Ten years ago, when I was working towards my master’s degree in Geographic Information Systems at Penn State University, I never thought my education in geospatial networking and statistical modeling would help me to appreciate how God designed biological neural networks. At that time, knowing how geospatial network data is constructed for GPS units was considered core curriculum, but today the industry is moving beyond this toward machine learning in order to predict traffic patterns using artificial neural networks.
Many atheists may look at my perspective on design in neural networks as an outgrowth of my faith and not something that can be proven in science. Is this a sound argument? The Creation model proposed by Reasons to Believe argues the following:
“Systems and structures produced by intelligent agents typically possess characteristics that distinguish them from those produced by natural processes. These properties serve as indicators of design. They will be apparent in biochemical systems of the cell if the biblical Creator is responsible for life.”Fuz Rana and Hugh Ross, Origins of Life: Biblical and Evolutionary Models Face Off, pg. 44
How do we identify the fingerprints of a designer? Fuz Rana, in his book The Cell’s Design, argues that in order to identify the fingerprints of a designer we need a set of criteria that universally describes the behavior of an intelligent designer and some understanding of the Creator’s properties and capabilities. (The Cell’s Design pg.29)
Although no universal criteria for the behavior of an intelligent agent exists the Bible does say that we are made in God’s image. This suggests that this likeness reveals itself in human creativity (art, music, literature, and technical inventiveness). In addition, being a reflection of our Creator suggests that our inventions will mirror those that were divinely designed. (The Cell’s Design pg.29)
Is it possible to detect the Creator’s activity if he works in a way that mimics natural processes or simulates chance? If we look at neural networks from a pure naturalistic perspective can we see a pattern of organization that can be easily explained by the evolutionary paradigm or would we run into an “evolution of the gaps” theory?
This article will briefly examine the relationship between human built neural networks and biological neural networks and how it points to intelligent design. We will start off by examining the teleological argument, which argues that neural networks are not the inevitable product of chance or nature’s fixed laws and early conditions. We will then discuss how GPS units are made and how the machine learning that goes into traffic prediction reflects the machine learning of our brain. We will then discuss, via the converse Watchmaker argument, whether nature’s laws or God’s laws are responsible for their production.
The Teleological Argument
According to the philosopher Ken Samples, the teleological argument (from the Greek word telos—“end,” “purpose,” “completeness”), which claims to reveal God’s existence from the design evident in the universe, makes the following statement:
Premise 1: The fine-tuning of the universe must result from physical necessity, chance, or design.
Premise 2: It does not result from physical necessity or chance.
Conclusion: Therefore, it results from design.
(7 Truths That Changed the World, pg. 113 )
Many distinguished scientists reject both chance and necessity. Regarding necessity, eminent physicist and cosmologist Paul Davies asserts in his book The Mind of God that the universe is not the inevitable product of nature’s fixed laws and early conditions. Instead he writes, “It seems, then, that the physical universe does not have to be the way it is: it could have been otherwise.” (The Mind of God, pg 113 – as quoted by Samples)
Concerning chance being the reason for the universe’s fine-tuning, physicist Roger Penrose has calculated the likelihood of the undirected formation of our universe to be one part in 1010(123). This number shows how incredibly improbable a chance formation is and that it is not a rationally plausible consideration. (The Emperor’s New Mind, pg 169 – as noted by Samples)
Given that both premises are not rationally plausible explanations, it seems that design is the best, if not the only, reasonable explanation. Philosopher Paul Copan has argued that “Design seems the preferable option given its greater explanatory power. If God exists, a delicately balanced universe isn’t surprising at all. If God doesn’t exist, shock is appropriate.” (Loving Wisdom, pg 85 – as quoted by Samples)
Samples has made the following observation regarding the rationally untenable naturalistic position:
“If one embraces the evolutionary view that the sensory organs and cognitive faculties of human beings are the result of strictly natural processes, then how can one trust that the things one observes correspond with reality?”Dr. Ken Samples, 7 Truths That Changed the World, pg 114.
