Sergey - stock.adobe.com
While the power and capability of current artificial intelligence applications is certainly proving to be highly valuable, we're still far away from the vision of AI that science fiction and fantasy predicted. Currently, we only have narrow applications of AI, which focus on designing systems that perform specific tasks using well-defined boundaries. In fact, all current use cases, ranging from AI-powered chatbots to facial recognition to self-driving cars, are under the category of narrow AI applications.
Rather than single-purpose systems that can do recognition, conversation or autonomous control, is it possible to design a generally intelligent system that can handle a wide range of cognitive tasks, while adapting without continuous retraining? Futurists and researchers are striving for much broader intelligent machines with artificial general intelligence (AGI). The goal, which currently still seems just out of the grasp of technologists and researchers, is an intelligent system that can respond similarly to intelligent beings.
The goal of AGI is to enable intelligence with similar cognitive capacity as a human being. Generally intelligent systems are designed to not only carry out basic or narrow tasks, but also handle a wide range of tasks in different domains and be able to adapt to complex settings. The grand vision of generally intelligent systems is to be able to do mostly anything that a human being can -- namely, to think quickly and react instantly to solve problems.
The state of AGI research
While we have not achieved the goals of AGI yet, there is widespread disagreement on how long it will take to reach that goal and whether or not it's even possible. Experts in the field differ on estimating the length of time it will take us to get to AGI. Some researchers say AGI is just around the corner, perhaps even within the next decade or two. Others feel we're hundreds of years away from true AGI, if it is even a realizable goal at all.
Despite the time frame, many companies are beginning to invest in AGI research either directly or indirectly through deep learning and neural networks. Most notably, DeepMind, an advanced AI research company that was subsequently acquired by Google's parent company, Alphabet, is spending significant amounts of money and time pursuing the goal of AGI and using reinforcement learning, among other approaches, as its means to get there.
DeepMind is famous for its 2016 AlphaGo program that beat a human world champion Go player, Lee Sedol, in a five-game match. In 2018, the vendor introduced its Impala technology as a singular system that had the capacity to learn multiple different tasks. While traditional narrow AI is more dependent upon single tasks that it is assigned, Impala was able to learn up to 30 different tasks. This groundbreaking technology is one step closer to creating AI systems that can truly think for themselves and adapt to new environments.
Since then, DeepMind has worked to find uses for this technology in other areas to see how it can expand its systems to learn and grow. Its various research efforts are aimed at leading a more constructive path for future AGI attempts to help the systems overcome any hurdles.
DeepMind isn't alone in pursuing artificial general intelligence. Microsoft recently announced a $1 billion investment in OpenAI, a company focused on the research and development of AGI. This money will be used to create machine learning models that will hopefully realize the long-term goals of AGI.
OpenAI has already released groundbreaking machine learning models, such as GPT-2, which has been able to create impressive amounts of machine-generated text that's almost indistinguishable from the works of creative human authors. Samsung, under its subsidiary Viv Labs, is investing in AGI research, with a special focus on AGI for its primary projects, including the eventual design of fully active virtual assistants.
The future of AGI
Because even researchers don't know all the facets of how the brain works, developing models for replicating it means company R&D efforts are taking different approaches to realizing the goals of AGI. Many researchers are focusing on a neuroscience-related base for their system designs in order to best build out a digital brain that can function on the same level as a human being. Others are thinking about building sophisticated neural networks combined with expanding knowledge graphs as another way to realize some of the goals of AGI. Still, others are bringing out humanoid companion robots that have been revealed to be just smartly developed machine puppets that are still at the whim of their creators.
This ultimate goal of AGI has attracted plenty of big-name supporters, from Bill Gates to Elon Musk and Alibaba founder Jack Ma. With that kind of interest and support, it is safe to assume that we can expect to see continued interest and investment in AGI research.
However, the timeline for actually reaching this goal is still up for debate. In the meantime, it is safe to assume that the more time spent trying to build the ultimate intelligent machine will result in interim technologies that will continue to prove their value. The journey to create fully functional AGI systems is perhaps the destination, whether or not we get a fully functional artificial general intelligence system within our lifetimes.