RESEARCH REPORT

In brief

In brief

  • Artificial intelligence (AI) is behind many of the advances we’ve seen in areas like healthcare, customer experience and predictive maintenance.
  • From cancer screening to reliable products to intelligent home speakers, advances in machine learning are improving many aspects of our lives.
  • Still, today’s AI models are narrow, and typically fail if applied outside the domain for which they were trained.
  • Although true artificial general intelligence (AGI) is still a distant prospect, businesses can benefit from improved machine reasoning.


Think like a human

The ultimate goal of artificial general intelligence is to replicate the broad range of human cognitive abilities. Sometimes also called “strong AI,” AGI aims to create machines capable of general intelligence—the kind we typically associate with broad competence such as common-sense reasoning.

Common-sense reasoning is a cognitive capability that we apply all the time. Unlike a conventional machine learning system, a child understands that if she goes to school, so do her toes. And that she won’t cast a shadow at home while she’s in gym. And that Beethoven never had an iPhone. Achieving this same type of flexible reasoning with AI has long been a goal, but it’s much easier said than done.

The value of artificial general intelligence

The success of machine learning on narrow tasks has sidetracked us from the goal of artificial general intelligence. Since machine learning is designed to handle classification problems and does it so well, we’ve gotten used to construing every problem that way. We rarely stop to consider problems of a different kind altogether: those that require broad reasoning capabilities.

Almost 70 years after the charter was set for AI, it might seem disappointing to note that we’re still relatively early on the journey toward true AGI, and that reaching it is likely a way off. But there’s good news for business: there’s value in the journey to AGI itself, and significant untapped potential waiting in the AI systems that are beginning to exhibit some early traits of what we might characterize as AGI.

In recent years, a series of “language models” developed using deep learning have been released. This capability can be used for applications like question answering, semantic search, and text generation with results that sound surprisingly natural.

Spot the right opportunities

There’s a lot of untapped potential in today’s machine learning approaches. And common-sense reasoning—the hallmark of what we’d call AGI—is already present in every employee and every customer. So do businesses even need machines with common-sense? Wouldn’t it be common-sense to rely on people for common-sense?

The answer is that there are many situations where even a small degree of common-sense reasoning can make a big difference to machine operations. As with other areas of automation, the best results are often achieved when humans and machines work together.

Better customer service

Accenture has been prototyping a chatbot capable of scriptless interaction involving modelling with enough granularity to capture fine distinctions.

Simple reasoning

We can extract knowledge from language models for common attributes. Then we can plug the informal knowledge gaps so much common sense depends on.

Better compliance

We’re working on a broad framework to allow structured domain-specific event monitoring applications that support a limited amount of reasoning.

View All

Exploring the AGI frontier

Machine learning and narrow AI solutions have been behind so many of the extraordinary recent advances we’ve seen in healthcare, customer experience, predictive maintenance and elsewhere. And they will continue to be a central driver of business value into the future.

As we progress toward the long-term goal of common-sense machine reasoning, AI solutions will gradually expand the breadth of situations they can handle. Emerging language models will improve, as will our ability to extract knowledge from them and apply that knowledge in productive business applications.

About the Authors

Andrew Fano Ph.D.

Managing Director – Artificial Intelligence R&D, Accenture Labs


Jean-Luc Chatelain

Managing Director – CTO Applied Intelligence


Shubhashis Sengupta, Ph.D.

Lead – Accenture Technology Innovation, APAC


Vivek Khetan

Technology R&D Associate Principal

MORE ON THIS TOPIC

Alepa Block Wish: Your wish, our command
Applying knowledge graphs for social good

Subscription Center
Stay in the Know with Our Newsletter Stay in the Know with Our Newsletter