Whether or not you’re a fan of his novels, you can’t deny Dan Brown has his finger on the pulse of the modern zeitgeist. His latest blockbuster, Origin, envisions a world where artificial intelligence (AI) has reached a level almost indiscernible from humans. Of course, Brown’s vision is a work of fiction, but just how likely is it the supercomputer he depicts in his book will ever become a reality?
Brown’s conceptualisation of the future is just part of the hype that surrounds AI – the subject everyone is talking about. Gartner predicts it will be one of the top 10 technology trends of 2018 and 85% of UK businesses are planning on investing in AI by 2020. The UK Government also set aside £75 million worth of funding for AI in its autumn budget.
Today’s AI systems are still a long way from truly mimicking human behaviour and capabilities. If science is to reach the point where it can create technology akin to Brown’s super-computer – what is often referred to as artificial general intelligence (AGI) – there are several challenges that need to be solved.
First, there is the data required to match the genetic code of a human – 1.5GB to be precise. But this doesn’t include the technology needed to build the physical body of a human, nor the memory required to add intelligence or brainpower. The human brain has a vast capacity – most computational neuroscientists estimate human storage capacity somewhere between 10TB (terabytes) and 100TB, though the full spectrum of suppositions ranges from 1TB to 2.5PB (petabytes).
Humans learn through experiences and it takes 21 years for the brain to fully develop into adulthood so it stands to reason the data sets required to not only build a synth but to train it on the fly are almost unimaginably vast.
While advances in cloud computing and emerging analytics platforms mean it is now possible to manage these enormous datasets, there are still other issues to overcome before AI can replicate human thought.
When Google’s DeepMind AlphaGo AI was trained to beat humans at the ancient and complex Chinese game of Go, it was given the rules of the game and then left to simulate millions and millions of games in order to calculate the best strategies. But after all those simulations the system could still only perform one task – albeit exceptionally well.
To achieve AGI, systems will need to perform millions of different tasks in the same way as a human brain. At present only small pieces of the puzzle that make up human beings can be replicated – such as object recognition, simple construction, and logic – rather than the functionality and the components to piece it all together. No true AI system has ever managed to pass the Turing Test, showing just how difficult it is for a machine to mimic a human personality.
AI-based systems currently used in customer service, such as RBS’ Luvo, are able to access vast swathes of information from CRM systems and answer simple questions, but still require human help when it comes to more complex queries.
There is simply not enough analysed data to train AI-based systems to think and behave like a human being, nor the capacity and processing power to analyse that data and respond to it in real time.
An alternative to replicating humans with AI would be to combine the two and attempts are already being made to mesh technology into human bodies, such as Elon Musk’s Neuralink.
It is hoped technology can be used to boost memory and intelligence, or enable humans to control objects with the power of thought alone, which isn’t as futuristic as it might seem as tech is already being installed in human brains to treat conditions such as Parkinson’s.
But medical intervention isn’t the only way human intelligence can be combined with AI. Predictive intelligence, highly receptive sensors, and fast processing mean machines can read humans without physically becoming part of them and this is humanity’s best way ahead with AI – for now at least. A great example of this is Cogito, which has developed software to monitor call centre conversations and suggest solutions to representatives based on its analysis, which even includes factors such as tone of voice.
Using AI for some tasks while humans oversee the machines and retain control is the best way to improve efficiency, and this is already possible using emerging technologies that can process extreme data, from multiple sources, in real time. And the possibilities are endless with synths just one end of the spectrum. Smart cities are also starting to use AI, as are the adtech and martech industries.
It remains to be seen whether Dan Brown’s vision that AI will become indistinguishable from humans could ever come true – and what the ethical implications might be on jobs, relationships, and the law.
In the meantime, governments and businesses need to look at ways of equipping humans with AI so they can become more efficient, rather than turning the machine into the human. The AI in Origin won’t be part of our lives anytime soon, but give it 50 years and who knows what could happen.
Sourced by Steve Marsh PhD, founder & CTO of GeoSpock