In November, an organisation to stop autonomous weapons . In it, a fictional spokesman for small, weaponised drones reaches out to grab one that鈥檚 hovering nearby. The drone easily evades him. 鈥淗ell of a pilot?鈥� asks the man. 鈥淣o, that skill is all AI. It鈥檚 flying itself.鈥� He extends his palm and the drone gracefully lands on it.
The drone鈥檚 toolset includes cameras, sensors, face recognition and three grams of shaped explosive. The man throws the drone into the air and it zeroes in on a mannequin and blows a hole in its head. The video then shows how a swarm of such drones can punch holes through windows and use artificial intelligence to target one 鈥渂ad guy鈥� in a crowd of people. What can possibly go wrong?
Advertisement
We find out soon enough. In one scenario, the drones begin targeting university students who have been exchanging videos on social media of corruption in high places. The speculative short film 鈥渟hows the results of integrating and miniaturising technologies that we already have鈥�, warns AI expert Stuart Russell of the University of California, Berkeley. 鈥淸AI鈥檚] potential to benefit humanity is enormous, even in defence. But allowing machines to choose to kill humans will be devastating to our security and freedom.鈥�
Far-fetched? Some of the ingredients are already available. , Raffaello D鈥橝ndrea of ETH Zurich demonstrated 鈥渄azzling flying machines of the future鈥�, mesmerising his audience with a swarm of tiny quadcopters that buzzed around like fireflies. And Facebook announced last month that it鈥檚 going to use AI to monitor social media posts, identifying people鈥檚 suicidal thoughts and getting them early help. 鈥淲ith all the fear about how AI may be harmful in the future, it鈥檚 good to remind ourselves how AI is actually helping save people鈥檚 lives today,鈥� wrote Mark Zuckerberg .
Deep learning
AI, machine learning, deep learning 鈥� whatever we call it 鈥� the technology is becoming ubiquitous, most obviously in Google search, Amazon鈥檚 Alexa and Apple鈥檚 Face ID.
This year also saw a slew of studies targeting neurological conditions. Machine learning algorithms analysed brain scans of infants with a family history of autism to predict whether these babies might show signs of autism at 2 years of age.
Similarly, AI can use brain scans to identify people who may go on to develop Alzheimer鈥檚, almost a decade before clinical symptoms emerge. Similar work is going on to help advanced diagnosis of other neurodegenerative diseases such as Parkinson鈥檚.
These advances raise ethical concerns. How will such information be used? Might such individuals be denied healthcare? Would you even want to know 10 to 15 years in advance that you will get Alzheimer鈥檚, given that nothing much can be done yet to halt the disease?
Even if you don鈥檛 want to know, those caring for you might. To that end, Andrew Ng of Stanford University and colleagues that can be trained using patient records to predict whether an individual patient is nearing the end of their life and will need palliative care鈥攁 decision that physicians can get wrong because they often overestimate how long the patient will live or simply because there aren鈥檛 enough palliative care physicians to do the diagnosis.
Hidden in such news are subtle problems. For example, algorithms that learn from human-generated data can pick up our biases. In 2016, Microsoft鈥檚 experiment in AI, a Twitter bot named Tay that was learning to tweet by feeding on social media data, failed spectacularly when it tweeted 鈥淗itler was right I hate the jews鈥�. In April this year, that machine learning algorithms could imbibe our racist and gender biases.
AI that can explain itself
There鈥檚 growing concern that we won鈥檛 understand the workings of these algorithms as they grow more sophisticated. In May 2018, the European Union General Data Protection Regulation will go into effect: EU residents will have the right to 鈥渕eaningful information about the logic involved鈥� in making automated decisions. Given such laws, an AI that can explain itself is going to become increasingly important.
The loss of jobs is another growing concern. AI-enabled automation (think self-driving cars and trucks) will usurp many blue-collar jobs. But they are not the only ones. Doctors, journalists, data analysts and many others are in AI鈥檚 sights too. Geoffrey Hinton, one of the world鈥檚 foremost experts on deep learning, told that 鈥淚 think that if you work as a radiologist you are like Wile E. Coyote in the cartoon. You鈥檙e already over the edge of the cliff, but you haven鈥檛 yet looked down. There鈥檚 no ground underneath.鈥�
To those in the know, the advance of AI is a foregone conclusion, as theoretical physicist Max Tegmark of the Massachusetts Institute of Technology pointed out with neuroscientist and philosopher Sam Harris. 鈥淭here is nothing special about human level intelligence,鈥� said Tegmark. 鈥淚t鈥檚 how smart you have to be before you are able to design AI systems. Once machines get there, they can start designing themselves鈥� If you think of progress of intelligence as this freight train moving along, once it comes to station human, it鈥檚 going to blow right through and keep going.鈥�
Powerful and pervasive AI will happen, and it will reflect who we are: compassionate, caring, conflicted, bigoted, self-absorbed, narcissistic and sometimes downright evil. It鈥檚 our choice as to whether the machines we invent embody our faults or our virtues.
