How far are we from replicating the Human Brain into a Computer?

In this blog, we’ll explore various interesting facts about the human brain, then compare those facts with recent advancements in AI and see where are we currently in developing an artificial (vs human level) general intelligence.

Before we start, here’s a quick introduction about myself: I started my career as an AI Engineer and been fortunate enough in building various real-time AI systems (Computer Vision and NLP). I was fascinated by the fact that the human brain is the “most complex thing in the universe” and developed a great interest in neuroscience (understanding the human brain) and I hope it’s the same interest that brought you here.

Let’s get started with our topics:

1. Understanding the relationship between the human brain and computers
2. Interesting facts about the Human Brain and Decision Making
3. Human intelligence and Artificial Intelligence: The Differences
4. Will AI take over Humans?

Source

Understanding the relationship between the Human brain and Computers:

In Classical Neurology, the Brain is considered as a machine / a computer that performs Classification and Categorization. But wait there’s something special about the human brain, the “personal” (i.e) the ability to feel things and continuously judge it’s world(environment) and thereby continuously learning. Removing this personal from one’s brain makes him defective and he'll just be a computer/machine.

Dr. Oliver Sacks in his book “The Man who mistook his wife for a hat”, explains about a patient(old man) who can recognize Einstein from a picture but not his friends or family members. And the same patient when experimented with a rose flower. He was able to describe it’s(rose) features like six inches in length, a convoluted red form with a linear green attachment but unable to map this to his memory of rose flower. Even a kid could do this, so what made him defective? Was he lost in the world of lifeless abstraction? This is how a brain without its self(personal) behave.

Are we building an AI system that’s as defective as this patient? We’ll see more about this in the latter part of the article.

Interesting facts about the Human brain and Decision Making:

To keep things simple, I’ll try not to use any biological terms for brain regions or parts rather use some experiments or examples from V S Ramachandran and Oliver Sacks

• Intuitive mind vs Rational mind: As explained by Daniel Kahneman in his book “Thinking Fast and Slow”, there are two primary modes of thinking to process information and make decisions. Intuitive mind (System 1) which is fast, unconscious, emotional and it’s responsible for 90% of our decision making. Rational mind (System 2) which is slow, conscious, logical and it comes into play when the intuitive mind doesn’t have an immediate answer or to validate the thoughts of the intuitive mind.

“The intuitive mind is a sacred gift and the rational mind is a faithful servant. We have created a society that honors the servant and has forgotten the gift.”
— Albert Einstein

• Multiple regions of the brain are responsible for single action: Earlier it was believed that for each and every action, there’s a separate part of the brain responsible for it to happen. However, later researchers found that there’s a lot of complex processing involved even for a simple action (like “smiling”). Example: You smile automatically when you see a friend and it’s completely natural but when someone holds a camera and asks you to smile, the same action becomes difficult and looks unnatural. This is because there are two different “smile circuits”.
• What and How pathway of the brain: Our Visual and Auditory systems have two different pathways that guide us with our respective vision and hearing actions. Let’s take the vision system as an example: “How” or “Where” pathway is concerned with grasping, navigation, and other spatial information while the “what” pathway is concerned with recognizing objects. These two systems work together in an optimized way thereby making sense of what you see or hear.
• Filling in: The ability of our brain to “fill in” missing information automatically is a fascinating thing, it’s faster and automatic. For Example: Just imagine seeing a cat’s tail out underneath your bed, you don’t panic that your cat’s tail is cut, instead, you automatically visualize your cat underneath the bed by filling the missing image of the cat.
• Penfield Mapping: It’s a neurological “map” of the areas and proportions of the human brain dedicated to processing motor functions, or sensory functions, for different parts of the body.

Source: Penfield map

When you look at the above image, you could notice two things:

1. The weightage given to different parts of the body differs. Hand and Face have a higher proportions.
2. The ordering of parts varies and completely different from our physical appearance.

Another Interesting fact is that, When one of the part gets damaged or removed (example: removal of hand) that particular part in this map remains vacant and over the period of time the neighbouring parts occupy them. This is one of the main reasons behind “phantom limb”.

Let us now relate these awesome facts about the brain with current Artificial Intelligence systems and highlight the similarities and differences.

Human Intelligence and Artificial Intelligence: The Differences

Any definition of Artificial Intelligence will have to be vague enough due to our inability to define Human Intelligence. — Dr. Dan Roth, University of Illinois at Urbana-Champaign

For most of AI’s history, AI research has been divided into sub-groups:

1. Artificial Narrow Intelligence(ANI) or Weak AI: This type of AI is neither conscious nor emotional. Almost all the AI systems that we encounter today are of this type. They are mostly designed to perform a single task(say, classifying objects) or in some cases, multiple ANIs combined to build a complex system like a self-driving vehicle.
2. Artificial General Intelligence(AGI) or Strong AI: This refers to machines that have the ability to understand and learn any intellectual task a human being can. We have seen this type of AI in sci-fi movies like “HER”, “Ex Machina” and we are nowhere near to build this type of machine.

