Every now and then technology advances in leaps and bounds in ways that surprise us all. Artificial Intelligence (AI) has long been seen of as either limited use in our everyday lives, or as some far off sci-fi vision of the future. Yet happening almost stealthily among our midst there is a revolution going on in machine intelligence that’s poised to change our world as we know it. Here we’ll gain a glimpse into a new dawn of AI super-minds, and why it’s being driven by the neuroscience to do with the way our very own brains work.

The Traditional AI Paradigm

The conventional approach to computing has basically remained the same ever since Alan Turing first developed machines to help crack the Enigma Code in World War II. This involves writing a computing script or set of behavioral rules, known as an algorithm, then serially crunching one calculation at a time. Although computing power has increased exponentially, following the infamous Moore’s Law, the methodology behind computing has for the most stayed unchanged. The main difference these days is that computers are vastly faster at crunching data points due to superior hardware. Take a modern smartphone for instance, its processor compresses literally billions of transistors into a tiny chip.

Brawn Over Brains

From an AI perspective this has fueled an increase in what’s known as brute force computation – as long as a programmer writes the right kind of algorithms, computers can tackle large problems just by the sheer speed and amount of calculations they can perform.  Most famously this led to the defeat of long running world chess champion Gary Kasporov by IBM’s Deep Blue.  Though such feats are impressive, there has long been skepticism of how useful this type of machine intelligence is. Dubbed Narrow or Weak AI for a reason, it’s generally only useful for tackling very specific problems that basically don’t translate to the complexity of the real world. This leaves little or no hope for emulating the type of creative intelligence possessed by human consciousness.

Another example is Google’s Deep Mind project which created Alpha Go, the AI that was developed to take on the fiendishly complex game of ‘Go’. In this game brute force techniques don’t work well, whereas humans excel through use of intuition. Although Alpha Go did manage to defeat world champion Lee Sedol, it did so by being fed massive amounts of games from elite Go players, copying and combining their strategies, and then executing moves without errors. Yes it was successful, but on the grander scheme of AI progress, Alpha Go is essentially limited by the knowledge that humans have figured out, with little prospect of going any further.

The New AI Paradigm

Though few people are aware, AI has been going through a revolution in recent years by taking a completely new and innovative approach to computing that actually emulates the way our brains solve problems. Rather than taking a rule based algorithmic approach, a novel method called ‘deep learning’ has taken a giant leap in evolution to create a new form of general AI that literally does not need to be told what to do. Instead it starts out pretty much like a new born baby, and from a blank slate it tackles problems by learning about its world through experiment after experiment. Then at each step it creates its own inherently new behaviors based on what it finds to be the best solution.

This gave genesis to Alpha Go Zero - zero meaning starting from nothing but the simple rules of the game. This seemingly innocuously change in name represents an AI that has redefined what computers are capable of.

Independent Learning

Alpha Go Zero starting playing Go against itself, experimenting with what worked and didn’t work, refining, and then playing again.  In just 3 days, and in stunning fashion, it used what it had learnt to defeat the version of Alpha Go that defeated Lee Sedol. However it didn’t stop there, and went on to beat the most evolved version of Alpha Go (Master), winning 100 games 0. What’s truly impressive, is that it wasn’t built specifically to play Go – it just seemed to like it.

So then it was given chess to play with. In just 4 hours of self-practice it became good enough to conquer the current AI world chess champion.

It did so in ways that dumbfounded human chess experts. This is because it created new strategies no one had ever seen the likes of. This included concatenations of novel tactics like sacrificing a queen to gain a positional advantage, and attacking with its king piece. Experts called it ‘alien chess’, or ‘crazy attacking chess’. Alpha Go Zero’s freshly discovered style of play changed the way humans actually perceive the game itself.

Deep Neural Networks

So how is this kind of creative and self-learning intelligence made and how does it relate to human brains? Well, it’s really about qualitative over quantitative calculations.  The human mind is what’s known as a complex system, from which intelligence and consciousness emerges from the collective interactions of billions of neurons talking to each other. Efforts to understand how it truly works involve Complexity Theory or Systems Theory. This is ultimately about the idea the whole is more than the sum of the parts. For example a single neuron has zero intelligence, so the classical reductionist approach to scientific progress doesn’t really cut it when it comes to how the brain works overall.

Humans, for the most part, are not built with a predefined set of rules how to behave. Instead we experience the world, learn, and then adapt. This is done primarily through the neo cortex, which uses non-linear, non-algorithmic processing to figure out solutions for optimal behaviors. These new discoveries can then even be coded become automatic behaviors, performed without actually thinking - imagine someone popping a balloon next to you.

The new revolution in AI takes an uncannily similar approach, where learning emerges through Deep Neural Networks, operating very much in the same way our neo cortex works. Rather than serially processing information one data point at a time, calculations are performing in parallel and through almost organics interactions. This method uses a lot less computational resources than traditional AIs, yet achieves much broader levels of intelligence. Most importantly, there’s no programming work once created - it’s simply a case of presenting the AI problems to solve.

Strangely, and much like the brain, how deep learning actually happens at a fundamental level, is still a bit of a mystery.

Beyond Board Games

Fascinating though these developments are, the ultimate question is will this new form of AI tackle real world problems? After all there’s not a lot of practical use in computers that just play board games all the time.

The answer is yes. Google’s self-driving cars and speech learning engines are cursory examples of applications that are being developed today, but expect this to be the tip of the ice berg. Corporate giants such as Google, Amazon, and Facebook are all investing huge resources into developing deep learning AIs as a core feature at the heart of their businesses. There is also the tantalizing prospect of a breakthrough in quantum computing, which holds promise for a gargantuan rise in computing power.

On the flip side, the new momentum in AI is sparking a fresh and serious level of concern that AI might not only replacing us, but is potentially becoming an existential threat to humanity. Even the likes of Elon Musk and the late but great Stephen Hawking have given public warnings on how very real such a threat is.

As we saw with Alpha Go Zero, the results of this next level evolution in machine intelligence will probably surprise us, but one thing is for sure - AI super-minds are coming and they will change life as we know it.

If you’re curiosity has been piqued on the neuroscience of the brain, then why not check out these blogs?

5 Reasons Why Neuroscience is Amazing

Your Brain’s Remarkable Neuroplasticity

‍