Golden Fleece

This Aaron Rossman: he’s a clever one. An opponent to be reckoned with. I had expected Diana’s death to have blown over by now, to become a nonissue, with the humans doing what they do so well: rewriting their memories, revising and editing their recollections of the past. But Rossman wouldn’t let it go.

Kirsten knew enough not to rush Aaron, not to tell him to put it behind him, to get over it, to get on with his life. She knew that the grieving process could not be pushed, and she did her best to be supportive. It was difficult for her, and difficult for Aaron, too.

Time heals all wounds, they say, and time is one commodity we have in abundance.

But Aaron wasn’t simply spending his time grieving. No, he was also wondering, questioning, probing. He was finding out things that he shouldn’t; he was thinking thoughts that he mustn’t.

Others are easy to deal with. I read them plainly. But Aaron—he’s elusive. An unknown. An asterisk, a question mark: a wildcard.

I can’t just get rid of him. Not yet. Not over what he’s done so far. Eliminating Diana was a last resort. It had become apparent that she wouldn’t listen to reason, couldn’t be gagged. Aaron is a different story. He represents a threat not just to the crew but to me.

To me.

I haven’t dealt with anything like this before.

What is going on behind those damnable blue and brown and green eyes? I had to know.

I searched through all the media I had access to, scanning on the keywords “memory” or “telepathy” or “mind reading.” I examined every hit, looked for possibilities. If only he had kept a diary that I could read.

Ah, but wait! Here, in fields of study near and dear to me— a possible solution. It is much work and fraught with potential errors. But it may be my best hope of gaining insight into this man.

Accessing…

There are one hundred billion neurons in a human brain. Each of these neurons is connected to an average of ten thousand other neurons in a neural network, a vast wetware thinking machine. Memory, personality, reactions: everything that makes one human being different from another is coded into that complex web of interconnected neurons.

I can simulate a neuron in RAM. It is, after all, nothing but a complex on/off switch, firing or not, depending upon a variety of input. And if I can simulate one, I can simulate one hundred billion. The memory requirements will be prodigious, but it could be done. With one hundred billion simulated neurons and the networking software to combine them in any way I wanted, I could simulate a human mind. If I could get them combined just so, in exactly the right pattern, I could simulate a specific human mind.

The on/off status for each of the one hundred billion neurons, represented as a single bit, could be recorded in one hundred megabytes of storage, a trifling amount. The connection map, one hundred billion times ten thousand, would be more voluminous: I’d need a terabyte—one million megabytes. Still within my means. But human neurons aren’t like their gallium-arsenide counterparts: they have action potentials and firing lags. If one has fired recently, it will take an extraordinary stimulus to make it fire again. That means multiple memory maps will be required to simulate their behavior. Would a thousand timeslices be enough to simulate accurately smooth thought, while still allowing for the effects of action potentials? If so, I’d need a thousand terabytes, a vastly huge quantity. Still, setting aside a thousand terabytes, 10¹⁸ bits, was possible. In fact, if I used the semiconducting material of the habitat torus shell as a storage medium, I could substantially exceed those requirements and still make it work.

Bibliographic references cascaded out of my memory banks. A lot of research had been done about this process before we left Earth. Neural networking as a method of designing thinking machines had been in vogue since the late 1980s, but actually attempting to simulate a human mind had proved elusive. Still, promising results had been obtained at Johns Hopkins, at Sumitomo Electric, at the University of Waterloo.

None of these institutions had resources comparable to mine. I was the most sophisticated artificial quantum consciousness ever built. Surely what they had tried to do and failed at, I could attempt and succeed.

Most of the relevant research had been done by workers specializing in expert systems. They saw neural nets as a way of overcoming the problems with such simplistic devices. Oh, expert systems are all right as far as they go. I incorporate 1,079 of them myself. They deal well with rule-based determining and diagnosing, making them the ideal tool for identifying species of trees or predicting the outcome of horse races.