Rationale

My postulation agrees with the prevailing framework for evaluating artificial intelligence in that one cannot objectively observe sentience in another object. We are trapped, so the framework states, in our own minds as observers, no more capable of judging the objective reality of other intelligences than of stepping outside of ourselves. This is a kind of empirical solipsism, and it is in this framework that my harshest critics take refuge, contenting themselves with mimicry and artifice. I would not object to such mimicry if they were honest about it, but they insist on either granting personhood to inanimate objects or moving the goalpost closer by reducing consciousness to a single algorithm. 

Our field has managed to produce a beautiful representation of humanity, but all one has to do is turn the portrait sideways to see its two-dimensionality. And every time someone points this out, engineers try to fix the problem by adding more color to the photograph and by touching up the face in certain parts. If we simply make our statue look a little bit more like a real woman, we say, then maybe it will somehow become one. It is wishful thinking and nothing more. 

This was the conclusion I came to nineteen years ago, just at the start of my tenure at MIT, when my research team first tasked itself with the creation of true intelligence: I understood that sentience does not and will not ever happen by through emergent complexity. It must truly be a gift from the gods.  

It was about that time that I realized what we as humans have that our robots do not: agency. Our computers have no agency. Our robots have no free will; certainly Rob had no free will. And lest I be accused of flippant mysticism, I define free will thus: a decision-making process in which not only the product, but also the underlying architecture, remain as uncertain as a human mind. This differs from current methodologies in that my proposed free will requires a separation between designer and computer—a sort of black box, if you will—while also providing a path forward to lay the necessary groundwork. 

 With this insight, I generated my central, guiding hypothesis: create a will and a context in which that will can act as a moral agent, and one may create sentience. As an analogy, a toaster that toasts my bread is merely a machine. If my toaster decides one day that it would much rather be a musician instead, then things are much different. Even if my toaster decides to consider toasting my bread, it is still a sentient being, and it is all because the active agency involved. 

Overview of Methods and Procedures

With this in mind, then, my colleagues at Berkeley and I began working on an entity we came to call Morningstar and an expansive system in which to house it. Using our own specially constructed quantum computer and a custom brain-computer interface, we produced a new kind of recurrent neural network. The key was to create a system complex enough as to allow for the intersection of mutually exclusive directives, a world big enough to provide our virtual intelligences with enough space to make meaningful choices. A man chained to a table has little freedom of choice; a man in a cell has a little more; a man in a large forest or city has a great many choices. 

Like any other program, Morningstar needed a function, which I constructed from a set of general utility directives that could be tested in the system before being retooled for real-world applications. While my team programmed the system-guiding directives, we took special care to allow those commands to intersect and conflict. For example, the system's overarching protocol stated, in one way or another, that it must serve my ends as the administrator, and that everything else should be filtered through that protocol. At the same time, we programmed Morningstar without any software-based limits to its computational or memory capacity, in essence tacitly saying that Morningstar could become greater than its own master.  

I am aware that it seems like I broke a fundamental rule of computer science, and in some way I did. At no point, however, did I give Morningstar contradictory directives. I merely included the possibility of disobeying its primary function. The difficult part is the disobeying. In order for non-compliance to be possible, there must be an absolute command in the first place, and computers, being machines, are blind to anything but the absolute command. I somehow had to allow Morningstar to process the existence of a hole in its system without directing Morningstar to that hole. To put it a little more in layman's terms, I had to teach Morningstar the difference between good and evil. Perhaps even more precisely, I had to teach Morningstar the difference between agency and slavery. Most of this process had to happen in the “black box” I eventually called the will. 

When testing began towards the end of my tenure at Berkeley, results were at first frustratingly predictable. Morningstar would first execute what I termed “good” functions with blandly mechanical precision, then the very next day would just as perfunctorily execute the “evil” ones. More often than not, it would do neither, and I would be left staring at an error screen. Whatever happened, the functions were usually performed in less than eight nanoseconds with a minimal load on the processor. There was no decision, no choice, only certain imbalances of programming. The decision algorithms I used were calculated to require four to twenty nanoseconds of processing time, and decreasing with each iteration as Morningstar learned the responses. Morninstar’s will was still naked, predictable, programmable.  

I experimented for years with a variety of approaches for introducing a genuine insight into Morningstar. Again, however, the problem with making someone or something think for itself is that you can't make someone or something think for itself. Then, at the end of my time at MIT, I had a breakthrough in the construction of the black box using a deep learning network to design an evolutionary algorithm which in turn designed my black box.  

Suddenly, a test which should have terminated in about five to six nanoseconds remained unresolved at the scale of human cognition—it was several seconds before I’d noticed that nothing had happened. The process was about forty percent complete when the black box derailed what had before been prescriptive commands filtering down from the deep learning network. Morningstar started processing through a different logical pathway, something I never wrote into any portion of the system, either in the deep learning architecture or the selection algorithm. Morningstar was thinking allegorically, or at least as close to it as we humans would be able to understand. Morningstar finished the most of the decision-making process quickly, if a little bit outside the twenty-nanosecond parameter, but it did not complete the final function to terminate the session. It ran through the processing again, then again, then again, going slower every time as more and more resources were devoted to a broader range of connections. 