How can information, knowledge, and truth come from a random, accidental source? Naturalism, essentially, purports that life and reason came from a source that lacked all of these characteristics. Therefore, we must return to our question of whether it is a sound argument by atheists to assume that observing design in neural networks is truly based solely on my faith alone and not on science. (7 Truths That Changed the World, pg 114.)
What is a geospatial network?
A network is a system of interconnected elements, such as edges (lines) and connecting junctions (points), that represent possible routes from one location to another. People, resources, and goods tend to travel along networks: cars and trucks travel on roads, airliners fly on predetermined flight paths, and oil flows in pipelines. By modeling potential travel paths with a network, it is possible to analyze the movement of all the agents on the network.
According to the Environmental Systems Research Institute (ESRI), a geospatial network is constructed to assist operators in answering questions like the following:
- What is the quickest way to get from Manhattan to Brooklyn?
- If a fire incident is reported in downtown San Francisco, what are the closest fire stations that can respond to the incident within five minutes’ drive time?
- What are the market areas covered by the warehouses in various cities?
- What is the nearest coffee shop from my current location?
- How can we route our fleet of delivery vehicles to minimize overall transportation costs and improve customer service?
- Where should we open a new branch of our business to maximize market share?
- Our company needs to downsize—which stores should we close to maintain the most overall demand?
- What are live or historical traffic conditions like, and how do they affect my network analysis results?
A network is a system of interconnected elements, such as edges (lines) and connecting junctions (points), that represent possible routes from one location to another.
People, resources, and goods tend to travel along networks: cars and trucks travel on roads, airliners fly on predetermined flight paths, and oil flows in pipelines. By modeling potential travel paths with a network, it is possible to perform analyses related to the movement of the oil, trucks, or other agents on the network. The most common network analysis is finding the shortest path between two points.
In order to model transportation networks, Network Datasets are created from source features, which can include simple features (lines and points) and turns, and store the connectivity of the source features. When you perform a network analysis, it is always done on a network dataset.
Also, Network Datasets contain Network elements. Network elements are generated from the source features used to create the network datasets. The geometry of the source features helps establish connectivity. In addition, network elements have attributes that control navigation over the network.
The following are the three types of network elements:
- Edges—Edges connect to other elements (junctions) and are the links over which agents travel. Line feature classes participate as edge feature sources.
- Junctions—Junctions connect edges and facilitate navigation from one edge to another. Point feature classes can participate as junction feature sources, but multipoint feature classes cannot.
- Turns—Turns store information that can affect movement between two or more edges. Turn feature classes participate as turn feature sources in a network. A turn feature source models a subset of possible transitions between edge elements during navigation.
Edges and junctions form the basic structure of a network. Connectivity in a network deals with connecting edges and junctions to each other. Turns are optional elements that store information about a particular turning movement; for instance, a left turn is restricted from one particular edge to another.
What is a Neural Network?
Microtubules play a very important role in the internal transport within neurons. They form a network resembling a road network that runs from the soma (the cell body) all the way down to the axon (the synaptic terminal). Different substances are moved along this network of microtubules with the help of two proteins – kinesin and dynein. These substances include synaptic vesicles, which contain neurotransmitters; proteins needed within the cell; lipids; and even organelles such as mitochondria.
Kinesin and dynein are able to transport substances from the soma to the synaptic terminal, but also in the other direction going from the synaptic terminal to the soma – this is known as ‘axonal transport.’ In addition, microtubules help to transport nerve signals as synaptic vesicles are shuttled down the microtubule road network all the way to the synaptic terminal, where the neurotransmitters that they contain are released into the synapse. http://www.youtube.com/watch?v=-qV__tYb4c4
Considering that materials can move back toward the soma, or signals can move to another neuron, we can equate the axon of the nerve cell with the edges within our geospatial road network; the microtubules within the axon with the lanes of a road; the soma with the junctions of our network; and the synaptic terminal with the turns and junctions of our network.
