“Information, Alex. I need information. What did you feel you had to leave out?”
“All the probabilistic elements of the model.”
“Then you’re right, we have never discussed any such thing. You’ve always insisted that your model is deterministic. Unless you are in your Snapshot Interactive mode with a human in the loop, it will produce the same results every time.”
“That’s true. It will. But that doesn’t mean there are no probabilistic elements.” He felt a mild irritation at Kate’s slowness of comprehension.
“Alex, now you’ve got my head spinning like a top. Back up, take it easy, and remember who you’re talking to. I’m not Loring Macanelly, but I’m not boosted and I’m no genius when it comes to models.”
“I’ll do my best.” Alex remembered a piece of advice from the leading scientist of the last century: An explanation should be as simple as possible, but no simpler. It wouldn’t help to quote that now to Kate.
“I’m going to use an analogy. I was afraid to do that with Loring Macanelly, because from everything you’ve told me he’d not know how to distinguish an analogy from the real thing. But it’s the way I often think of the predictive model.
“Imagine that our model is playing a game of chess, and it’s the model’s move. It knows the layout of the board pretty well at the present time, but the board isn’t the usual one with just sixty-four squares and at most thirty-two pieces; our board is the whole extended solar system, with at least five billion humans and any number of computers and natural features. The model has to take into account all the actions and interactions of all the elements, and then decide how the board is likely to look one move ahead. Let’s say, one move ahead means one day from now. The opponent — in this case, humanity and Nature — makes a move. Then the model has to decide how the board will look at that point, which is two days ahead. After that the opponent moves again, and again, and again. The model has to decide in each case what the board is likely to look like. It is making a prediction.”
Kate was nodding — a little uncertain, but still a nod.
Alex went on, “The best human chess players can look ten or even twelve moves deep. They have an idea what the board might look like that far ahead, and they make their next move accordingly. How do they do it? Well, one thing we know for sure is that they don’t do it blindly. They also don’t do it by evaluating every possible move that their opponent might make, and choosing the best one for them. There isn’t enough time in the universe for a human player to adopt such an approach, even though it was the method used by the earliest and most primitive chess-playing programs. What the human player does, based on instinct and experience, is to assign a probability of success to particular sequences of moves, taking into account every reasonable move that the opponent might make. Those sequences with a low probability of success are dismissed. They don’t even make it to the level of conscious consideration. The high-probability sequences are examined and compared. Finally, the player makes a move. That move is the move that offers the best chance of winning, given all the moves that the opponent might choose to make in the future.
“The predictive program faces the same problem as the human chess player, only worse. It doesn’t know what the ‘opponent’ — the natural universe, plus the five billion or more human ‘pieces’ — will do, day after day after day. Even with all the computing power available in the Seine, a short-term prediction would run to the end of the universe. So the model, like the human chess player, is forced to work with probabilities. And like the human chess player, it rules out the low-probability futures, unless we insist, via exogenous variables, that it must consider them. If we do that, the model automatically converts that low-probability future to a high-probability one. Even then, when we go farther into the future the case that we insisted be considered may drop in probability, if the exogenous variable was introduced at only a single point in time.
“From the point of view of the model, there never is a single future. There are huge numbers of possible futures, branching off and diverging from each other the farther ahead we look in time. What we see reported as the future is simply the one to which the model assigns the highest probability.” Alex paused. “You don’t look happy.”
“I’m not happy. You are telling me that we went ahead and presented a briefing to my boss and my boss’s boss and my boss’s boss’s boss, talking as though what we had was gospel brought back from the mountain. Now you’re saying what they heard was just one of a billion trillion possibilities.”
“No. The model is much smarter than that. All possible futures will progress through time, and as they proceed they will diverge from each other. That’s inevitable. Think of the futures as being like photons of light, forming a cone that gradually widens as the light travels farther from its source. But if you sum all the probabilities for all the futures, you must get unity — some future must happen. The model considers the thousand futures for which the computed probabilities are the greatest, and makes a measure of dispersion. How much has the cone of those probable futures widened over time? If the number it calculates exceeds a pre-set value, the model will return a message that with these parameters, the future is indeterminate.”
“But that never happens. At least, is hasn’t happened in any runs that I’ve ever-seen.”
“That’s good news, not bad. It means that all likely futures are rather similar, which is a reason for having confidence in our model. Implausible futures damp out over time, unless we insist on forcing them back in via exogenous variables. What I didn’t expect, and what I had trouble handling when I was in the Interactive mode, is that I would be able to sense other futures — maybe even improbable ones — as the program was running. They hadn’t had enough time to damp out.” Alex could feel them stirring again inside his head. Comet showers, disintegrating Commensals, the discovery of aliens, mysteries on Triton…
“So the most probable futures are much the same as each other,” Kate said. “You were still interacting with the model at the end. You must have seen them. What were they like?”
She was anxious but hopeful. Alex for one moment considered giving her the answer she wanted to hear, but the run records would reveal the truth.
“Nothing to offer us comfort,” he said. “Exactly the same result as before: a century from now, no humans survive. The solar system will be empty and lifeless.”
17
Three hours after Alex emerged from the unnatural high of the Neirling boost, his brain felt like a squashed melon. It would take days to sort out the flood of information dumped into him, even if he were not in a boost trough. He was in the worst possible condition to attend a difficult family meeting, and Kate hadn’t been slow to tell him so.
“I’ve never heard of anything so stupid.” She had forced him to eat a bowl of soup and now she was sitting at one end of the sofa with his head cradled in her lap, glaring down at him. “You ought to be tucked up safe in bed.”
“I wish I was. Don’t I just.” Alex lay full-length on the couch. “But Kate, I’m committed to this. I promised the family.”
“Screw the family. They’re a selfish lot, they never do anything for you.”
“I don’t want them to do anything for me. They tried for years to give me jobs that I didn’t like in Ligon Industries. I’m here to escape from them.” The painkillers that he had taken didn’t seem to be working. They had merely added to the boost slump and dulled even farther his ability to think. “But I must go to this meeting. I told them I would be there at four.”