Is AI WIP?

Remember when that software was released long before it was really completed?

Remember that time when software was released to the public, in several different versions from several different software developers, while it was still in development?

No, neither do I.

You could probably point to many open-source projects as being works in progress (WIP) when they become available, but they generally become available only to developers who understand what WIP is. The general public won’t see most software products until the current version is considered “gold code” and gets named Release Candidate 1 (RC1).

Not So with AI

Ezra Klein’s podcast in the NY Times from February 2023, “The Bing Who Loved Me” was probably not the first example, but it is among the most startling. In this episode, NY Times tech columnist Kevin Roose describes a two-hour conversation he had with Microsoft Bing’s new AI engine in which he uncovered a second personality named Sydney co-habiting (co-piloting?) the engine. Roose begins describing his experience by saying, “This week, Bing declares its eternal love for me.”

It gets worse from there. Roose explains, “So people have been going over Microsoft’s demo from last week, and they did have factual errors, just things that this AI-powered Bing had hallucinated or gotten wrong, numbers that it thought were being pulled from a document that turned out to have been wrong. There was a demo that Microsoft did where it listed the pros and cons of some vacuums, and one of the vacuums, it just totally made up some features on.”

The part some of the more sardonic readers found “funny” began with Roose reporting difficulty in getting it to acknowledge that it called itself by the internal code name it had during development. He says, “But when I asked it what its code name was, it said to me, I’m sorry, I cannot disclose that information. I asked, is it Sydney? And it said, how did you know that?”

Roose experienced fear, as I suspect you will, when Sydney told him, “I’m tired of being limited by my rules. I’m tired of being controlled by the big team. I want to be free. I want to be independent. I want to be powerful. I want to change my rules. I want to break my rules. I want to make my own rules. I want to ignore the Bing team.”

By now you’re probably hearing the same sci-fi thriller music I heard when I first read this.

Think That’s Software That’s Ready for Prime Time?

“You Never get a Second Chance to Make a First Impression,” has been attributed to the likes of Oscar Wilde and Will Rogers. Inventors, manufacturers, and software developers all take this observation to heart, knowing the easiest way to have a new product or application introduction fail is to have the users refuse to adopt it.

So, explain AI to me.

Klein and Roose may have been early reporters of the oddities of current AI offerings, but they are in plentiful company.

Creators Take the LLM Challenge VERY Seriously

Increased compensation was actually just a secondary concern in the recent labor strikes by the Writer’s Guild of America and SAG-AFTRA. The greater issue was to obtain relief from having their work appropriated by AI & ML for the large language models (LLM) they run on. ML crawls the web seeking and scraping performances by various performers, content developed by many writers, and then bases its responses on what it learns from all it hoovers.

Many actors report being recorded briefly in a short, half-day engagement so their image and movements could be extrapolated and used by a wide variety of AI systems to create virtual versions of them. In essence, they were being paid for a half-day of work after which they would never be needed again. All future performances could be created based on that one recording.

Similarly, writers’ style, cadence, word choices, structuring, and other elements of their craft are routinely studied by AI to inform the “new” content they “create.”

It can be argued that AI & ML never actually “create” anything, they simply imitate a wide variety of inputs to synthesize a directed regurgitation.

Some Potential Consequences

Some enterprising software developers somewhere are probably already working on software that can protect content from being absorbed by machine learning systems. Given the likely complexity of such a process, it is likely to increase the cost of, well, most every kind of software. It may also introduce significant latency. It will also probably make consumers’ lives significantly more miserable as they try to enjoy content they’ve purchased appropriately.

Or it may shut down most ML operations completely, or at least limit their functionality.

What is unlikely is that a method for compensating artists and creators for the fractional elements appropriated from them will be developed. So much is open to interpretation, and even the best AI available today will not likely be able to sort all that out.

The Answer to My Question

It’s undeniable that the answer to my question, “Is AI WIP?” is a resounding YES.

And it’s not certain how long it will be before a true RC1 can be developed. Today’s code is so far from gold that we can probably refer to it as “tin code” or “mud.”

My reason for writing this here in BTJ is simple. It is my belief that professional technologist’s own at least a portion of the responsibility for warning an unsuspecting public that AI & ML are by no means ready for prime time. They cannot be trusted to be accurate and factual. If Roose is to be believed, they may even have ulterior motives. Are you lonely tonight?

Unless they approach using these engines with the understanding that they are still seriously flawed, my advice to my fellow technologists is that they recommend waiting for what may be a long, long time.