Creative Machines: We Need Synthetic Friends19 min read
On a very general level (not necessarily bound to the field of climate change), we will develop and implement technologies that allow tapping into the collective wisdom of many.
These “machines,” as we call this synthesis of technologies in a somewhat simplistic manner, will support the generation of creative ideas where well-intended efforts to conquer existing frontiers of imagination are currently stymied by being confined to the mental limits of an organization, a particular research group, or a single individual.
Through the connection of a huge number of brains—biological and artificial—the machines can help evoke solutions from totally unexpected origins, and they may even produce novel ideas on their own. Collectively, we can come up with the most out of the ordinary but still qualified hypotheses—with well-substantiated fantasies, one might say.
An Organism-Like System of Brains
Emulating Our Own Cognitive Machinery
We believe that collecting, aggregating, distributing, and transforming information as required for enhanced insight generation through an organism-like system composed of biological and artificial brains will depend on our ability to emulate the functional power and organizational ingenuity ingrained into our very own cognitive machinery. Only if we can manage to scale the functional units that give rise to a single human being’s workflows of mental reasoning up to the level of a vast collection of people and computers interacting with one another, will we be able to address the convoluted challenges of climate change that we are facing.
If the constituents of this machinery are designed to provide for easy integration with human modes of perception and information processing, the researcher, engineer, developer, or designer using that technology will be liberated from the mundane aspects of her or his respective craft (like trying to understand one another by means of the error-prone and painstaking act of exchanging words). Only then can she focus on the creative aspects that require uniquely human capabilities such as intuition and non-inferential thinking. Without these fuzzy forms of creation, we will stay slaves to incremental progress, and the limits of reality will be shielded from our views for longer than we can afford. But if our own ingenuity can be multiplied by a machine that not only supports our own thinking by delivering us from the burden of crude information processing, but also by fetching those marginal concepts that inspire the inimitably human kind of analogical, abductive reasoning, we will find a solution for every problem.
“That Is Crazy—Dangerous Even!”
Well, yes, in a certain way (to the first exclamation) and no, not necessarily (to the second outcry).
The idea of connecting brains to build a networked brain of unprecedented thinking-power is not crazy at all. Think about it (with your singular brain for the time being): Our brains are massively interconnected devices already. An enormous number of neurons is connected via synapses. Also, while some nerve cells work together more closely than others and organize into distinct regions to let us taste things or recognize faces, all of these regions still run in a coordinated manner. Yet another connection would be that of the left and the right hemispheres. Albeit being two separate entities really, these functionally distinct brain-halves interact heavily via the Corpus Callosum that builds a bridge between them.
From a structural perspective, connections are the main feature of a brain! A brain is connected at so many different levels that one might righteously ask, why adding some more connections would be a crazy thing to do. And we are not talking about thousands of added connections (much less multiple hundred trillion, which would be the number of connections that an average adult human brain already maintains). When the most pervasive machine in the domain of online social networks carries 200 connections per participant on average, we can assume that the number of brain-to-brain connections potentially being established by creative machines will not be much higher.
For now, connecting one’s brain to the brain of someone else seems to bring about much more hardship than hitting the Add Friend button on Facebook does. Connecting brains is more complex than connecting social profiles to the degree that renders any attempt of explaining that gap a silly thing to do. So we leave it right there. Hence, what actually is crazy is trying to build these creative machines against all the odds.
“And what is about the danger of such machines taking matters in hand themselves?” That is a valid concern. But let us say two things about the fear of autonomous (and potentially evil) machines that have at their command some form of superior intelligence of the artificial kind. First, we do not intend to create anything that may function in an autonomous way, meaning that it is self-sufficient with regard to its operating energy requirements and that it makes decisions for itself. A creative machine as per our conception, albeit having artificial components engineered into it, must always comprise a vast number of biological brains too. We, the human population of our planet, must make for a core ingredient of that machine with our own brains for this hybrid of carbon- and silicon-based processor to become really creative. And as we will be part of the central nervous system of a creative machine, we will be the ones to dictate its actions. Second, and that is going to be no more than a statement, we advocate searching for ways to use artificial intelligence in a beneficial way instead of just focusing on its danger potentials.
Artificial intelligence will be a reality, and we think that it can be massively useful when applied in concert with our inherited intelligence of the biological kind. So we might be better off to think of these machines as synthetic friends than to demonize the underlying technologies, that are sometimes (simplistically) referred to as AI.
