In Artificial Intelligence, the most difficult problems are informally known as AI-complete or AI-hard.

The difficult nature of these computational problems are equivalent to solving the central Artificial Intelligence problem—making computers as intelligent as people, otherwise known as strong AI.

Strong AI’s goal is to develop artificial intelligence to the point where the machine’s intellectual capability is functionally equal to a human’s.

There is a common misunderstanding with AI. When we solve a challenge, like mastering the board game Go, we often believe that we are on the brink of building superintelligence, when what we are really doing is solving narrow challenges and operating within a limited and pre-defined range of functions using brute-force computational strength.


AI-complete problems cannot be solved until humans come up with a deeper solution for developing human-like artificial intelligence. Problems include human-level image filtering or human-level natural language processing, or dealing with unexpected circumstances while solving a real-world problem.

AI-complete problems will more than likely be solved using Artificial General Intelligence.

Artificial General Intelligence (AGI), also referred to as “strong AI” is currently non-existent software that can successfully perform any intellectual task that a human being can, using the intelligence of a machine.

Artificial Narrow Intelligence (ANI), also known as “weak AI”, is an artificial intelligence that is purely focused on one narrow task. All existing systems considered artificial intelligence are weak AI at most.

There are a number of tests for AGI:


Brittle: hard but liable to break easily.

Brittle systems are defined as systems that are characterized by a sudden and steep decline in performance as the system state changes.

Examples of brittleness in software:

  • An algorithm that allows a divide by zero error to occur.
  • A curve-fitting equation that is used to extrapolate beyond the data that it was fitted to.
  • The use of data structures that restrict values.


Brittleness occurs in many AI systems because many systems rely on significant assumptions about the input data. When these assumptions are not met, the system responds in completely unpredictable ways.

Human computation

Current AI-complete problems cannot be solved with modern computing alone, we still require human computation.

The term “computer” or “one who computes” has been in use since the early 17th century. Until only recently it referred to a person who performed calculations for a living.


NACA High-Speed Flight Station "Computer Room" (1949)

Human-based computation is where a machine performs its function by outsourcing specific steps to humans, usually as micro-tasks. It helps us to achieve symbiotic human-computer interaction.

In traditional computation, a human directs the computer to solve a problem by providing a formalised problem description and an algorithm to a computer and receives a solution.

Human-based computation reverses the roles; the computer asks a person or a large group of people to solve a problem, then collects, interprets, and integrates their solutions.

Solving AI-complete problems

AI-complete problems cannot be solved until humans come up with a deeper solution for developing human-like artificial intelligence.

One company working on this is Google DeepMind with their PathNet project, which is a new Modular Deep Learning (DL) architecture that blends Meta-Learning, Reinforcement Learning, and Modular Deep Learning, into one solution.

Check out the paper on Arxiv.

Lets end with a quote from the paper:

For artificial general intelligence (AGI) it would be efficient if multiple users trained the same giant neural network, permitting parameter reuse, without catastrophic forgetting. PathNet is a first step in this direction. It is a neural network algorithm that uses agents embedded in the neural network whose task is to discover which parts of the network to re-use for new tasks.

Check out our previous blog post Machines that learn with Transfer Learning and Machines that learn with Reinforcement Learning.

To understand how Adverai can help you create the meta-understanding of marketing, explore our website or get in touch by email.

Leave a Reply

Your email address will not be published. Required fields are marked *