Nearing the Singularity, Thanks to Our Brains?

It can be daunting, looking at the field of Artificial Intelligence and wondering if you fit into a future that involves so many elements of uncertainty. It’s our aim here to provide you with the information that you need, to learn just about anything in the AI space. When you’re getting started with any industry, it’s logical, to begin with, its’ foundation, which in this particular case means artificial neural networks.

Artificial Human Brains

It’s been said by experts in the space that the best way to understand how artificial neural networks work is to think of them as “artificial human brains,” in the way that they function. Some in the space like Cade Metz of Wired Magazine and the New York Times says that this isn’t a perfect analogy. In attempting to make this direct connection between artificial and natural neural networks, we leave out the fact that we are nowhere close to fully understanding and mapping everything about the human brain.

It might be most accurate, however, to state that the fields of cognitive and computational neuroscience have the best chance of discovering these answers that we are missing. If we were to make another connection between the field of AI and the study of the human brain, we could simplify cognitive neuroscience down to the study of the algorithms that run our cognition or, the way we think and why we think that way.

What is Computational Neuroscience?

Computational neuroscience appears similar and yet different at the same time. Firstly, it appears to consist of two major research paths. Scientists in the field use mathematical models and theories to study the way the brain works. They also combine parts of electrical engineering, computer science, and physics to try to discover more about how our nervous systems process information.

Given the connection between these fields and the ideas behind artificial neural networks, certain leaders in the AI field like Demis Hassabis, who founded Google’s Deepmind AI, have argued that for a merger between neuroscience and AI.

You might be wondering now: how exactly are ANNs connected with neuroscience?

It’s actually quite simple. As mentioned above, it’s not quite reasonable to make a direct connection between them and how the human brain works. Think about this idea more like the community’s best attempt to conceptualize a process that they don’t even fully understand. This is not to say that Artificial Neural Networks aren’t similar to the human brain in any way. One Ph.D. of neuroscience said that the two can be compared in terms of five qualities: synaptic learning, synaptic structure, brain regions, neural connectivity, and individual neural dynamics. Going completely into all of these qualities on a more basic level would take much longer than the scope of this piece.

Neural Connectivity

For the sake of brevity and clarity, we’ll stick with talking about the neural connectivity in a network as a whole. Most existing ANNs are structured as feed-forward neural networks. On a basic level, this means that some sort of input value is fed through nodes on a digital network, which can be compared to neurons. From here, in the simplest sort of FFNN, the values are then fed into an output layer and put through a mathematical function involving the input values and the weights, which are special values that feed the inputs into the output layer.

If this is a bit confusing, try to simplify all of it to a mathematical function. Information flows through the function it until a value comes out that tells the system a decision to make. In the overall field of machine learning, this decision-making process often relates to the Delta rule, which is a mathematical function that essentially helps AI systems to learn from their mistakes.

In other words, they take in data and decide a certain way to act on it but as they take in more and more data, they improve their decision making based on learning by doing. A prime example of this may be seen in speech recognition systems like Amazon’s Alexa and Echo as well as Apple’s Siri. All of these systems were tested with real human voices in connection with functions similar to the Delta rule to learn from their mistakes, as they go. Understanding these concepts isn’t easy. Even so, we hope that with this basic understanding of Artificial Neural Networks and how they relate to the human brain, you’ll continue on this journey with us through the world of Artificial Intelligence.