When to expect arrival of true Artificial Intelligence?

How, what is considered, AI today will form the basis to build true artificial intelligence in the future

The excitement around AI as the next frontier of computing is ubiquitous and well deserved. However, there is a lot of confusion about the current state of AI evolution. There is even more confusion about how close are we to building true artificial intelligence. In this post, I will share my thoughts around the subject.

Setting the bar too low for AI is misleading

First, I don’t believe we are any where close to building true artificial intelligence. In my opinion backers of the theory, that we are really close to building AI, are setting the bar too low to declare a machine artificially intelligent. A machine cannot be declared artificially intelligent just because the machine can perform a certain task better than human beings. Machines have been able to solve mathematical computations better and faster than humans since the days of a calculator. Clearly, a calculator cannot be considered artificially intelligent. Similarly, expert systems that can perform specific tasks better than human beings have been in existence for decades. But these expert systems cannot be considered artificially intelligent either.

Critical threshold for intelligence

To understand how AI will evolve in the next decade or so we need to be very clear about two things. First, we must understand that intelligence is not a binary domain (I will do a separate blog post on this). If anything, intelligence varies on a scale more than a binary outcome of a yes or no (hence the imperfect measurement of I.Q.). Second, we must not set the bar too low on the intelligence scale to declare a machine artificially intelligent.

So what is true artificial intelligence?

In my view, true artificial intelligence is something that is capable of doing three very specific tasks in its domain of existence (or surroundings).

1. Recognize and perceive elements in the domain of existence
2. Create/manipulate knowledge base and context of elements in domain of existence
3. Manipulate domain of existence well enough necessary for survival

Recognize and perceive elements in domain of existence

To be truly intelligent, a machine must be able to recognize elements in the domain of existence. In human beings, this job is done by the eyes and the visual cortex. Humans are able to “see” what is around them and recognize the objects as separate elements. For example, we are able to see a tea cup sitting on a table.

Not surprisingly, we are feverishly working on computer vision to enable image segmentation and object detection in machines. Machines are getting better at recognizing objects through newer techniques such as deep neural networks etc. However, I believe machines are about 5+ years away from reaching human capabilities of image segmentation, object detection, motion tracking etc.

Create/Manipulate knowledge base and context of elements in domain of existence

Without context and knowledge base, identification of elements has very little value. For example, if a human opens their eyes and all they see is water around them, without knowledge and context, the human could be taking a bath in a bath tub or drowning in a lake.

There are a couple of major problems we need to solve before machines can reach human capabilities of creating and manipulating knowledge base and context. First, machines must be able to acquire and store an enormous amount of data. The world around us is vast and complex, it is unclear how machines can store that much data. Second, suppose even if we can store that much data, it is also not clear how we can manipulate and create context around that data. Intelligent machines will need to store and manipulate enormous amount of data in complex and readily accessible structures (possibly mimicking human memory i.e. Random, Episodic & Relational).

Lots of research is happening in the knowledge representation and context creation; but I believe we are about 10+ years away from creating something that can match human capabilities of knowledge base storage and context creation.

Interact with domain of existence well enough to survive

Perhaps the most important task any being in existence does is to interact with surroundings such that it can survive. For example, if we are able to recognize and perceive that we are standing in the middle of the road and a car is coming at 65 miles an hour in our direction, we will manipulate the environment by moving such that we are not in the way of the car anymore.

Human beings have been building robots for a while now to manipulate the environments. Robotics capabilities are becoming better every passing year and I foresee a linear improvement in robotics to match human capabilities in about 15+ years.

Damned if you do and damned if you don’t!

However, the fundamental issue is that whether we allow the machine to interact and manipulate the environment necessary for survival. There is no easy answer to this question. If we allow the machines to manipulate the environment for survival, one day the machines can decide human beings need to be eliminated for the machines to survive. On the other hand, if we don’t allow the machines to interact and manipulate the environment for survival, the machines will never become artificially intelligent as they will not survive. The machines must be able to survive to keep perceiving, keep acquiring knowledge/context and keep interacting and manipulating the environment for their survival; Only then the machines will be able to become smart enough to be considered truly artificially intelligent.

I foresee human beings to struggle with this dilemma…. forever!