How close are we to creating artificial general intelligence?

In 2005, the popular and controversial futurist Ray Kurzweil released one of his most popular books: The Singularity Is Near. What did he mean by singularity? The moment when machines would have processing power superior to that of the human brain, something that until then had only been glimpsed in science fiction and in perhaps not-so-friendly scenarios, considering examples such as The MatrixandThe Terminator.

On the other hand, films such as Her present a milder side of artificial intelligence and its relationship with human beings. There, we have an example of how an artificial intelligence can become creative and even demonstrate some kind of consciousness. According to experts in the field, this level of potential for an artificial intelligence would already place it on a new level, that of Artificial General Intelligence (AGI).

In an article for the Futurism website, Dan Robitzki comments that for decades, scientists have been testing different methods to create this AGI, with techniques such as reinforcement learning and machine learning being some of the possibilities for exploration. However, as Robitzki states, none of these techniques have been sufficient to reach this AGI level. And worse: apparently, no one knows very well what steps to take next.

In August, however, scientists gathered in Prague for the Joint Multi-Conference on Human-Level Artificial Intelligence to discuss their goals, hopes, and the progress made in developing an artificial intelligence that resembles human capacity alongside global leaders. For some, talking about an artificial intelligence that resembles a human is the step before we finally achieve artificial general intelligence.

But although there is still no consensus on how we will reach this AGI, experts believe that we are getting closer and closer to it: about 37% of the responses to a survey conducted among event participants indicated that an artificial intelligence that resembles a human being will be created within the next 10 years. Meanwhile, 28% of respondents believe it will take another 20 years, and finally, only 2% believe that this type of technology will never exist.

Some of the questions that loom over achieving this scenario are, for example, who will fund such an endeavor and how we will combine algorithms to acquire an AI capable of reasoning like a human being. To this end, Robitzki interviewed some artificial intelligence experts who gave their opinions on the subject.

Irakli Beridze, Head of the Centre for Artificial Intelligence and Robotics at UNICRI, in the United Kingdom

“At the moment, there is absolutely no indication that we are close to an AGI. And no one can say with any kind of authority or conviction that this will happen within a certain period of time. Or worse, no one can say if it can even happen. We may never have an AGI, so we need to take that into account as we discuss the subject.”

Seán Ó hÉigeartaigh, Executive Director of the Centre for the Study of Existential Risk at Cambridge

“There is still a lot of work to be done; there are many things we don't understand yet. Once we understand them, it might be possible to make AGI happen in the next 50 years.”

John Langford, Principal Researcher at Microsoft AI

“I think we should enjoy the technology more as it advances. We should look for a future scenario where we want to land. But, on the other hand, it’s not like we have human-level artificial intelligence right now, and I don’t think that will happen very quickly. I think, if I’m lucky, it will happen in my lifetime.

An intelligence level of a worm is already quite possible to emulate. If you take into account vision and planning, that kind of thing is quite possible. The integration of planning and learning, planning itself is already quite well resolved. But planning means that the way [machine learning] works is still not very well resolved.”

Marek Rosa, CEO and CTO at GoodAI

“I think we’re almost there. I’m not predicting that we’ll have AGI in the next three or thirty years. But I’m confident that it will happen any day now.”

Ben Goertzel, CEO of SingularityNET and Chief Scientist at Hanson Robotics

“I don’t think we fundamentally need new algorithms. I think we need to connect our algorithms in different ways than we do today. If I’m right, then we already have the basic algorithms we need… I believe we are less than ten years away from creating human-level AI.”

Hava Siegelmann, Program Manager at DARPA

“I don’t think we are very close to achieving AGI. I think artificial general intelligence is almost a badge for a generalist idea. Lifelong learning is an example of this, and it’s a very specific type of AI. We already know the theoretical principles for it, we know how nature works, and it’s all well-defined. There is a clear direction, there is a metric. I think we will be able to achieve it in a short time.”

‍