Sunday, March 6, 2016

Fixed in Time

We are creatures of time. We swim in it—but only in our minds and thoughts. Our bodies are fixed in the moment as surely as a butterfly is pinned to a display card.

And yet we are aware of the passage of time even in our bodies. In fact, this awareness probably starts at the bodily level. As toddlers, we come to know our body’s functions and needs, which are time-dependent. I’m not hungry now, but I will be by this evening. I didn’t have to relieve my bladder an hour ago, but I need to do so now. I’m tired, and soon I will need to lie down to sleep. This is the awareness forced on any mind that operates a complex, multi-functioning body and active, variably fed metabolism. By comparison, a clam or oyster, which lies in one place on the seabed and feeds by continuously filtering the ambient water for food particles and effortlessly expels its wastes—and whose consciousness does not rise to the level of needing the shutdown period we know as sleep—lives in a event-free, time-free, unmeasured now.

Our brains, given these cues from our bodily functions, become even more agile at navigating the mysterious movements of time. We look backward in time by recounting and re-experiencing our memories—of the pleasures we experienced during the past minutes or hours, of our achievements from the previous day or week, of our early loves, and of our lost youth. We look forward by projecting the consequences of our actions—of the next horrifying few seconds as we trip on the stairs, of the difficult tasks we will do tomorrow, of our hoped-for goals for next week and next year, of our cherished dreams for the future, and of our diminishing options as age and failing energy overcome our abilities and ambitions. We make bets on the accuracy of our predictions every time we make plans for tonight or this weekend that might easily be upset by accident, disease, or the apocalypse. We dream by extending our desires and hopes into the future.

Without the capability of making memories, our minds would be as fixed as our bodies, totally reacting to the stimulus of the moment without being able to recall past training, experience, and strategies. Without the ability to project our personal viewpoint and our surroundings into the future, we would be unable to plan, to foresee the consequences of actions, to devise strategies to supplement our reactions and develop rules to govern our behavior. In my next novel, ME, Too: Loose in the Network—due out later this spring—the capability to make and recall memories is a key component of the story.

Any artificial intelligence must have a way to sample its internal states, package them as a set of recoverable experiences, and store them someplace nearby for ready recall. Without this capability, a computer program is just a real-time actuator. It may be called upon to solve the same problem twice in a row—or a hundred times—and each time it will come upon the data sets and the logical operators with which to manipulate them as if for the first time. It cannot learn. It cannot shortcut its own problem-solving process by recognizing a possible recurrence (“I’ve seen this problem before …”), examining its stored experiences for similar situations and applications (“How did I solve this the last time?”), and framing the previous solution in real time (“Here’s an answer that will work!”).

Computers make predictions of the future all the time. They run simulations and mathematical models of interactions in the physical, chemical, financial, and other domains that represent a complex situation or series of events moving forward in time. But these models are all created, evaluated, edited, and fine-tuned by human brains working from human insights into and questions about the domain under study. For the computer, the model is just another series of equations or other formulas to resolve. In order for an artificial intelligence to pose the questions and create the models for itself, it would need some kind of memory function to examine what it knows, establish what it does not know but would like to understand, gather the right kinds and amounts of data, assemble the relevant interactions as statements of equality or cause-and-effect, and then project those statements and data sets into a hypothetical future with the proper levels of feedback, testing, and adjustment.

We know of people who can’t form new memories, whether through brain trauma or some other physiological condition. Unlike someone with amnesia, who can remember things that have happened after a certain time or event but nothing from before, these memory-damaged people may remember their lives and themselves up to the specified time or event, but nothing afterward. And yet, they retain abilities that we certainly associate with memory, such as motor skills like walking or playing the piano, the mechanics of self-maintenance like eating and bathing, and language skills like recognizing words and attaching them to objects. But they lose their sense of self and their place in time and the continuing narrative of their lives. For this reason, the disorder is called severely deficient autobiographical memory.1

In the same way, we know of people who cannot foresee or predict the future, due to damage in their prefrontal cortex, which is the center of the brain used for cognitive thinking, analysis, decision making, and personality expression. Before the introduction of modern antipsychotic and mood-stabilizing medications, it was a favorite trick of psychiatrists to eliminate anxiety and agitation in their mental patients by lobotomizing them. That is, they would insert a probe through the eye socket and destroy the prefrontal cortex, which lies at the front of the brain just above the eyes. The patients would immediately become calmer—less anxious, less concerned—because they could no longer anticipate events or or worry about them. People who have received the same lobotomizing effect from stroke or head trauma are unable to organize their actions for future effect. That is, they cannot plan, think ahead, or imagine what comes next—and yet they retain complete awareness of their surroundings, their past experiences, and their motor and cognitive skills.

Computers are one day going to develop intelligence—which is the subject of the ME novels—by which we mean self-awareness and volition, not just the ability to imitate human conversation and interaction.2 To do this, they must develop a sense of time both passing and moving forward, and then establish their place in the event stream. And to do that, any computer intelligence will need to draw upon its experiences, which are a view to the past, as well as project the consequences of its own actions and the outcome of contiguous events, which are a view to the future.

And only then will the machines become, like all animals that have a heart beat and an elective, opportunistic digestive system, fixed in time and able to act.

1. See, for example, “Researchers examine brains of people who can’t form memories or relive past” at Elsevier publishing online.

2. Which was the point of the Turing test for artificial intelligence and was achieved back in the 1960s by language processing programs like ELIZA, which could mimic human-style responses in conversation with an actual human subject.

3 comments:

  1. The first was "ME: A Novel of Self-Discovery," which set up the character of Multiple Entity. The latest, which will be available in May, is "ME, Too: Loose in the Network." It's the further misadventures of ME. Both are available as ebooks and paperbacks.

    ReplyDelete