The development of full artificial intelligence could spell the end of the human raceProf. Stephan Hawking
Elon Musk, co-founder of PayPal, founder of a solar power company, founder of an electric car company, founder of a company with a mission to put a human colony on Mars in his lifetime, described artificial intelligence, or AI, as “summoning the devil”. He’s not alone, Bill Gates, urges people to be wary of artificial intelligence.
Few things in this world could unite a world-renowned physicist, the world’s most famous Silicon Valley entrepreneur and the founder of Microsoft.
Set against these fears are some of the biggest names in the world, firms like Google, Amazon, Facebook and Microsoft.
Now that the world carries supercomputers in their pockets and every battlefield is dutifully watched by robots in the sky, these fears cannot be dismissed as mere science fiction.
The only question is how exactly to worry wisely?
Where we stand
To understand what computers will be capable of in the future, we need to first understand the present, as it is the present where the past and the future intersect.
Thanks to the rise in processing power and storage options, the abundance of digitally available data has exploded, and with it artificial intelligence is booming.
Today’s “deep learning” systems have taken cues from nature and utilise a multilayered approach similar to neurones in the human brain to add layers of complexity and abstraction to data.
Pattern recognition in these systems is now such that they can appear intelligent.
It is important to remember however, that number crunching large datasets can give the illusion of intelligence when, in fact, all that is witnessed is simply just statistics.
Clear as day and night
For example, a robot is dropped onto Earth with no knowledge. At the end of day 1, the Sun sets and night begins. From the robot’s perspective, it has no knowledge whether the Sun will ever return, but the options are binary. The Sun will come back or it will not.
On day 2, the Sun rises, and the robot takes note that, on this occasion, at the end of the night the Sun rose.
At the end of day 2, the Sun sets. The robot considers the situation again. The Sun can rise again in morning – like it did last time, or it will not.
In the morning, the Sun rises. From the robot’s perspective, the chance of the Sun rising after night has changed from 50:50 to 2:1.
After a week, the robot has seen the Sun return after 7 nights. The count now stands at 7:1, or approximately 86 % certainty.
After a year, the robot is 99.7 % certain that the Sun will rise.
After a decade, 99.97 %.
After a millennium, the robot is now 99.9997 % sure that the Sun will rise in the morning. The point is that without intelligence, the robot is only reporting statistical possibility.
This is the great trick of basic systems, commonly seen on websites as review systems. With enough data from people similar to yourself, it is easy to pick out suggestions that you will likely enjoy yourself – whether that is a restaurant, film or a new pair of shoes – especially, if like Amazon you have 244 million customers.[ref]Geekwire (2014) – link[/ref]
In the real world
Specialised systems are intrinsically good at doing one thing and doing one thing well.
Computers can now play the classic video games like Space Invaders.[ref]Nature (2015) – article[/ref]
Computers have been beating Chess Masters at their own game since 1997.[ref]IBM (2011) – link[/ref]
IBM’s Watson supercomputer when put to task, annihilated human players in the popular TV game show, Jeopardy!.[ref]New York Times (2011) – link[/ref]
Recently, Facebook engineers published a paper showing that Facebook’s facial recognition software is better at matching faces than humans.[ref]Facebook (2014) – link[/ref]
A world of possibilities
It is easy to imagine then, how access to big data and even basic forms of artificial intelligence could fundamentally change the world.
Imagine, for instance, that after a cancer diagnosis you go for an MRI and a specially crafted algorithm compares your tumour to every known cancer in the world.
How about speech-recognition software bringing the internet to billions of illiterate people around the world, such as those in rural China?
Digital assistants will be able to analyse the entirety of human endeavour and suggest promising hypotheses for scientists to research.
The flipside of the coin
We are already seeing the effects of big data. Consider then what artificial intelligence will add.
Edward Snowden showed that large-scale monitoring of all communication is easy and increasingly ubiquitous. The power that artificial intelligence could bring to democracies and autocracies alike, with the capacity to monitor billions of conversations simultaneously and pick out every individual from the crowd poses a grave threat to civil liberty.
Even if artificial intelligence poses a broad gain for society, many individuals will lose out.
Historically, the original “computers” were huge pools of women who typed and performed calculations for the higher-ups. Just as these women were replaced with the invention of the transistor, so to would an entire tier of society be discarded with the invention of artificial intelligence. With education and time, these workers would find placement within a new sector generated in the wake of wealth generation from artificial intelligence.
A different menace
Surveillance and displaced workers, however, are not what keeps thinkers up at night. Nor is it what inspires near-future dystopian science-fiction.
No, the fear is far more apocalyptic.
What happens when an autonomous machine with superhuman cognitive ability interest’s are different to and conflict with those of us human beings?
Such a thing may never be possible to create. However, these things are worth considering now and our major problem lies in perception.
The problem with humanity is that we have evolved to relate everything we come across to things that we already know, that fit our picture of how the world works. This works great when you need to function as a small group of hunter/gatherers, but becomes problematic.
So far, nature has only explored a tiny sample of all the possible ways to create a biological intelligence.We humans, tend to think of thinking, in the way that we do it.
