There have been many discussions in the blogosphere predicting the doom and gloom of the inevitable rise of Artificial Intelligence (AI). I’ve seen theories on everything from how it will result in massive layoffs to the ultimate downfall of humanity itself.
Some prominent names – such as physicist Stephen Hawking, Bill Gates and Elon Musk, to name a few – have put forth ominous warnings of the effects AI could have on mankind. Let’s take a break from this kind of hype and take a hard, realistic look exactly what AI can bring to the table instead.
In the “ABC’s of AI,” I discuss how AI is made up of supervised and unsupervised machine learning. I outline how supervised learning is the foundation for predictive analytics and that it simply consists of giving the computer known answers up front through training sets. The math being used to provide this intelligence has been taught in undergraduate classes since the ’50s!
There’s been a lot discussion about how IBM Watson won out in a major showdown against some of the best Jeopardy players in the world. The current capabilities of AI today are to provide an already known answer to a question. This is exactly what Watson did in the Jeopardy showdown, captured all the answers to all the possible questions that Jeopardy could cover. No small feat in and of itself.
What’s missing, and what would give AI credibility, is for it to be used within the intelligence community: the CIA, NSA, Homeland Security, FBI, etc. I would expect to see some headline in the blogosphere along the lines of “Secret Agent Watson 001 cracks ISIS terrorist ring.”
You could argue two points. One: this type of capability is top secret, and you don’t have “need-to-know” clearance for the purpose of national security. Two: the current capabilities of AI just aren’t there yet.
I suspect that there are truths in both points. There probably are very specific AI solutions in the intelligence community. Like the IBM computer that had 480 special purpose VLSI chess chips that defeated Garry Kasparov in 1997 chess tournament.
On the other hand, however, general purpose AI systems doesn’t possess the capability of reasoning. What do I mean by reasoning? Take the nursery rhyme “Cella Ree and Tommy To.” Tom is red and Cella pale, His blushes are of no avail. Who is blushing? You and I instinctively know that it’s Tom. This is a simple classification problem of determining gender. Some vendors will argue that their products, using statistical probability, could resolve this type of problem. There is no doubt in this example, when you’re simply talking about a 50/50 chance of calculating the right answer. This nursery rhyme also presents another challenge concerning the relationship between red and blush. This is a different type of classification that is associated with synonyms.
However, trying to understand a news headline presents a more challenging solution than classification. As an example, take the fictitious headline “White House Approves Signing Treaty with Russia.” An AI system needs to understand the context of the sentence. The White House is a place, not a person. Places do not approve of making treaties. Also, which White House; the one in Washington D.C. or the white house that happens to be the primary office of the government of Russia?
Given that you do have an AI system that can understand the White House is really an inference to the President of the United States, now you need to answer which President of the United States. So, to have an AI system that can reason, you need to have a system that can reason entity identification within spatial and temporal dimensions. Not exactly a trivial task that is being handled with any great certainty in the market place today.
So, this notion that is floating around in blogs where AI is going to be the silver bullet that will save the U.S. healthcare industry is far from being reality. The argument that if we just let AI have access to everyone’s medical records then we can reduce the overall healthcare costs. Not to discount that AI can help in many areas of healthcare, but to suggest that incorporating AI into our healthcare system is going to cure our nation’s healthcare woes is not looking at the healthcare problem in full context.
Time magazine published an article back in April of 2013 called “Bitter Pill: Why Medical Bills Are Killing Us” (subscription required). The article brought to light the illness that is slowly killing our healthcare industry is not the lack of knowledge of how to diagnose and treat patients, but not having transparency at all levels of the healthcare system.
So, if you’re looking for AI as the silver bullet to save the healthcare industry, you’ll be regretfully disappointed! To truly fix our nation’s healthcare woes will require that you and I and every working American who is reliant on the healthcare system needs to voice our concerns to Congress. We need a healthcare system that is fully transparent and can meet the open market requirements of supply and demand.
As for AI causing high unemployment and the downfall of mankind, yes, this may come to fruition sometime in the unforeseeable future. In the meantime, the real threat to mankind is not AI, but the Internet itself. You don’t have to go any further than Sony to understand that we are one firewall away from going to World War III.