Thoughts on Imposing Creativity on a Machine
Emulating Our Own Cognitive Machinery
Much like the human brain, the machines’ primary task will be to build new connections where none have existed before. They would then be capable of spanning structural holes between domains of knowledge and expertise. Building on these newly grown connections, the machines must allow for conveying information of most different types to flow through that vast network seamlessly. Beyond facilitating information flow, we must find a way to make use of the complex of knowledge that is veiled by a vast heterogeneity of information. And then, the machines to be created must enable us to increase both the speed and the scope of reasoning dramatically. After all, creativity is also a question of pure cognitive strength.
As we currently understand it, three things will be required for that form of enhanced information processing in an extended network of brains to take place. These are our thoughts.
A Super-Structure under Constant Remodeling
First, we need a Network of Neural Networks. In order to reap the best possible solutions from the least expected of places, machines must somehow be connected to that place. Ideally, we have an interwoven super-structure that encompasses as many information repositories of biological or artificial kind as possible. As it must adapt to the evolving nature of both the problem (of climate change) and our general state of knowledge and expertise (on everything), the Network of Neural Networks must be designed for constant remodeling and growth just like our own brains adapt through the process of learning through neural plasticity.
Me, Snatching at Your Cognitions
Second, we need a Semantic Web of Cognition. Without the capability of exchanging information among its nodes, a network is worth nothing. Our brain would rather be a lump of protein and fat than a powerful computer were it not for the highly functional communication between the neurons (and the glia, as some will legitimately argue). What makes the task so complex in a research and deployment endeavor such as that pursued by Archai is the fact that the nodes of our intended Network of Neural Networks do often not understand one another. While nerve cells within a single brain communicate in a highly standardized manner either through a chemical reaction (via neurotransmitters) or through electrical impulses (using charged particles), brain-to-brain communicating across knowledge domains, cultures, or even the trenches of structural design in the case of human-machine interaction warrants a multitude of media and mechanisms.
The task gets even more complex when instead of binary non-/activation signaling as conducted by our brains such vague states of cognition as convictions, talents, decision patterns, or motives must be conveyed through the Network. Given the intricacy of climate change (and any other compound problem that has emerged from the unfathomable evolutionary steps that gave rise to the universe’s present state), we think that it is fair to say that the heterogeneity of information that must be assimilated somehow is massive.
In expectation of the highest possible level of compatibility among different forms of knowledge representation, and to guarantee interoperability of knowledge domains, we are seeking the equivalent of the Internet’s Semantic Web in the realm of cognition. Just like its source of inspiration, the Semantic Web of Cognitions holds at its core the aspiration of making all information amenable to machine processing. But it expands on that notion by accounting for the need to inject into the calculations those intermediate states of thinking and acting that are not typically legible to a non-human intelligence.
Through the Semantic Web of Cognitions, creative machines must learn to detect concept analogs across realms as diverse as mental states, memes, or discrete information. Doing so requires the machines to extract and represent relevant concepts, generalize these representations in a way as to make the underlying concepts accessible to a unified processing mechanism (like finding a language that both can understand), and then determine the degree of overlap or matching between these concepts. An overlap may then indicate a promising approach to solving a problem of mine by snatching at a cognition (or a sensation, idea, or intuition resulting from the process of cognition) that was born in your brain. Since I am a biologist and you are a musician, such intellectual cross-fertilization will allow me to tap into the very creative space of distant ideas.
Hence, while artificial brains might struggle to overcome the linear nature of purely inferential reasoning ingrained into their digital minds, they might be well suited to detect the unifying features of composing a fugue—a rather non-linear process that requires a high level of intuition—and trying gain new insights into, let us say, the folding of proteins.
By deploying those faculties that come with the Semantic Web of Cognitions, artificial brains can invigorate the act of generating well-substantiated fantasies as conducted by those that are capable of non-inferential, creative thinking: the biological brains.
Breaking down of Frontiers between (Un-)Related Territories
Third, we need that Network to be endowed with an Amplifier for Spreading Activation. A thinking human being (or another animal sufficiently capable of cognitions for that matter) does that thinking by starting from a concept and building from there to where the network of associations carries her or him. Along the way, new concepts are evoked and activated in the thinking mind, meaning that distinct neuronal patterns representing these concepts “catch fire.” Many of these newly aroused concepts were not originally linked to one of the concepts with which the whole deliberative process started. But if a suchlike evoked concept turns out to be useful for the deliberation at hand, it gets interwoven into the stream of current thought; otherwise, it will be discarded. So, if two concepts have not been linked originally, now they are—and that added link gives rise to a potentially new idea.
