Fearing an omniscient AI is a trend. Countries are (allegedly) racing towards developing general AI, and whoever gets to the ring first - will rule them all!
I fear our future AI overlord about as much as a meteor strike. If/when it happens, it will have a potentially cataclysmic impact on civilization, however the chances of it happening in our lifetime are extremely low.
Understanding AI: Narrow vs General
AI is an imprecise term. Generally a layperson would take AI to be some sort of thinking machine with generalizable reasoning abilities similar to a human. We have nothing of the sort. We have narrow AI, or a term preferred by some, machine learning. Furthermore, the algorithms we use today are direct ancestors of the work done in the 1950's. The history of this field is one of incremental improvements, an AI winter, followed by a fortuitous combination of a data explosion and a discovery of GPU based processing of perceptrons.
Calling learning "deep" doesn't make it new or revolutionary in mathematical terms. Deep learning is nothing more than a fancier multilayer-perceptron (an algorithm first developed in the 1940's & 50's).
So, no - I don't think we are close to creating a silicon being capable of hijacking the internet, launching nukes, and taking all our jobs. Or that such a being is inevitable or even possible (especially without quantum effects). At the same time, I agree with people like Elon Musk and Bill Gates who argue that we should focus on developing regulations now (both within the industry & at government level) in order to address the potential risks. Just like I think that we should continue to invest in and enhance early monitoring for near-Earth objects. The latter is more fearsome for unlike the potential nightmare of general AI, catastrophic meteor strikes are an actual risk.
The Real Opportunity: Healthcare Transformation
Let me come back to healthcare. It may come as a shock to you and I hope that you are sitting down for this: we have major issues of access, quality, and affordability in healthcare. Machine learning is one of a few reasons to be optimistic that we can make a dent in these problems.
Diagnostic Insights and Image Classification
One of the biggest areas of current & near-term potential has to do with diagnostic insights from image classification. A recent scientific article in the journal Nature shows that an image classifier can detect breast cancer more accurately than trained radiologists. There is potential for better early screening for diabetic retinopathy using an AI-app on a smart phone that can prevent blindness. This can particularly help resource-poor areas of the world.
These are just a couple of examples from many. Stories like these cause click-bait headlines of the sort "AI to replace radiologists/ophthalmologists/etc.". I don't think that is the case. Rather, I think that AI or machine learning will aid them.
In general, I think that AI is not going to eliminate healthcare professionals but will increase the productivity, quality, and affordability of certain clinical processes. The throughput of the healthcare industry will increase dramatically, lowering costs.
Future Contributions of AI in Healthcare
Allow me to speculate on future contributions of AI in healthcare:
Drug Discovery: AI can help with drug-discovery (especially in combination with quantum computation).
Healthcare Administration: Within the reach of current technology is the ability to significantly simplify healthcare administration. Very few healthcare professionals are passionate about paperwork!
Personalized Medicine: AI or machine-learning will continue to be employed to help consumers understand, manage, and improve their health with highly personalized recommendations.
Medical Knowledge Integration: Another great application for AI is ingesting all historic and current medical research (only a tiny fraction of which can ever be consumed by a human), and developing best practices for treating old and emerging health issues.
Surgical Robotics: The Game Changer
I think a game-changing if not inevitable leap will be machine-learned machines, or robots enabled with AI. These machines will be capable of performing surgeries better, faster, and cheaper than humans. Think of the incredibly powerful hardware-software feedback loop that can propel such machines far beyond human capability.
This is already beginning to happen in dentistry, and one can grasp why. Far better vision coupled with far better dexterity in a "brain" plugged into all available experiential data. Of course such machines will be a better surgical tool. We will still need trained human surgeons to wield these tools, maintain, improve, develop new procedures, and to help ensure the outcomes that we want from these technologies.
Democratizing Healthcare
Machine-learning can have an incredibly positive and scalable impact on healthcare. Furthermore, it can be a democratizing force, bringing the latest medical advancements to parts of the world without huge investments in human training and research.
Final Thoughts
Fear, uncertainty and doubt (or 'FUD') is natural, but we must not let it derail legitimate scientific inquiry and progress. We should, however, channel our FUD to develop better and efficient regulatory frameworks to help us separate snake-oil from the medication that our healthcare system sorely needs.