This is the first article in the series 'Artificial Intelligence & The End Of Human Race'. This series will explore artificial intelligence and their impact on human life. We will gather evidence and stories that point towards a plausible future of artificial intelligence. We will try and answer one of the most controversial questions surrounding AI development - could artificial intelligence actually spell the end of the human race? Follow the series here.
Part 1 - How Can Machines Be Intelligent?
Computers are constantly getting better, faster, and stronger, and they’re really great at things we’re terrible at. For example, tell me what’s 187673 times 6835? Right now. Go! Can't? I thought so. An ordinary computer will be able to compute this in less than a second and yet we still consider them dumb. If being amazing at raw calculations doesn’t make them smart, at what point is a machine considered intelligent?
The most famous test of artificial intelligence is the Turing Test. Basically, a computer passes the Turing Test if a computer can chat with a human for 5 minutes and fool the person into thinking it’s human at least 30% of the time. Aside from one controversial effort in 2014 where a chatbot snuck past the judges by lowering expectations, the Turing Test remains unbeaten. That could very well be because not many AI developers are trying to beat it.
The Turing Test isn’t broad enough in scope to be the be-all end-all benchmark for AI. For example, he Turing test doesn’t test if a machine can make a judgment about an image, or use common-sense and reasoning to solve common problems.
So more challenges have been set forth, and little by little machines are starting to pass them. Researchers think an intelligent machine should be able to interpret visual information, like when something is a face. Lo and behold, there are several phone apps that know what shapes to search for when you are taking selfies so it can slap dog ears on your face.
Computers are also beating us at our own games lately. Google’s AlphaGo is an algorithm that plays the ancient Chinese game of Go like a boss. It was good enough to beat 18-time world Go champion Lee Sedol in four out of five games played.
Playing Go like a pro and recognizing a face is impressive, but a machine that does one specific thing at a human level can’t really be called intelligent. We call that Artificial Narrow Intelligence. In order to have a truly intelligent machine, we need it to have Artificial General Intelligence. To meet that standard, a machine would have to do many things at our level, a human's level.
The human standard for intelligence is to be like a human. It sounds egotistical but it’s not.
Anyone who is interested in artificial intelligence and/or human mind should definitely read How the Mind Works by Steven Pinker. It is a 1997 cognitive science classic. The book explores our mind's functions and compares it to a computer. It explains our minds and machines in simplified evolutionary terms and contemporary anecdotes. It will guide you on why an intelligent machine needs to have a mind as good as humans.
The human mind is formidable. One is often perplexed by our (humans') beautiful minds. We can learn about so many completely unrelated things, and we can do it even if we’re not explicitly taught. A single human is capable of playing Go and recognizing faces and having an open-ended conversation. An extraordinary feat to be achieved by a single computer so far.
What makes us humans so intelligent is the fact that we can teach ourselves to do the things we want. But is the same true for computers? Yes, moderately. That’s where machine learning steps in. Today computers can teach themselves. Given enough data, modern algorithms can start to write their own code, refining it over and over.
The most basic facial recognition code, that an algorithm writes, might look for simple things like edges. The next level adds the ability to combine those edges into shapes like circles and rectangles. Go a layer deeper and it can start to distinguish eyes and noses. These algorithms need to look at hundreds of thousands to millions of examples of faces. Eventually, their codes will be so good at looking for facial features that they’ll be able to tell when two pictures are of the same person; even if the lighting has changed or the angle is different or even if they’ve grown a beard. It’s why Facebook knows your friends are in a picture before you even tag them.
A smart machine will not just recognize specific people but also be able to generalize. Even if a face looks like nothing they’ve ever seen before, they’ll still recognize humans from other species.
Eventually, if a machine can teach itself many different human traits and has the raw horsepower to do it quickly, it’ll be considered artificially intelligent.
But I doubt it’ll stop there. It won’t be long before they surpass human intelligence and obtain artificial super-intelligence. When that day comes, maybe it’ll be the machines that test humans to see if we can fool them. That would be such a dangerous role reversal!
Remember what Professor Stephen Hawking said -
The development of full artificial intelligence could spell the end of the human race.
I'll let this story of Artificial Intelligence continue! Stay tuned.