Having seen the various types of AI and facts about Human Intelligence, let’s start comparing their differences and look where we are now in solving real-world problems with AI.

• World or Environment: The world that we are in or the environment that surrounds us is continuous and infinite. And humans by nature continuously judge and learn their environment and get adapted to it. This is how we’ve evolved. In contrast, the AI systems that we have today works in a pre-determined world.

It is not the strongest of the species that survives, not the most intelligent that survives. It is the one that is the most adaptable to change. — Charles Darwin

• Context or Scene Understanding: What we see or understand is mainly influenced by our prior-knowledge and expectations. And it’s highly impossible to program this into a computer. Let’s get deeper into this with the example below,

Scene Understanding Example

There’s something common in these four pictures, yes it’s the “candle”. And at the same time, you’d have noticed that these pictures imply different emotions or scenes. Picture 1: shows(memorial and gratitude) children paying tribute to APJ Abdul kalam, Picture 2: shows protest against rape, Picture 3: shows(romance) candlelight dinner, Picture 4: shows a child praying to god. Let us now imagine an AI system built for Image captioning. Do you think it can understand these emotions or context and write relevant captions?

• Decision Making: In the second section, we saw how naturally we make most of our decisions. But we’re currently building an AI system that is neither conscious nor intuitive(or emotional). Hence it’s difficult to model our decision-making ability with AI, but on specific tasks, like “chess”, “Go”, Narrow AI seems to outperform humans.
• Fooling humans and AI systems: Humans are easily fooled by optical illusions whereas AI systems are easily fooled by adversarial examples. Let’s look at the image below,

Adversarial example(left) to fool face recognition systems, Optical illusion(right) to fool humans

The image in the left shows a woman wearing a t-shirt which is specifically designed to fool face recognition systems, an example of an Adversarial attack. The image in the right is more popular on the internet, the popcorn bag she holds tricks you and at first sight, you’d have thought her legs are extremely thin as if something is not quite right. To summarize, both AI systems and human beings can be fooled easily.

• Multi-tasking: Significant efforts have been made to build AI systems that are capable of taking a single input and performing multiple tasks at a time. One such example is “DecaNLP” by Salesforce. However, the tasks must be more relevant. In contrast, Humans are naturally capable of performing multiple tasks and some people train themselves to master the art of multi-tasking.
• Concept of Filling In: The very first thing that popped into my mind while reading about the “concept of filling in” in the human brain is how natural it is and if we can’t model this into an AI system, How are we gonna handle uncertainties?. Given that, the goal of our narrow AI systems is to model uncertainties specific to the task it performs, how are we gonna achieve it?
• Computation: Computers usually have higher precision and can perform 10 billion operations per second. However, most of the things are executed sequentially whereas the brain process information in a massively parallel way with very little power consumption compared to computers. If you’re more interested to learn about computation comparisons, I’d highly recommend you to read this article

We’ve seen a lot of facts about the Human Brain and Artificial Intelligence in general. Now let’s jump into the discussion on Technological Singularity.

Will AI Take over Humans?

We have seen a lot of sci-fi movies like “Eagle Eye”, “A.I. Artificial Intelligence” where AI takes over human beings but in reality, we are not even close to building “Artificial General Intelligence” systems.

Currently, we have “Narrow AI” which can only perform specific tasks in a pre-determined environment. So instead of replacing a human being in general (like “HER” movie) it can only facilitate human beings by making their work and personal lives easier with assistants like “Siri”, Alexa” or “surveillance” and “autonomous driving” systems or “health monitoring” systems.

However, it can replace specific professions like customer service, similar to the effect of automation.

There are some potential threats where AI can be used by humans for cyberattacks, AI-enabled Terrorism.

The last thing that I wanted to touch upon is Bias say Gender and Racial bias. Humans and AI are highly correlated by bias, except the fact that there have been significant efforts in addressing and reducing bias in AI.

Four polls of AI researchers, conducted in 2012 and 2013 by Nick Bostrom and Vincent C. Müller, suggested a median probability estimate of 50% that artificial general intelligence (AGI) would be developed by 2040–2050. Let’s wait till then to witness Technological Singularity.

Check out related blogs,

How businesses use AI to thrive in the new economy

Happy Learning…!