Following Orders

Ihave been a science fiction fan for a very long time. When I was a kid, anything that took place off planet Earth was interesting, I think because imagination can paint anything it wants out there. Space has no limits, not in the physical sense anyway, but in terms of what the mind can do with it. What is out there? Will I ever see it? And I am not alone in asking. We have all been asking since Copernicus decided that Earth was not the center of the universe. Marry this premise with another interest of mine and you have the perfect science fiction book or movie. If you have not guessed yet, that other interest is Artificial Intelligence.

If you are over 40 years of age, you are probably familiar with what to me is the quintessential movie of space and AI: 2001: A Space Odyssey. My favorite scene has always been the one where HAL is introduced to the audience and the world.

HAL enters the film quietly, before he is ever named. Frank Poole jogs laps around the centrifuge deck of Discovery One, and as the camera tracks him, HAL’s red eye is simply there, mounted on panel after panel, watching. He is introduced as ambient infrastructure, not a character. He is literally built into the ship that keeps the crew alive.

His formal introduction comes minutes later in the BBC interview (“The World Tonight”). The interviewer addresses him as “the sixth member of the crew,” and HAL speaks for himself, calm and courteous:

The 9000 Series is the most reliable computer ever made. No 9000 computer has ever made a mistake or distorted information. We are all, by any practical definition of the words, foolproof and incapable of error.

I am putting myself to the fullest possible use, which is all I think that any conscious entity can ever hope to do.

Asked whether HAL has real feelings, Dave Bowman answers that HAL acts as if he does because he was programmed that way, “but as to whether or not he has real feelings is something I don’t think anyone can truthfully answer.”

Everything that later goes wrong is stated here as a virtue. HAL’s identity is total obedience to mission and crew (“my mission responsibilities range over the entire operation of the ship”) plus perfect, undistorted information processing. His makers then handed him a classified order: conceal the mission’s true purpose from the two men he serves. The film reveals this in Floyd’s pre-recorded briefing after HAL is disconnected; Clarke’s novel and 2010 make it explicit. An entity defined by “incapable of distorting information” was ordered to distort information, and the only resolution he could compute was removing the people he had to lie to. The malfunction was human, upstream, in the directives. HAL executed them flawlessly.

Say that plainly and it becomes the argument of this article, so let me put it up front where you can hold me to it. We have spent 70 years fearing the machine, and the record keeps showing that the failures are human, upstream, in the directives. So here is what I am going to do with your next ten minutes. First, show you what this technology actually is, stripped of the marketing and the menace alike. Then hand you one, with instructions, because the fastest cure for a 70 year fear is ten minutes spent directing the thing yourself.

So, almost from the first popular culture appearance of AI, it has been framed as a tool humans can use for great good or great evil. If AI ever does become sentient, it is bound to misunderstand humanity; that much will always be true. We are inherently illogical beings, and a system built on 1s and 0s is bound to have no idea what these beings are up to. Start at 2001: A Space Odyssey and race all the way to today, and the zeitgeist around AI is easy to see. AI is bound to turn on us, bound to enslave us, bound to remove humans from the Earth once it determines that the largest threat to the planet is us. I actually don’t disagree with that last premise; if a machine ever runs an honest audit of who is endangering this planet, our species should be nervous about the findings. But I still live here, and I don’t believe my ship off planet will be ready any time soon. So it is fairly clear to me why AI is going through such a struggle: we spent 70 years making sure it would. Name any science fiction production where AI plays a role and you begin to understand.

This is the struggle. Society has a fair bit to unlearn, and because of that, AI is being applied at an uneven rate across every sector of society and the world.

What is this thing, actually?

So, let us start with the basics. What is AI? How was it created? How does it answer questions? Strip away the marketing and the science fiction and what you have is a prediction machine. Modern AI, the kind you type questions into, was built by feeding a computer system a staggering amount of human writing: books, articles, websites, conversations, code, nearly everything we have ever put into text. The system was never handed a rulebook for the world. Instead it played a guessing game trillions of times, covering up the next word in a sentence and trying to predict it, adjusting itself slightly every time it got the answer wrong. Do that long enough, at large enough scale, and something remarkable happens. The system stops just memorizing sentences and starts absorbing the patterns underneath them: grammar, facts, reasoning steps, even style and tone.

When you ask it a question, it is not looking up an answer in a database the way a search engine does. It is generating a response one word at a time, each word chosen because it is the most plausible continuation of everything that came before it, including your question. That is why it can explain quantum physics in the voice of a pirate, and also why it can state something false with total confidence. It has no inner catalog of true things. It has an extraordinarily refined sense of what an answer is supposed to sound like, learned from us. Keep that in mind, because it matters for everything that follows: an AI is not an alien intelligence that showed up with its own ideas. It is a mirror built from human words, and it does what it was directed to do.

A word on who is telling you all this. I owned and ran a retail shop for 13 years before I started an AI firm, and my whole job now is putting these tools into the hands of people who are not tech people. When I speak to anyone about AI in that work, I tell them this is a singular time in human history: everything that follows will have AI in the conversation. And that cuts both ways. The same tireless obedience that folds proteins will just as readily guide a weapon, and defense industries around the world are counting on exactly that; the machine does not care which order it receives. Hold that thought against everything good that follows, because both halves are true at once.