In fact, we can even equate the electrical and chemical synapses with the types of roads – highways equal high-speed electrical synapses and local streets equal lower speed chemical synapses. The electrical synapses also send an ion current that flows from the cytoplasm of one nerve cell to another via a gap junction. In the case of chemical synapses, the neurotransmitters are moved through synaptic junctions to communicate with another neuron. http://www.youtube.com/watch?v=VitFvNvRIIY
The speed of chemical synapses is strategically sub-optimized, and this is how the number of signals within the high-speed network is controlled. It’s the same reason why we don’t allow bicycles and mopeds on our highways here in America: bicycles and mopeds are slower and will clog up a road. On the other hand, they can reach areas of the network that a car or truck may not be able to reach. http://www.youtube.com/watch?v=VitFvNvRIIY
When building a geospatial network within a computer software program, strict rules are encoded to communicate the directionality, minimum speed, and road surface properties of the network. For instance, a rear-wheel drive sedan can adhere to the directionality and – probably – the minimum speed of a road, but if that road is riddled with potholes and ice patches, the sedan may not be able to traverse it successfully, thereby blocking access to the four-wheel drive trucks that could easily traverse the road. This points to another reason why there are electrical as well as chemical synapses in our bodies. The chemical synapses – our sedan – do not have the speed or agility to traverse the network that leads to the heart. Without an electrical synapse – our four-wheel drive truck – we run the possibility of only having half of a heartbeat. http://www.youtube.com/watch?v=VitFvNvRIIY
Artificial Neural Networks Help to Forecast Road Traffic
Artificial neural networks (ANNs) are mathematical models that simulate the working facets of biological neural networks. Each network includes simple processing units and a set of connections between them. Neural networks are non-linear statistical tools, which model complex relationships between inputs and outputs in order to find patterns in data. https://www.britannica.com/science/cognitive-science/Approaches#ref1121999
For my final capstone project at Penn State University, I utilized a software program with machine learning capabilities to help me forecast repeat burglaries in the District of Columbia. I used an unsupervised machine learning technique to allow the computer to tell me what ecological factors — i.e. distance to police stations, level of street lighting, level of ingress and egress, etc. — most likely contributed to a criminal burglarizing the same location more than once.
The first step in the process was to build an ANN for deep learning. I did this by first creating a training dataset to allow the computer to build a signature of where future events may take place. The training dataset included half of the known repeat burglaries in one year (from January through June). After the computer built the model, I compared the known repeat burglaries from July through December with the model’s forecast. The model successfully forecasted the repeat burglary locations for July through December. I then ran the model through several years of data, and although some years did better than others, overall the model successfully adapted to changes in criminal behavior.
Being able to forecast criminal behavior via machine learning is just one example of how ANNs are contributing to artificial intelligence. They are also being utilized by many companies to help enhance geospatial road network data. By using historical data and monitoring existing users, apps like Google Maps are turning traffic forecasting into something of a science. Google is accomplishing this by monitoring millions of users around the world and using their location and average travel speeds. The more people use the app at any particular place and time, the more accurate the traffic forecast becomes. https://interestingengineering.com/neural-networks-are-being-used-to-help-predict-road-traffic-more-accurately
The Converse Watchmaker Argument
The designs we observe in biology, such as the neural network, are exactly the types of designs we would anticipate seeing if a Creator is the reason for life’s origin, history, and design. Yet, “evolutionary mechanisms (based on unguided, directionless processes that rely on co-opting and modifying existing designs to create biological innovation) are expected to yield biological designs that are inherently limited and flawed.” Would engineers really turn to biological designs to inspire their work if biological systems were truly generated by an unguided game of chance and necessity? (Design Principles Explain Neuron Anatomy)
The converse Watchmaker argument is a term coined by Dr. Fuz Rana of Reasons to Believe and promotes the idea that biological designs can inspire engineering and technology advances; thereby, allowing for a new argument to be made for God’s existence.
