and a shortcut to some sort of minor literary visibility, at most.

This tradition of respecting IBM’s insight into responsibility is sadly lacking in so may boardrooms and management suits. I can’t help but think that many of the executives in corporate offices simply do not understand the limitations of the technology they implement. I am awaiting the first of the very large lawsuits that will arise when a class action is filed for mismanagement. It will happen, but that certainty is tempered with the knowledge that these court actions might never come to public light, or even be suppressed, due to the size of the companies involved and the monies paid for representation pales in regard to the operational income of these systems.

Frank Herbert’s DUNE series of books is predicated on a (fictional) history of a machine rebellion of AI entities that was only successfully defended after a somewhat Pyrrhic war. After that, it became a near-religious rule that “a machine should never be designed in the form of the human mind.” This was written in the late 1950s, and computer technology of the time couldn’t be even considered “insightful,” let alone “intelligent.” Herbert’s own insight, however, was brilliant, and even prophetic- possibly why his novelizations and movies are seeing a resurgence in popularity in recent years.

I’d hate to rely on the old trope of “Education is the only solution,” but I cannot think of anything else that would offset the issues we are dealing with at the moment. But while money is the only consequence of one day seeing harder legislation setting boundaries of AI management accountabilities and failures, we’re going to continue to see continued Western-culture labor and economic losses. AI is not the perfect replacement for human judgement and accountability.