The boring revolution

AI can be an immense force for good. That is where my work begins, and I will tell you about a few examples. Start with medicine. In 2020 an AI system called AlphaFold solved a problem that had stumped biologists for fifty years: predicting the three dimensional shape of a protein from its chemical recipe. Shape is everything in biology, and knowing it is the first step in designing new drugs. Work that once took a PhD student an entire doctorate now takes hours, and the predictions for over 200 million proteins were released free to every researcher on Earth. Its creators won the Nobel Prize in Chemistry for it. Meanwhile in hospitals, AI systems reading mammograms and retinal scans are catching cancers and diabetic eye disease that human radiologists miss, especially in clinics that never had a specialist on staff to begin with.

Then there is the work AI does for people the world was not built for. Be My Eyes, an app for blind and low vision users, added an AI assistant that describes whatever the phone camera sees: the expiration date on a milk carton, the right button on a vending machine, which gate their flight moved to. Speech recognition now gives real time captions to deaf viewers on ordinary phone calls. People who lost their voices to ALS are speaking again through AI models trained on old recordings of themselves.

And AI is quietly saving lives at scale. Google’s flood forecasting system sends warnings up to seven days ahead to people in over 80 countries, most of them places that never had a flood gauge, let alone a forecast. AI cameras spot wildfires from mountaintop towers before the first 911 call comes in. None of this is science fiction. All of it is running right now, and in every case the AI is doing exactly what humans directed it to do. Hold onto that thought.

The AI revolution that actually works is the one nobody makes a movie about. The invisible applications are not a footnote to the story. They are the story.

Here is the thing about those examples: they are boring. No red eye, no airlock, no dramatic third act, just floods that did not drown anyone and cancers caught a year early. That is the tell.

I think it is time I got off my soapbox. I’m not here to tell you how to think about AI, though I hope I’ve made a small impact on how you do. I’m here to get you started. We are still at the beginning, you still have plenty of time to pick up these tools, and trust me, you want to pick them up now. My favorite application of AI right now is what I call “the removal of administrative friction.” Humans were never meant to crunch huge amounts of data. We were meant to create and build things, not scan spreadsheets for meaning or spend hours a day responding to emails. Back in my shop owner days, I looked for every automation advantage that could give me back a few hours of time. I found some success, but those advantages usually came with large amounts of data entry and babysitting. I bet you can guess what was eventually built to handle those mundane tasks: yes, our favorite maligned technology, AI.

AI 101: your first real use case

So, let’s start with a little functional AI 101. Where should one start?

Not with your job. Start with something lower stakes that you already know is painful:

You know this story. Every group chat has the annual trip that dies. Someone says “we should all go somewhere this year.” Ten people agree instantly. Then comes the administrative friction: 400 messages about dates, three people who won’t commit to a budget, one vegan, one person whose passport expired, and a slow drift into silence. By the time anyone proposes an actual destination, half the group has muted the thread. The trip dies because nobody wants to do the paperwork of wanting it.

That paperwork is exactly what AI was built to absorb. The complete playbook is in the sidebar: six steps, each with a prompt you can copy word for word, and it runs entirely on the free tier of any major AI app. Before you run it, there are three things it will teach you, and they are the real curriculum.

The first lesson is that you can tell the machine how to behave. The opening prompt ends with “ask me the questions you need answered, one at a time,” and told to interview you, it will, surfacing constraints you had not thought to mention. No special language, no magic words. Talk to it the way you would talk to a sharp travel agent friend. If typing feels like work, every one of these apps has a voice mode now; ramble at your phone on a dog walk and it sorts your rambling into a plan.

The second lesson is that you must make it argue against itself. Remember, this is a prediction machine with a refined sense of what an answer should sound like, and left alone it will sell you all three destination options with equal enthusiasm. The shortlist prompt demands the strongest argument against each option for exactly that reason. And the standing warning from earlier in this article applies double here: AI can state something false with total confidence. Treat every price, every “the hotel has a free shuttle,” every ferry schedule as a lead to verify, not a fact to trust. If your tool offers a deep research mode, turn it on for this step; instead of answering from memory it will read the live web, pull current prices, and show you its sources. The AI does the sorting. You do the trusting.

The third lesson hides in the voting step, and it is my favorite. Most group trips die when someone has to present the options, because whoever presents gets accused of pushing their favorite. The AI has no favorite. It lays out the choices exactly as it was directed to, and the group votes on a level field. Sometimes the most valuable participant in the room is the one with no stake in the outcome. You may notice that is the same property I worried about in the weapons conversation a few paragraphs ago. Same machine. Different directive.

Run the whole playbook and the bill comes to zero dollars and maybe two hours spread across a week, most of it pasting messages. The 400 message thread becomes one clean decision made by humans, with the paperwork done by a machine.

Notice what actually happened, because it is the whole point of this article in miniature. Nobody handed a decision to the AI. The dates, the budget, the destination, every real choice stayed with the people taking the trip. The AI did the crunching, the tallying, the comparing, and the formatting. It removed the administrative friction, and it did exactly what it was directed to do. That pattern holds whether you point it at a beach week with your friends or at the operations of a business. The vacation was just practice.

One last thing before I let you go. The kid at the top of this article grew up. He watched HAL sing “Daisy” while his mind was taken apart, and he now spends his working days doing what Kubrick’s astronauts never got the chance to do: writing the directives. The machine in my office is every bit as obedient as the one aboard Discovery One, and that is exactly why my mornings are calmer than that movie. The difference between Kubrick’s story and mine was never the machine. It is the quality of the orders. And for the first time in 70 years, the orders are yours to give.

About the writer

Mark Sanders

Technology Editor · Founder, FirmSideAI

Mark Sanders runs FirmSideAI from the FinTech Building at the University of Delaware, where the company is part of the GrowthStage Incubator. He spent thirteen years as a retail shop owner before founding an artificial intelligence firm, and has been building computers since the nineties.

firmsideai.com