This act of creating associations builds on a process called spreading activation. The idea inherent to spreading activation is that of a cognitive process starting from an initial pattern of firing neurons (excited by an internal or external stimulus), with the activation so sparked traveling along a network of neurons and neuronal patterns that are evoked in the course of that particular cognition. From its origin, the firing-cascade proceeds in multiple directions simultaneously and only comes to a halt when the activation potentials that carry over from one neuron to another dry up somewhere down the line. Learning and insight then happen when cellular connections are built and solidified as the firing of certain neurons—and entire neuronal patterns—repeatedly spikes in temporal dependency.
The problem for a thinking human being is that we can only activate mental concepts that are there—meaning that one only knows what one knows—and that we can juggle just so many activated ideas at one time. These are pretty severe restrictions with regard to the intent of generating breakthrough ideas, and even the smartest among us are subject to these constraints. We, therefore, think that machines should be capable of amplifying that process of concept activation and associative creation by expanding the limited associative network of a single thinker to many thinking entities and by allowing us to integrate more thoughts into one powerful idea at once. And it is not that just more thoughts stuffed into an idea make for a better one.
The amplification of activation spreading across a variety of thinkes also holds the promise of colliding with a concept or a thought that was not meant to pop up, was it just for one-dimensional associations. But now, with the support of machines, the ability to more easily integrate information from the intellectual outskirts of a problem-solving attempt, allows for a new quality in the synthesis of raw material into new ideas. Potentially, we are presented with the opportunity to regain a sense for multi-disciplinary work that seems to have gone entirely lost in our age of hyper-specialization and risk-averseness. What we are asking from machines is to help unmask those positions, at which multiple dimensions of knowledge and experience interfere and give rise to novel insights. While Arthur Koestler claimed the intersection line of two “associative planes” to be the source of creativity (rather than speaking of associations he, consequently, coined the term bisociations which he used to explain such different phenomena as humor and the scientific inquiry), we call for the coalescence of many frames of reference when trying to make those discoveries, that help solve the most complex of problems. When arguing for building into our creative machines an Amplifier for Spreading Activation, we do this with the same goal in mind that Arthur Koestler promoted by stating that what is needed is “the breaking down of frontiers between related territories, the amalgamation of previously separate frames of reference or experimental techniques; the sudden falling into pattern of previously disjointed data.” To Koestler’s quote (p. 229), we would add that the territories can be, should be even, unrelated for the transformational idea to arise. It is along these lines of recombinant intelligence, that a super-organism of artificial and biological brains can dramatically increase its reasoning performance and its creative prowess.
“The Concept of Bisociation” is described by Arthur Koestler as “the perceiving of a situation or idea, L, in two self-consistent but habitually incompatible frames of reference, M1 and M2. […] L is not merely linked to one associative context, but bisociated with two.” [italics in original] (p. 35)
Machines as a Production System
Apart from being creative in the above-outlined sense, our machines may also allow for the productive integration of manpower and computing power by dividing labor intelligently among a varied collection of individual contributors. Much like every other (physical) machine, our creative machines must comprise a mode for efficient production. Enhancing productivity may be a rather simplistic task to ask from a machine compared to facilitating creativity (or even being intrinsically creative), but we still think that it is of tremendous importance as it will allow for our engine to operate at high speed with the least possible amount of resources. In its productivity mode, the machine controls the division of labor among all participants of a given problem-solving attempt, and it navigates the participants along the fastest route to convergence.
Creativity and Productivity—All at Once
Consequently, as per their creativity enhancing modality, Creative Machines will increase the probability of finding solutions from within an immensely large space of viable approaches. And as per their productivity features, these machines will empower each individual contributor to control for the resources he or she is willing to invest into any given project without compromising the venture altogether.
In simple terms, the machines we are developing can point to solutions that may be hidden behind the curtains of seemingly unrelated knowledge domains, and they allow for a productive aggregation of insight-labor provided by a large crowd of individuals.
More of Technology
No Results Found
The page you requested could not be found. Try refining your search, or use the navigation above to locate the post.