August 11, 2019 by Austin Torres

AGI is the intelligence of a machine that can replicate human cognition and perform any task that a person could. Artificial General Intelligence means a machine must be able to do more than complete a task using a purpose-specific algorithm. Instead, it needs to be capable of learning to solve real-world problems much like a human learns through experience.

One way to potentially achieve this level of intelligence in a machine is to simulate the functionality of a human brain. This is done by scanning and mapping every synaptic connection in a human brain and copying these connections into one powerful state machine. Theoretically, this would allow for the development of connections that mimic the computations done by the human brain. The result would be a machine that thinks, learns, and lives as a human would.

Baum, S. D. (2017). A Survey of Artificial General Intelligence Projects for Ethics, Risk, and Policy. SSRN Electronic Journal. doi:10.2139/ssrn.3070741

Testing the ability of AGI

One challenge in developing artificial general intelligence is defining what exactly differentiates an AGI-capable machine from a more traditional machine that merely uses algorithms to solve specific tasks. An AGI machine needs to be able to learn, adapt, and perform actions at the level humans do. This is why experts have developed tests that can help define the requirements for considering a machine to possess AGI.

Turing Test

In this test, there are three separate terminals. Two are controlled by humans and one is controlled by a computer. One of the humans is given the role as the questioner, while the second human and the computer are participants that must answer these questions. After asking a preset number of questions, the questioner must use the responses to try and determine which participant is a computer and which is human. If the questioner fails to identify the computer at least half of the time, the computer is considered artificially intelligent because it is identified to be just as human as the human participant.

The Coffee Test

A machine must enter a home and be able to make a cup of coffee on its own. To do this, it has to locate the water, cups, coffee maker, coffee, and carry out the action of making coffee. This requires high-level processing in order to distinguish between and locate objects in a new environment, and the ability to perform a task using equipment that the machine is unfamiliar with.

The Robot College Student Test

In order to pass this test, a machine must enroll in a university, attend classes, and do all of the work assigned in the class. The machine needs to pass all of its classes and obtain a real college degree.

The Employment Test

This test requires a machine to be matched to a human with the same job title with performance that can be measured economically. In order to pass, the machine must perform at least as well as the human.

Completion of these tests displays a machines ability to learn, reason, plan for the future, communicate, and work towards solving a problem or reaching a goal. If a machine can pass tests like these, it is assumed that the machine is as capable as a human in society, thus proving the achievement of AGI.

“What is Artificial General Intelligence (AGI)? | 4 Tests For Ensuring Artificial General Intelligence”. Talky Blog. 13 July 2019. Retrieved 17 July 2019.

Current State of AGI

The fact that humans still do not completely understand the complex functionality and connections of the human brain, we are unable to accurately model and replicate a brain in the form of a machine. Yet, as the Church-Turing thesis states, any problem can be solved by developing an algorithm to solve it given infinite resources such as memory and time.

Although true AGI is out of our current reach, there are 45 known research and development projects aiming to further research in the broad field of AGI. One of the largest research efforts is the Human Brain Project. The Human Brain Project is attempting to build a functioning simulation of the brain using incredibly powerful supercomputers. Their scientists hope to further the understanding of the human brain and also build neurorobotics that can be controlled by these brain simulations.

Research like this is crucial for achieving AGI in the future by laying a solid foundation and furthering the understanding of the human brain.

Joshi, N. (2019, June 10). How Far Are We From Achieving Artificial General Intelligence? Retrieved August 11, 2019, from https://www.forbes.com/sites/cognitiveworld/2019/06/10/how-far-are-we-from-achieving-artificial-general-intelligence/#540a19336dc4

Computing for Brain Research. (n.d.). Retrieved from https://www.humanbrainproject.eu/en/massive-computing/