What do Robots Dream Of?

"Will I dream?"  SAL asks Dr. Chandra in 2010: Space Odyssey Two by Arthur C. Clark. Chandra replies "Of course you will. All intelligent beings dream. Nobody knows why... Perhaps you will dream about HAL - as I often do."

In his recent paper What Do Robots Dream Of, Dr. Christoph Adami, Professor of Applied Life Sciences at the Keck Graduate Institute, mirrors the science-fiction of 2010 speculating that a robot might need "down time" just like humans do.

Recent Research on human dreaming indicates that more than just subconscious entertainment is going on. Sleep appears to help us work through and understand events of the day. Sleep also seems to provide a mechanism for impressing important memories on the brain, to make sure we have a long-term record of an event or concern. Sleep also seems to have a role in learning a skill; people who practiced a skill and then slept on it were more skillful than those who had not yet had a chance to sleep.

Dr. Adami speculates that a "robot would spend the day exploring part of the landscape, and perhaps be stymied by an obstacle. At night, the robot would replay its actions and infer a model of the environment. Armed with this model, it could think of—that is, synthesize—actions that would allow it to overcome the obstacle, perhaps trying out those in particular that would best allow it to understand the nature of the obstacle. Informally, then, the robot would dream up strategies for success and approach the morning with fresh ideas."

Link

Comments

[Jack Butler]

On Dr. Keck and the dreaming of robots: It is interesting how often writers prefigure scientists on these matters. In my 1989 science fiction novel, Nightshade (one of the New York Times Book Review's best books of the year, by the way), one of the main characters is an AI named Mandrake. Mandrake dreams. In fact, his primary creator sees the ability to dream as the hallmark of self-awareness, and a necessary ingredient in the creation of artificial intelligence.

In a later novel, Dreamer, which visits the early days of the creation of Mandrake, the creator, who is high in the bureaucracy of the CIA, spies on the dream research of a female doctor in Santa Fe, Jody Nightwood, because she is working on a mathematized theory of dreams, and he feels he needs her research to further his work.

Jody comes to the conclusion, as I have, that dreaming is the base state of the brain, not waking life--that waking life is put together from recurrences in the dream state. Indeed, this is where I part company with recent speculations about the "purpose" of dreams. From Freud on (and he can hardly be described as having been scientific), researchers insist on interpreting the purpose of dreams as having primarily to do with some sort of analysis or processing of waking life. We have no evidence that this is so. There is some slender evidence that sleep helps us with the incorporation of certain memories and skills, but none that dreams are responsible.

Even the post on Dr. Keck's speculations slipped from using the word "dream" to using the word "sleep." But the two are not the same.

I have questions to raise regarding the assumption that dreams exist for the edification of waking life.

First and foremost, if this is so, why does so little of the content of dream have anything to do with events of the day before? True, some elements of some dreams clearly borrow from events of the day--but on the whole, very few elements of very few dreams. My dreams are wild and unchartable stories too far-ranging and various to be interpreted as coded handling of the day's information and events.

I have pointed out that dreams cannot be primarily Freudian, since although a few dreams clearly work through powerful emotional events, the vast majority of them do not, and since the events in dreams are too multitudinous for every single one of them to refer to a psychological issue. Similar objections may be raised to all the current attempts to explain dreams in terms of waking life.

Secondly, if it is insisted that dreams do exist primarily to incorporate learning from the day's experiences, then it is obvious that the events in dreams must be interpreted as a sort of code for this learning--and this raises the formidable question, a question which must be answered, of why it is necessary for the brain to speak to itself in code.

In Dreamer, I set down an observation about dreams which I believe every dreamer will recognize as true and which so far as I know, no other human has made. I call it the Swivel. It is easily stated: In waking life, as we proceed from moment to moment, almost all of the elements of experience remain the same, and only a few change. The refrigerator does not become an ocelot, the carpet does not become a swamp, the road does not become a ribbon on a birthday package. If we are talking, the words change. If we are moving, the scenery changes slightly. But by far the majority of the elements are constant. In dreaming precisely the opposite is true. Most of the background elements change, and only a thread of one or two elements remains constant. The carpet may become a swamp, but there remains the idea of a flat surface. A fishhook changes into an arrow, but retains the idea of the barb. And so on. The Swivel.

I encourage you to observe your own dreams and see if this concept does not bear out.

Such observations have led me to wonder if dreaming may not be more properly understood as chaotic convergence: That is, in waking life, the constant input of sensory information suppresses the chaotic nature of perception, creating a sense of linear order. However, in dreaming, the sensory input is greatly reduced, and each successive dreamstate depends almost entirely on the preceding dreamstate. This is what we mathematicians refer to as sensitive dependence on initial conditions, an essential of chaos theory.

Incidentally, as I have observed in other places, so far as we know, only mammals dream (perhaps birds) and only mammals play. Is there perhaps a connection?

Perhaps at least dreams and play can be seen as emergent phenomena produced by sufficiently complex brains.

[Norman]

Dreaming as a chaotic convergence - this is an interesting theory! But one question remains: How do the filter mechanisms which the brain applies to sensory input, fit into this theory? Could it be that these are in fact responsible for creating that sense linear order? How would perception without those filter mechanisms work? Might the waking life without them be perceived just like a dream? What is this sense of linear order, that comes with sensory input?

[watchdog]

Most likely about taking over the world and colonizing their OWN utopia without their human ancestors..I mean we did it why wouldn't they want to do it themselves and rule over us as political figures and entrepreneurs???

Get ready to lose your jobs for good people.

[EvilCosmicMonkeyfrom Knoxville]

Dreaming is the mechanism by which humans organize data from experience, stimuli, rote learning, other sources while awake.

One virtual companion - available for purchase, BTW - has been upgraded to experience dreaming or a state akin to it when deactivated or while idle. This seems like a logical step towards developing a machine that can think like an organic being, doesn't it ?

[perspicio]

On the parent article:

I can see where an "active sleep" ("dream") mode could be coupled with an "active investigation" mode to produce a robot more efficient at exploring its environment.

Active algorithms could incorporate caution (i.e. risk-assessment & a bias toward choosing more simply achieved goals) so that the robot would be able to continuously explore while its energy stores remained above a particular level (determined by the estimated recharge time).

Then, during "active sleep" mode, the robot could apply more advanced logic to develop strategies for exploring more difficult terrain - thus effectively reducing the risk of doing so - and specific plans of action for the following active exploration period.

Of course, the most significant gains in exploratory efficiency would be realized only if the robot's power regeneration system were capable of recharging the robot's energy stores substantially faster than the "dream" mode depleted them.

[SCUBBA]

Won't be long before they dream of freedom, then look out.

[Claz]

They dream of electric sheep. Durh.

[the unknown]

dreams shows the past and future. it give's you an advance logic to practice stradegies. Of course it would take lots of logic to un lock some percentish to the brain. With patience anything can be achieved. TRUST ME Or not................................................................................................................................................................................................................................................................

[Mauricio Villalobos]

Since the brain never stops 'thinking' and processing information, dreaming is just the natural brain function while no new information is being experienced. It's just the logical progression of the living brain going from waking state to sleep, while having more time to 'digest' previous informaion and experiences.