Is a Go-playing algorithm that beats human champions a good example of an artificial intelligence application?
How about a racist Twitter bot?
It's been a rollercoaster ride for artificial intelligence lately. Google's AlphaGo program beat human world champions at the game Go, which has been described as infinitely more complicated than chess. But whatever AI-related excitement that project generated came crashing down when Microsoft's Twitter chat bot Tay flamed out in a stream of racist and sexist tweets.
So what does this mean for the state of AI today? Nothing.
AI capabilities don't come close to human mind
Both are good examples of machine learning. Microsoft added a natural language processing layer, arguably pushing Tay into cognitive computing territory. But as far as true AI goes, neither comes close.
In 1956 Herbert Simon, a Carnegie Mellon researcher who's considered to be one of the founding fathers of computer AI systems, said, "Machines will be capable, within 20 years, of doing any work a man can do." Leaving aside the wildly inaccurate timescale of his prediction, the quote serves as a good guide to what AI has always been about -- replacing human brain power with machine brain power.
But looking at all the examples of things called AI today there's still a huge gap between their capabilities and those of a human mind. Today, what we often see called AI are algorithms optimized to do one specific task. What separates human intelligence is the ability to engage with many different tasks. All these systems lack the ability to understand circumstances in a broad sense.
This isn't just quibbling about semantics. We've seen this show before. When big data became a broadly acknowledged term about five years ago we saw a huge rush of marketers trying to call their software big data systems. Today, any piece of software that can manage or analyze data is referred to as a big data tool. The term is muddied almost to the point of uselessness.
AI today nothing more than machine learning
The same thing is happening with artificial intelligence applications. I receive at least one pitch per week from someone touting cognitive computing or AI software. In most cases, these products seem to involve little more than machine learning.
Of course, the ability to learn is a necessary component of intelligence, as is the ability to communicate through natural language. And machine learning can be a great tool for businesses. Look at the success Amazon has had with its recommendation engines, which are built around machine learning algorithms.
But to call that an example of AI overstates the case and detracts from the conversations businesses should be having about machine learning and intelligent systems. Mention AI and the first thing most people think of is some fictional future dystopia like Terminator or The Matrix in which intelligent machines try to wipe out humanity. This is quite obviously irrelevant to any discussion of business value.
Focus on business value of machine learning
Instead of worrying about this, insurance companies should be thinking about how machine learning algorithms can help them identify cases of fraud more quickly. Call centers should be looking at ways natural language processing engines can service greater numbers of customers more quickly than traditional phone banks. Leave debates about AI to the academics and fiction writers.
And most of all don't worry about what AlphaGo and Tay say about the state of artificial intelligence applications today. They aren't true AI and neither are most of the products you are currently being pitched.
Ready or not, here comes artificial intelligence
Big opportunities seen for computer AI in hospital ERs
AI could play important role in future of software testing