“If biological designs are the work of a Creator, then these systems should be so well-designed that they can serve as engineering models and otherwise inspire the development of new technologies.”Dr. Fuz Rana, Reasons to Believe, Design Principles Explain Neuron Anatomy
Even though the evolutionary paradigm assumes that biochemical systems evolved via an unguided process of chance and necessity, human-built neural networks and biological neural networks do have identical mathematical structures. Where did this information come from? Bioengineers are still convinced that there must be a naturalistic design principle that explains these identical structures despite there being a lack of evidence to support this view – “evolution of the gaps” theory. (Design Principles Explain Neuron Anatomy)
The paradigm that evolutionists embrace demands that they view biological systems as flawed and imperfect. Yet, these systems appear to be designed for a purpose – the Watchmaker is not blind. Biologists routinely use design language and seek design principles when describing and studying these systems. They “operate as if life was the product of a Creator’s handiwork, though they might vehemently deny God’s influence in shaping biology.” In the case of biological neural networks, the de facto design paradigm of evolutionist researchers has paid off with the advancement of deep machine learning via artificial neural networks. (Design Principles Explain Neuron Anatomy)
Issac Newton and most of the other founding fathers of science argued that the universe can only be fully explained with a combination of natural and supernatural rationalizations. Intelligent Design proponents only invoke God in origins when a supernatural action is required according to the laws of science. (Evolution – Not Creation – Is a God of the Gaps)
Even Richard Dawkins had this to say about the origins of life:
We have no evidence about what the first step in making life was, but we do know the kind of step it must have been. It must have been whatever it took to get natural selection started . . . by some process as yet unknown.Richard Dawkins, The Greatest Show on Earth, pg. 419
Dawkins’ statement is a classic example of evolution being a “god of the gaps” elucidation. He is clearly invoking the god of evolution to fill in the gap and proclaims that natural selection “must” have begun somehow. “Natural selection by itself cannot create anything; it can only select from things already created.” (Evolution – Not Creation – Is a God of the Gaps)
Many distinguished scientists reject both chance and necessity. Given that both premises are not rationally plausible explanations, it seems that design is the best, if not the only, reasonable explanation for such organization in neural networks. How can information, knowledge, and the ability to learn come from a random, accidental source? Naturalism, essentially, purports that life and reason came from a source that lacked all of these characteristics. However,the Creation model argues that “Systems and structures produced by intelligent agents typically possess characteristics that distinguish them from those produced by natural processes.” (“Origins of Life“, pg. 44)
“If God exists, a delicately balanced universe isn’t surprising at all. If God doesn’t exist, shock is appropriate.” (Loving Wisdom, pg 85 – as quoted by Samples) There have been numerous scientific surprises that have befuddled origin of life researchers over the years. For instance, MIT cosmologists have been stumped as to why our planet is not a water world. According to their models, we should have about a thousand times more water than we do. Evolutionary science has failed to explain this (and numerous other examples), but the Bible has not. Rana and Ross argue that Genesis 1:2, Deuteronomy 32: 19-11, and Luke 1:35 all speak of the Spirit of the Lord hovering over the the Earth protecting something precious on the planetary surface – the seeds of life. (Fuz Rana, Setting the Stage for Creation)
Neuroscientists have successfully grown miniature brains in the lab that have human-like neural activity. Does this mean that God was not involved in the production of neural networks? Certainly not! While undirected chemical processes can produce neural networks under carefully controlled laboratory conditions, early Earth’s state was anything but pristine. How do these scientists know whether the methods they used to build the neurons would have operated in the early Earth environment? Is it possible that an all knowing divine designer, who hovers over life, is responsible for the formation and organization of neural networks?
Machine learning via artificial neural networks mirrors how our brain learns and adapts to its environment and thus contributes to a converse Watchmaker analogy. There is an unbelievable amount of computer code that goes into building a basic geospatial network, much less a machine-learning neural network. Machine learning code involves understanding how to build appropriate algorithms for a system to learn and adapt within a constantly changing environment. “Evolution of the gaps” suggests that like the clogged, inefficient roads of India, evolution somehow cobbled together a neural network that will work well enough to keep people alive.
The intelligent design argument, however, rebuts this paradigm by showing a need for an all-knowing computer programmer who knows when to encode rules for an electrical synapse – a trucks-and-cars-only road – or a chemical synapse – a road designed for bicycles and mopeds. The Divine Cartographer specializes in building the neural road networks of our mind. He knows how to build complex rules for the neural road to ensure that materials can move back and forth along the axon road network and that signals can move between neurons. He even knows when to sub-optimize synapses in order to prevent signal traffic jams. Psalm 147:5 says, “Great is our Lord, and abundant in power; his understanding is beyond measure.” (ESV) Praise be to The Great Cartographer!