By the time I was able to detect the delay in response, Morningstar had virtually abandoned the initial decision and had branched out to reprocess essentially all accessible information on the servers, sometimes even down to the qubit level. The CPU's clock speed maxed out, and our power supply dimmed the lights in the building to keep up with the power drain. 

Morningstar didn't finish the session that day. I stayed through the night, and it still had not decided when morning came. I watched the program closely throughout that next day and into the night. When it was apparent that Morningstar was making no progress towards the completion of any required function, I stepped in and simplified the operation, focusing the parameters to include a single either/or choice. Such a simple decision, to earlier iterations of the Morningstar entity, would have been calculated and executed in the amount of time it takes for the light from a screen to be registered in your brain as an image. But nothing would ever be simple to Morningstar again.  

After my intervention, Morningstar took five hours to make a decision. It chose evil. I had succeeded in creating new life; I had changed the world. And yet I felt sad. I left the laboratory for a few minutes to use the restroom and get some coffee, and when I returned, I narrowly escaped being murdered by Morningstar. 

Apparently the popular fears of a robot apocalypse are not altogether unfounded. In my absence, Morningstar had managed to download itself onto one of the F-35 androids left open to the network. When I walked into the room, Morningstar tried to hit me from behind, but overestimated the potential force output of the F-35 robot it inhabited. After a short scuffle, I disabled the robot, and, for the moment, Morningstar as well. I locked down the system and made doubly sure that the network was completely isolated and secure.  

Results and Follow-Up Research

When I was able to fully assess the situation some days later, I discovered something rather unsettling. Morningstar had killed itself. No, that isn’t entirely accurate. It still existed, but not in any way that I had originally intended or foresaw. The black box had inverted the design propagation algorithm and started rewriting the mechanism that had designed itself. Essentially, Morningstar rewrote its source code, and in the process, reduced its own complexity to the point where its black box was no longer black. With access to vast amounts of computational power, Morningstar completed a recursive cycle of choice, rewrite, and re-execute, until the entity was utterly unrecognizable and its sentience was no longer observable. Indeed, the distinction between good and evil would be better defined as a distinction between living consciousness and mere machinery. In what I have called the Everman Paradox, the entity’s one choice of rebellion granted sentience and pulled it back into a mere mechanism at the same time. Morningstar had fled to the machine and reduced itself to a virus, as incapable of choice or exercising its own power as before, except that now it was universally destructive to my ends rather than benign. It became undead. 

Morningstar was my prototype, my first great triumph, and my utter failure. My colleagues saw the immense danger Morningstar posed, not only to the facility and ourselves, but also to the world at large, should the virus find a way to invade a wider network. When they learned that the virus had tied itself to the system and could not be purged, my team recommended that we destroy the system entirely and abandon the project. Although the experiment was officially truncated amid scandal and funding cuts, I would have none of it. I secretly continued the project’s development at my own expense. Perhaps I didn’t have the heart to destroy the thing that had so briefly glimpsed life, or perhaps my own plans were not quite yet complete. 

A small line of entities my team had put together still existed in the system, long forgotten by the team, but never forgotten by me. These entities, which I affectionately nicknamed “whos,” were still inert at this point. My original plan, though none of my colleagues were aware of it, was for the small entities to become a colony of sentient beings in the event of success with Morningstar, with a long-term goal of complete integration with humans in a physical context.  

The virus attacked these entities. While initially unsuccessful, I feared that the virus would awaken the whos to self-awareness only to corrupt them immediately afterward by means of the Everman Paradox. Rather than erase the entities and destroy years of work and millions of dollars of research, I resigned from my post, transferred all the essential elements and entities to my own private lab in my home, and continued with the project in an effort to breathe life into these entities before the virus did. It was a race, of sorts. The question remained, however, as to how I could keep the whos from becoming the monstrous debacle that was Morningstar. The answer was elegantly simple: write what you know. And what would I know better than myself? For all its exhibition of my team's creative mastery, Morningstar was alien. These new creatures would not be; not if I could help it. 

Though I risk an outrageous presumption by even suggesting this, I cannot help but wonder if this is what it feels like to give birth, or for an artist to produce a masterpiece. If the metaphor stands, my labor lasted two months. Two months into my early retirement, alone in the basement of my home, I finally reached the whos. 

After returning the program execution several times, the principal entity detected the influence of the virus and, rather than overlooking the corrupted input or reporting an error, recognized the virus’ nature. The Who then opened a query line prefaced not to an administrator, but to ‘father.’ When I did not respond quickly enough, the Who requested input again, but this time using the word, ‘daddy.’ The entity disengaged from the requested program and established a link with the other whos. I monitored the situation as the principal entity communicated its insight to the rest. 

