A lot of public figures, chief among them Elon Musk, Bill Gates, and the late Stephen Hawking, endorse that AI might pose an existential risk (think Terminator’s Skynet); but we are nowhere near the point where we really must be afraid of it.
While advancements like AlphaGo beating a Go world champion do make us ask if the computers have really surpassed humans, the point we have to understand is that the machine is essentially trying to mimic logic based on the millions of games it has seen. These games all have a lot of training data present, and the degree of freedom when it comes to the actions taken by the AI is vastly limited compared to the real world. Another thing is that these AIs we have are not multi-functional. This means that while they could be great at doing a single thing, they don’t do anything else. Making something as powerful as a machine that could think on its own and have real-world consequences is virtually a thing of the future. In fact, it might be better to think along the lines of Andrew Ng, who states that AI existential risk is “like worrying about overpopulation on Mars when we have not even set foot on the planet yet.”
We are pretty far from a dystopian future.
That being said, we are making some progress towards realizing it. And, figuring out how to put filters or restrictions in place now to make sure an AI doesn’t get carried away and wipe out humanity should absolutely be considered. We are already struggling with AI bias based on how they are trained and by whom, so there are still some barriers in place before we realize a fully functioning, autonomous AI of the future.
Researchers are already working on deep learning for computer vision (CV) which can help a robot and robotics applications understand their environments using various sensors, and help them navigate through obstacles in their path. This is called the mapping challenge, and machine learning and CV technologies can really help with this. This is going to be an essential part of autonomous, self-driven cars which do have to learn how to see in new conditions even when no training data is present.
A lot of research has also been happening around speech recognition which can be used to convert spoken language into machine language, and Natural Language Processing (NLP) which can provide a machine with the ability to understand and talk to a human. A very inferior version of such technologies is already utilized by some of us through our Alexa, Portal and Google home devices. In fact, GPT-3, a new language-generation model that has been in the news for creating interesting write-ups, is again powered by using deep learning in natural language processing.