(Animated Artificial Life)
3.6 Discussion
3.6.1. Nervous Systems
It is possible that brains are local inflammations of early evolutionary nervous systems, resulting from the benefit in early animals to have state and generate internal models in order to negotiate a complex environment. This is an artificial life-friendly interpretation of brains.
The simulations I have described do not include extensive adaptive behavior in individual organisms. An enhancement in this design methodology would be to wire up the creatures with more connectivity and reactivity to the environment, and then to eventually plug-in evolvable neural nets (and the capability for them to grow) which complexify as a property of the creature's coupling with a complex environment. The Vehicles of [Braitenberg, 1984] supply some inspiration for this bottom-up design approach.
This design philosophy does not consist of building a brain structure prior to establishing the nature of the organism's connectivity to the physical environment. If adaptive behavior and intelligence are emergent properties of biological life, then it is more sensible to design proto-bodies, and to wire up a proto-nervous system in which brain-like structures might emerge in direct response to the environment. The "Physical Grounding Hypothesis" [Brooks, 1990], states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world.
In the current system, there are hardly any brains to speak of. It includes a simple mental model which is (at this point) predetermined and purely reactive: sensors in a creature detect aspects of its relation to the environment, which can directly transition its mental state to another mental state. For instance, while in "looking for mate state," a creature will react to certain visual stimuli, which can cause it to enter into a new state, such as "pursuing mate." A future enhancement to this scheme includes neural material which can evolve to become a mediator of the sensor-actuator pathways. The neural material would evolve internal structure and representation in the context of the creature's direct coupling with the dynamic environment. So, for instance, instead of an environmental stimulus directly and unconditionally causing a predetermined change of mental state or body motion, the stimulus is processed along the way, resulting in many possible outcomes, depending upon some internal state and some environmental situations.
3.6.2. Muscles, Bones, and Nerves
Here is an idea for a possible artificial life project: imagine a simulation which models a simple environment, and uses three basic creature materials: muscles, bones, and nerves. Imagine concocting an embryology that uses a genotype to build structures (proto-creatures) out of these materials. Primordial, un-evolved proto-creatures would consist of chaotic assemblages in which longish bone elements are connected hinge-wise end-to-end. Muscles (bone-bending agents) would exist at bone connections. Nerves would consist of linear elements with neurons at their ends, wired up to bone ends as well as each other. These proto-creatures would be stimulated by internal motors and a few environmental triggers. Some of the nerve endings could be sensors that could respond to stimuli corresponding to the presence or positions of objects. This proto-creature idea is not unlike the method used by Sims [1994a] for the design of his virtual creatures' control systems.
Imagine further that many of the nerve endings are connective neurons(the connectionist stuff of neural nets. While some nerve endings could receive stimuli, and while other nerve endings could stimulate muscles to bend bones, many nerve endings would essentially become internal layers, and the embryology would be capable of constructing potentially many such neurons. This simulation would be capable of evolving neural-like structures which are three-dimensional and completely wedded to the "body" (muscles and bones), evolving physically along with the body, through selection. If one were to successfully build such a simulation, capable of evolution, would brain-like or central-pattern-generator-like nerve bundles naturally emerge in local areas of the body? Would nerve-bundles appear in crucial local areas of the body, where a little subsumtive computation might help? (Figure 17). If such brain-like structures were to emerge, this might serve as a nice illustration that the brain and body are not separated by a "Cartesian Lesion," but are co-evolved, and unified as a single Bodymind.
Fig. 17. Emergent Brains.
3.6.3. Humans
We humans have always been extremely interested in Homo sapiens, and have rendered the likeness of this species in a myriad of ways throughout history. A recent motivation to render humans is accompanied by the arrival of a new communications medium: real-time 3D cyberspace for chat. Enter the avatar, a proxy for your identity; a computergraphic object representing you and your intentions. Since we are humans, we usually want our avatars to be humanoid, but of course they don't have to be. Humans are not only potent subject matter for art and communications media, they are the most familiar beings in our lives, and so our perceptions of things human are finely tuned. Humans are intensely responsive to faces and can read the most subtle of expressions; they are sensitive to skin tone, and can appreciate the well-crafted hues in a fine portrait; they are able to recognize familiar voices, and can easily detect mechanically-generated imitations; and finally, humans are superbly good at reading body language. Have you ever recognized a friend from a long distance away, simply from his or her distinctive gait? Body movement is one of the more unappreciated channels of expression and communication (especially now since person-to-person communication is so mediated by technology, and at such low bandwidth compared to the real thing).
While many of the human characters rendered in computer animations and computer games do impressively model key aspects of humans, they are almost always recognizable as synthetic. The successful Turing Test is a long way off. This might be because so much human modeling is simply a further extension of classic computergraphic modeling techniques, whereby the surfaces of volumes are broken down into polygons, and shaded using a lighting model. Motion is achieved through techniques such as scripted keyframe animation or snippets of motion-capture. However, one notices how artificial these models appear when set to motion, as compared to real humans. Motion-capture can achieve relatively good results, but does a motion-captured script really represent any unique situation that a synthetic human is in at a given moment?
One reason that state-of-the-art virtual humans are unconvincing is that they are built "skin-first." If you slice one of these computergraphic models open, you will notice that it is hollow, whereas if you slice open a real human (and I am not suggesting you do this) you will notice a complex expression of billions of years of evolution in the biosphere, with a motor control system whose design history is punctuated by strategies for fleeing from predators, navigating rough terrains, leaping between tree branches, and performing mating dances.
It appears as if the skills enabling one to play chess and use grammar are a very recent evolutionary trend, whereas climbing a tree is a much older skill, perhaps more complex, and in fact possibly an activity whose basic cognitive mechanisms are the primordial stuff of higher reasoning. The simulations described in this paper do not aim to model humans. But if that were the goal, it may be best to design a vestibular system and an environment in which vertical, bipedal locomotion might be advantageous. Best to design an evolvable neural system (and to allocate lots of computer memory!) and a highly complex, dynamic environment, whereby something brain-like might billow as an internal echo of the external complexity. Best to design a simulation in which objects existing in the world might be exploited as tools and symbols, extended phenotypes, cultural memes. What might evolve then could be something more human-like than the wiggling forms shown in the illustrations of this paper. But in the meanwhile, there is no rush on my part to get to human levels with these simulations. And at any rate, imaginary animals can be very thought-provoking.
next section:
Conclusion
Conclusion
