Conjure up an image of a machine with general intelligence and HAL from 2001: A Space Odyssey or Samantha from the movie HER might come to mind. Putting these fictional ruminations aside, though, Artificial General Intelligence (AGI) is currently the holy grail for computer scientists. An AGI application would use adaptability, learn-ability, and a broad range of applicability to solve arbitrary challenges from different fields and in varying environments, without explicit instructions.
First, let’s be clear: AGI doesn’t exist yet. Researchers are making inroads, but experts put it anywhere from 10 to 40 years away. (If you’re interested in how we’ll know when it’s here, there are a range of tests, including the long-standing Turing test and the more recent “coffee test” and “robot student” tests, to name a few.)
Still, when AGI finally arrives, opportunities for enterprise applications will be vast. Think about cybersecurity, where threats from attackers are pervasive for companies across industries. AGI solutions will dynamically reason and react, apply countermeasures, verify success, modify approaches from feedback, and automatically strike back against emerging threats.
Or think about a universal translator, a la Star Trek—such an AGI will learn vocabulary, decipher grammar, and discover patterns automatically. This is about more than just helping people talk to each other; it can also allow smart machines to talk to each other and create a “self-organized coherent unit” that can make decisions and operate as a team.
The opportunities are enormous. But with so much noise around artificial intelligence, how do you distinguish AGI from other emerging categories of AI? The idea that Human Level Intelligence (HLI), Strong AI, and AGI are synonymous, for example, is a common mistake. HLI exhibits human-level capabilities specific to a certain task. Strong AI solutions can sometimes demonstrate or even exceed human capabilities, but only in very narrow applications. Unlike these, AGI is a “broad spectrum” solution: when it becomes reality, it will potentially be able to apply itself to any type of problem.
If AGI isn’t here yet, what AGI work are we doing at Accenture today? We’re laying the groundwork that will allow AGI to be successful at scale.
Accenture and Accenture Labs are developing applications that will provide some of the core capabilities needed for AGI to become reality, like understanding, inferencing, reinforcement learning, domain transfer learning, and common frameworks with knowledge graph underpinnings.
Knowledge graphs and upper ontologies, for example, enable a common-sense inferencing mechanism, store learnings and domain insights, allow for semantic linking of concepts, and enable searching knowledge at scale. We’ve already adapted knowledge graph solutions for customer care, where our application enables semantic understanding of customer complaints, Customer 360, and customer relationship management, while also incorporating reinforcement learning based on previous customer interactions for rapid resolution of complaints.
In oil exploration, we’ve built a knowledge graph solution that uses documents, reports, events, and oil well configurations to provide an explainable foundation for understanding why certain events occur, discovering similarities, and transferring learnings from previous oil well deployments into anticipating supply needs.
We’ve also developed conversational agents that can reason with the help of a knowledge graph. For example, a smart travel advisor can suggest nearby sightseeing options to a tourist based on that tourist’s profile and by making inference through a combination of knowledge graph and real-time traffic and weather data.
And we’re incorporating reinforcement learning into supply chain decision support, where a human expert can provide real-time rewards and penalties that encourage learning with feedback within a system.
All of these applications have business value and impact today—and are laying the groundwork for AGI at scale down the road. We believe that instead of a “big bang” AGI era, the ever-faster and wider evolution of computer systems and algorithms will gradually introduce a set of fundamental changes to how machines gather and structure knowledge, think, deduce and reason.
The realms of knowledge and semantic technologies, transfer and reinforcement learning, and brain-inspired computing will coalesce into an exciting set of business possibilities. Systems will graduate from mere trainable pattern recognition systems to the machines that “learn to learn.” At Labs, we’re building the path to the AGI future. Stay tuned to see what else we’ve got in store.
For more information about our AGI work at Accenture Labs, visit the Artificial Intelligence and Systems & Platforms R&D group websites, or contact Shubhashis Sengupta and Colin Puri.