Will computer reach human intelligence in the near future?

Answer 1

This is a difficult question to answer because it can be hard for those in the industry to separate fact from opinion.

The facts are that machine learning and deep learning are amazing tools and capable of AI at levels that have not been previously thought possible. With all facets of technology developed by humans, we aim as a whole to continue to develop and make these better as time goes on.

With this in mind, we inevitably will get more and more sophisticated forms of AI as time goes on - however it took a very long time to get to even where we are now. Machine learning isn't new, in fact theory on the subject was being written 50+ years ago, with the term being coined in 1959.

It has taken from then until now to develop out these tools into the usable state we have them currently, but of course not without their flaws.

Machine learning requires a deep understanding of mathematical modeling and neural networking algorithms which are all not easy subjects to become proficient in by any means. Furthermore, machine learning and deep learning is not an exact science just yet. (One could even argue that computer programming as a whole isn't an exact science as of yet either).

Machine and Deep learning require large amounts of data in order to run simulations using this data to "learn" a behavior, object/image/audio/video classification, etc... With companies like Google, Microsoft, Amazon, Facebook and Apple having access to so much of our personal data it can be jarring to think about what sorts of AI's they could be training for their own needs.

The process of running these simulations on data is called "training." Training models in machine learning is a difficult thing to do. If your data or classifications are wrong or poorly structured then your resulting trained neural network will also be flawed.

I'll end with this story: A group of researchers were developing a neural network to tell the difference from a wolf and a domesticated dog. It was able to attain a 90% + success rate in doing so. However, the team noticed that all pictures of dogs in a white background were classified as wolves and not dogs. The team found this odd and tested further, discovering that the model they had training was not looking for the difference in physical features of the animals in the photos, instead it was recognizing it could achieve a 90% + success rate by simply classifying any animal with snow in the background as a wolf and any animal without snow in the background as a dog. This, of course, was not what they were looking for and needed to re-train their neural network with better and more diversified data.

In conclusion, while it is always scary to think about what humanity would do with simulated human intelligence we are still leaps and bounds away from that level of simulated "thought". What the future holds is unclear and how quickly we will advance is also hard to say. There's a good chance we're another 20+ years away from even coming close, since this technology is primarily being used for drawing sales insights and optimizing resources.

I hope this helped!