Intelligence is something you can never really have too much of. Intelligent people bring an umbrella when rain is forecast, avoid dark alleyways at night-time, and, given the right training, they can land rockets on the moon. Through technology, can supplement our own intelligence with silicon-based ones, and use them to make better decisions.
If you’ve ever bought something on Amazon or watched something on Netflix because it appeared in your recommendations, then you’ve made a decision with the help of an artificial intelligence. And by acting on that recommendation, you’ve helped to improve the AI responsible for those decisions.
The human brain is the most complicated single structure in the known universe. We’ve barely begun to understand just how it works – and psychology departments across the world are forever arguing about the subject. So what chance do the designers of artificial intelligences stand of replicating it?
The simple answer is that they don’t. Instead, they automate the learning process, allowing machines to effectively educate other machines. Machine learning allows an AI to progressively improve at a certain task without the need for human programming. All that’s required is an idea of what success looks like, and the ability to randomly mutate the connections of a number of bits of code. Given enough generations, something like intelligence begins to emerge.
The earliest beneficiaries of artificial intelligence are likely to be the disabled. While a person who can see might be able to quickly assess the state of traffic on a given street before crossing it, a blind person cannot. An artificial intelligence might use a camera to assess the same street, and then provide an audio cue that it’s safe to cross. A similar artificial intelligence might be placed into driverless cars which travel along the same street. If the two intelligences were able to communicate with one another, then the chances of an accident could be greatly reduced.
Driving is an obvious example of a failure of human intelligence. When it comes to this task, none of us are really able-bodied, because human beings are not well-adapted for safe driving. The mechanical limitations of our limbs prevent us from shifting gears quickly, or braking suddenly. Our attention wanders, and our field of vision is narrow. We have adrenal glands that make dangerous behaviours appealing, and warped priorities that encourage us to take unnecessary risks.
Now, these decision-making flaws don’t just affect our driving ability. The same criticisms could be levelled at human stock traders, trauma surgeons and business leaders. With their access to enormous amounts of data, artificial intelligences could curb the worst impulses of human decision-makers. In time, they might replace those decision-makers altogether!
A machine which is more intelligent than a human is called a ‘superintelligent’ AI. Superintelligences are already with us. Pocket calculators can instantly perform arithmetic that a human would find impossible. But this intelligence is very narrow; a human user must input the numbers and press equals to generate the results. To answer maths questions in the same way as a human might, the calculator must be supplemented with language-processing capabilities, and speech-generation software through which to express the result. When narrow intelligences are interconnected, a more generally-competent intelligence emerges. That’s the endgame of AI.
Human beings are said to be generally intelligent. They can not only perform basic maths, but they can read intentions and body language, understand spoken language, form mental maps of a spaces. This combination of functions is what allows restaurant staff to intuit when a diner is likely to try to leave without paying, and which route they’ll take to the exit.
A ‘General Artificial Intelligence’ (or GAI) would be one that could reason, plan and learn, communicate and access knowledge. It would also be able to bring all these skills together to further a goal. To qualify as generally superintelligent, the machine would need to do all these things better than a human could.
A general superintelligence is, according to many silicon-valley experts, a potential existential threat to humanity. It would be as good as a human at everything, and capable of improving itself. Thus, within a short timespan it would be able to make itself much better than a human at everything, including improving itself. The result would be an exponential explosion of intelligence, whose motives we could no more understand than a cat might understand a game of tennis.
Artificial Intelligence is at a tipping point. Even the most conservative forecasters anticipate a radical changes in the way we live and work in the not-too-distant future. Provided that the transition is properly prepared for, these changes will be almost entirely for the better.