The Mass Outsourcing of Human Reasoning

I think myself a romantic person. Maybe not in my love life, but towards life in general. I am a simple sysadmin, there is little that I specialize in; I am a jack of many trades and master of none, but that's my job. I need to wear many hats to get the job done and if I can do anything, it's change hats to complete my assigned tasks. I dabble in many things and have sundry hobbies to fill every last second of free time and drain every last red cent from my bank accounts if I don't keep my habits in check. Seems odd to find myself thinking that I romanticize life after my last entry in which I described the feeling that life often times leaves me behind, but tonight things are hitting just a bit different.

The devil is in the details. If I am given enough time with a task or endeavor, I will continually improve on it to the maximum extent possible over time. I am rarely, if ever, happy with a final product until I have had days to look at it from every angle and scrutinize every detail. It's a process. I am not a graphic designer but I like to think I'd have made a good one had I chosen it as a career path. My dropbox is full of directories each containing their own projects and sub-projects, each of those in turn containing numerous revisions of a single file with such suffixes as "RevA", "RevF", "RevF9", and so on and so forth. One of my aforementioned hobbies this year is Christmas cards which I designed and had printed by a local print shop, along with return-address envelopes. I shot the front and inside pictures I used in the card on my Mamiya M645 and Canon A1, respectively. I worked on a new signature to use in the cards, and had a signet made so I could seal the envelopes with wax. On many of the cards I added personal notes to my friends, and had I thought about it before I began the process, I probably would have designed a rubber stamp to use inside the card under my signature (I'm saving that idea to add for next year's cards). I enjoyed every moment of writing, addressing, signing, sealing, and applying postage stamps to each card. Every time someone mentions the cards (which are still likely being delivered as of this writing), it also brings me joy when they mention the small details which they enjoyed about it. Despite the fact that I don't specialize in any particular thing, I enjoy details.

I recently came across a reddit thread in which the poster explained their frustration that a college professor had cancelled their final assignment, instead ending the course by giving the entire class full credit for it. For most of the students, it was probably a relief, but the poster was frustrated because they had spent weeks of work on the assignment which now wouldn't really matter, even though they received the credit for it. The professor's actions were not born of kindness, but instead were a salient warning towards his students on the dangers of using generative AI to complete their coursework. You can read the contents of the post and the professor's message to the class here. The excerpt from his warning that struck me the most was the sentence: "Writing is not an arbitrary hoop to jump through; it is one of the few durable intellectual skills that will serve you across professions, relationships, and civic life. When you outsource that work, you weaken not just your ability to write, but also your ability to reason, to communicate, and to advocate for yourselves."

In my job, I have a coworker who has worked in the same position I work now on a different project for the better part of a decade now. Over the past few years of with him, I have noticed that more often than not when I stop by his desk, he's got OpenAI's ChatGPT open on one of his monitors. I will readily admit that I have used LLMs on more than one occasion in my job and it's helped me a few times, but in my sparing use of it I have seen just what it is capable of doing, or more accurately, what it is not capable of doing. Like any other tool, the results that a large language model produces are only as good as the parameters put into it. The first parameter in our context is the data on which the model is trained. I don't have any hard evidence but I assume that the models I've used are trained on a small amount of publicly available documentation and a large amount of stackoverflow and stackexchange threads. The problem with the latter (if my assumption is correct) is that the model isn't just training on the correct answers, it's also ingesting the wrong answers. The second parameter I'll mention quickly is the information you provide it. The more information about the problem you provide, the closer it will get to useful output (but that also comes with its own data sovereignty concerns). LLMs are most useful to me in helping find niche knowledge that is buried in the deep pages of my search engine of choice. Very rarely is the output of the model ever accurate when it comes to code, but there have been a few times where it did bring up an obscure bug or little-known issue that helps me look at the problem from a different angle and eventually find the solution. Like everything else in the administrator's backpack, LLMs are a tool. When you use a tool in the right context it can be incredibly helpful, but when you use it in the wrong way it can be extremely detrimental to your progress towards a goal. Using the wrong size phillips driver on a particular phillips head screw can result in rounding out the the head of that screw.

Back to my coworker: without going into detail that isn't necessary to divulge in these pages, like many others he seems to have become dependent on ChatGPT to complete his tasks. The problem is that those tasks are now rarely ever completed. Due to some extenuating circumstances, one of these tasks had to be assigned to another coworker who noticed how easy this problem really ended up being to solve. He also noticed just how far away the coworker in question was from even stumbling upon the obvious issue. An employee whose work has historically been well-regarded has all of a sudden become unproductive to the point that it is affecting other team members. This coworker has, by outsourcing his work, lost his ability to solve basic problems. Tying it back to our screw driver analogy, he rounded out the indentions on the screw head for so long that he completely missed the fact that the screw head is also perfectly shaped to fit a socket.

There is no logic behind the output of these models, it is just regurgitating what it was trained on. By contrast, man should learn the underlying logic of whatever he's learning so he can become an effective practitioner of that craft.

The devil is in the details, and when you offload the details to some other entity you will lose the ability to see them. When you look at the details in AI generated works, they just don't line up. The fingers will be portrayed in unnatural positions, the faces will be inhumanly smooth. The paragraphs will lead you down such abstract and detrimentally colorful prose that you will completely miss the fact that the conclusion was so ambiguous that Malachi's house cat could have found some kind of inane meaning in it. This is where artificial intelligence fails by living up to its name. The output is designed to look human, but cannot be human. LLMs and generative models generate human sounding and looking works, but they will never be able to accurately portray the essence of life in the way that we can. Any cursory scroll of any number of social platforms will reveal that we're already treading the dangerous path the professor so aptly warned his students about. We are surrendering our agency to automated systems designed not only to abuse our preferences, but systems that are created to learn the best way to bring our minds into captivity to be monetized for the gain of faceless corporations and investors.

It's the details that give human touch to artistic works and life to our output as people. It's the small inconsistencies and variations in the pen strokes of a handwritten note that imprint the life of the author on the paper. It's the way the leaves hang from a tree in a photograph, all dangling from the branches in their own individual ways; how a skilled writer guides the reader towards a destination, and how an artist somehow pours soul into a painting without causing atrophy of the observer's imagination. There is a disorder in life that is an order all its own which can only be begotten of life. Artificial intelligence can only create artificial life, and artificial life is just that: artificial.

I do not know what the solution this problem is. All I can do is try to improve my own ability to reason, think for myself, and hope that by painting life with my own canvases, I can help entice those who continually exist in an artificially manufactured mind back into reality.

If we as a society continue to outsource the life we should be contributing to the reality we were created to inhabit, we will cease to exist in it.