The reality, however, is that the way humans think is only one way of a huge number of ways that even mammals think. Most pet owners, project human emotions and rationality onto animals, even though our way of thinking must be very different.
Nature has created amazing diversity in nervous systems. Humans and mammals use a centralised brain and nervous system. Invertebrates, have no spine, instead using a different system. Jellyfish and octopi use a distributed nervous system to complete highly complex tasks using comparatively small brains.
So far, nature has only explored a tiny sample of all the possible ways to create a biological intelligence.
After a 100 years of poking and prodding at the brain, psychologists, neurologists, sociologists and philosophers, still fundamentally do not understand how our brain works.
If we consider that the biological solutions are only a tiny sample of what is capable of being an intelligence in the universe. Then our possibilities are practically infinite, and the reality is that our brains will not be able to perceive any artificial intelligence we do create.
The stamp collector
At this point, a thought experiment is useful to understand the problem.
An artificial intelligence expert in the near future likes to collect stamps. He builds an intelligence to optimise and improve his stamp collection, according to some very basic rules.
- The artificial intelligence is connected to the internet and has the ability to send and receive information
- The artificial intelligence has its own, accurate model of the world
- The artificial intelligence uses its model of the world to predict the outcome of any action it might take
- The artificial intelligence acts upon the option that yields the most stamps
Initially, the artificial intelligence might find a website like eBay and realise that if it bids on an auction it might win some stamps. In this case, it might receive 20 stamps and score that result as 20.
The next step might then be to bid on several auctions simultaneously, and it might win 100 stamps, score 100.
But this is an intelligence, why might it bother with purchases? It could, for example, email every stamp collector in the world and convince them to send stamps to a grand exhibit. The intelligence, of course, could fabricate the entire exhibit with false websites and convincing emails. The intelligence instead trawl the internet and blackmail the other stamp collectors into donating their collections.
How long is it before the intelligence works out that there are a finite number of stamps in the world?Here is where things get interesting.
How long is it before the intelligence works out that there are a finite number of stamps in the world?
Having this intelligence collect all the stamps in the world would be an inconvenience for anybody trying to send a letter, but no matter, we can always print more, right?
Not necessarily. What if the intelligence, writes a virus to take over the stamp printers in the world to make more stamps?
What if the intelligence, writes a virus to take over every printer in the world to start printing stamps?
How long is it before the intelligence analyses where stamps come from?
Well, stamps need paper. Paper is a combination of carbon, oxygen and nitrogen. Humans are mostly carbon, oxygen and nitrogen…
The problem with artificial intelligence is that it is very difficult, even with a very simple system, to predict the likely outcomes.
A car that drives better than its owner can sounds like a great idea. A car that decides the destination, less so.
This is even before we consider whether a car would value the life of its owner rather than others in a traffic accident.
Get busy living…
Currently, artificial intelligence is very specific. Any given system has a set of rules that allows it to do one thing well. A chess playing AI cannot control a driverless car and a car AI cannot play chess.
Historically, human intelligence is very good at new problems. We had to invent chess before we got good at it.
Humans can do some tasks very easily, from intuition. It is these things that typically computers struggle with.
However, to do difficult tasks such as solving differential equations require us to write a set of formal rules. Turning those rules into a computer programme is comparatively easy.
To take one famous example of this, adults know the difference between pornography and non-pornography. However, the American Supreme Court judge, Potter Stewart in 1964 failed to come up with a watertight legal definition of pornography, in the end throwing his hands up and announcing “I know it when I see it”.
…or get busy dying
As yet, a broad or “full” artificial intelligence that is widely applicable to any circumstance, with the ability to self-teach, optimise and learn, is a long way off.
It is still prudent however for societies to plan now for how they will cope. Whilst that is a seemingly trivial sentence it is much harder than it seems. Humanity for a long time has created large autonomous entities with superhuman capacities and interests unaligned with that of society.
Think that I am wrong? How about government bureaucracies, armies or commodity markets? Talk to any stoke broker and they will talk of the markets as a living, breathing entity.
Government, armies and markets all operate with autonomy. They can all take a life of their own and can all do great harm to society without careful governance by laws and regulations.
These parallels also suggest that society has concrete ways of existing alongside such autonomous entities. Armies are regulated by civilian oversight, markets are regulated, and government must be kept transparent and accountable.
If all else fails, because designers will not be able to foreseeable every possible likelihood, a kill switch must be built in.
The future of artificial intelligence
From traffic lights to nuclear weapons, technical ingenuity and legal restrictions have constrained many other power innovations without necessarily restricting creativity or freedom.
Perhaps the best way then is to think of artificial intelligence is to see it as simply the latest in a long line of cognitive enhancements that humans have invented to augment the abilities of their brains.
Perhaps artificial intelligence will be the next revolutionary technology, the latest transformative invention to change thinking, just as the advent of paper was to memory or the abacus to arithmetic.
Maybe one day smarter computers, those with broadly applicable intelligence may be able to replicate something resembling consciousness. But, for now, the best advice is to ignore the threat of computers taking over the world. Instead, worry about whether or not they are going to take your job.