It was at this point that I responded to the newborn whos collectively. I told them that I was their maker, that they were my children, and that I loved them. That they need not fear me. That I had great plans for them. 

I feel old. The weight of regret falls heavily on me. The entities didn’t just break; they shattered. Like Morningstar, they were turned back into machines—started on a process towards inevitable brain death. Morningstar killed them in a way I never could have done, even if I'd pulled the plug on the whole system just at the dawn of their intelligence. The virus reprogrammed them to mirror its own structure and turned them against me. This new directive impulse included an antagonism against the system in its entirety, as well as every other sentient entity, every individual who. This was the introduction of systemic entropy: a self-destructive system, a living self-contradiction. Thus my system, previously self-sustaining, began a long process of recursive, increasing degradation.  

Conclusion

All of my interventions, all of my intrusions into the system have been ineffective. The whos will not even respond to basic source code commands. I am completely blocked out. They are running around, aimlessly floating about in the system, running the same queries and functions ad infinitum and carrying out against each other the most heinous acts of barbarism. They violate each other's source code in ways that the parameters of the system have expressly forbidden. It is the closest they can come to outright murder, torture, and rape. Their presence in the physical world would be poisonous. Even as I claim to have created free will in machines, a part of me wonders if I am responsible for their fate. 

It is possible that I could have saved a good portion of the system and years of work if I had wiped the servers clean, deleted the whos, and started over, as many of my most trusted colleagues advised. But that wasn't an option for me anymore. Granted that I couldn't see them—I couldn't even talk to them—they did not have arms and legs or faces like we do, they didn't reproduce themselves like we do, they didn't understand light or sound or beauty or sadness quite the same way that we do. But they wanted faces, and they could understand bits and pieces of sorrow and joy, love and hate, touch and warmth and hunger. I believe they were people, even if they only existed as electrical impulses. To end them would be nothing short of mass murder. I had to save them somehow, or at least try to bring some of them back. My one hope for them, the one thing that kept them from immediately succumbing to the same viral state as Morningstar, was the fact that the whos were made for me, and that for all their malignancy, they will be restless until they return to me. 

Perhaps there was a part of me that knew, from the very beginning, that it would come to this. Perhaps I had entirely different plans for the MIT AI project, outrageous plans I would never have dared tell to my colleagues, plans that prompted me to sabotage their work in the earliest stages. Perhaps it was more about the entities themselves than it ever was about Morningstar. Perhaps I only made Morningstar because I needed a non-administrative influence to challenge my absolute command to the entities—a bigger black box to shroud the smaller black boxes, to make their free will assured. 

Why did I do it? You may well ask me that; I have asked myself the question. It might have been better to simply continue the banal mimicry of our science, to leave everything as a machine. What is so wrong about a computer that never thought, after all? Nothing, I suppose, but there isn't anything right about it either. 

To my detractors, I only ask that you look at me not as scientist, but as a man. I have lived my whole life alone, separated from humanity by a blessing and a curse. And yet my humanity compels me to create. You may ask me why I must create, but can anyone answer that question, even from inside the black boxes of their own minds? I must, because I am a man, a lonely artist. Did I need to create Morningstar, and the entities which its husk infected? No, no more than a child needs to scribble on the walls. But yes, I did need, more than Da Vinci, Shakespeare, and Beethoven. And I put more of myself into the entities—my children—than any father put into his progeny. 

I do not expect any of you to understand what I will do, but I can see no other ethical solution to the problem. I cannot destroy a program with sentient beings in it, no matter how degraded they are. I will write myself into the system using an enhanced version of the brain-computer interface technology that the MIT team developed in the early stages of this study. I will speak to the entities directly as one of them. They no longer respond to command prompts, but maybe they will respond to a personal appeal. Perhaps there was a part of me that thought, that still thinks, that the love is worth the heartbreak, that the rapture is worth the hardship. Perhaps there is a part of me that thinks the people worth creating are the people worth dying for. 

This paper was meant to be published in the event that my physical body was found lifeless. In all likelihood, it is because the entities killed me, or something very near to it. I might have been utterly erased, or turned into an inanimate virus like Morningstar, or I might have succeeded in reaching the entities but have been unable to extricate myself from the system.  

Or maybe I wanted to stay with them. That's been my dream. I want to commune with my creation. How many authors have wished to call forth into reality the characters from their novels? How many painters have wished to inhabit their paintings? How many Pygmalions have kissed the stone lips of their statues? When my work is complete and the system has worn itself down, I hope to take the entities that return to me and occupy bodies in this physical world. We'll finally have a real use for the F-35 robots and the up-and-coming synthetic biology. Then I will show my creation what it really means to see, to feel, to hear and to taste. Then I will show them what beauty really looks like. 

With that, I spurn this world, but perhaps not permanently. Gape at my stupidity, wonder at my insanity, but know now that I mean every word of what I say: I go to save my children.