Is AI winter Coming?
Are neural networks and deep learning useful? Absolutely!
And they are definitely worth learning. That said, there is a good chance you have seen media, politicians, and tech celebrities hail deep learning as some general artificial intelligence that can match, if not outperform, human intelligence and is even destined to control the community telling programmers they will be out of a job in few years because of machine learning and AI will take over writing code.
Nearly a decade later, these predictions are still running late in 2022 because they are simply not true. We have seen far more AI challenges than breakthroughs, and I still have to drive my own car and write my own code. Neural networks are loosely inspired by the human brain but are by no means a replication of them. Their capabilities are nowhere near on par with what you see in movies like 'The Terminator', 'Westworld', or 'WarGames'. Instead, neural networks and deep learning work narrowly on specific problems, like recognising dog and cat photos after being optimised on thousands on images. As stated earlier, they can not reason or choose their own tasks, or contemplate uncertainty or objects they have not seen before. Neural networks and deep learning do only what they were programmed to do.
This disconnect may have inflated investment and expectations, resulting in a bubble that could burst. This would brig about another "AI winter", where disillusionment and disappointment dry up funding in AI research. In North America, Europe, and Japan, AI winters have happened multiple times since 2960s. There's a good chance another AI winter is around the corner, but it does not mean neural networks and deep learning will lose usefulness. They will continue to be applied in the problems they are good at: computer vision, audio, natural language, and a few other domains. Maybe you can discover a new ways to use them! Use what works best whether it's linear regression, logistic regression, a traditional rule-based systems, or a neural network. There is much more power in the simplicity of pairing the right tool to the right problem.