Deep learning. A self-learning mechanized method by which an AI machine discovers the rules by which queries can be answered in a given context, i.e., it functions as a researcher. “Deep-learning software attempts to mimic the activity in layers of neurons in the neocortex, the wrinkly 80 percent of the brain where thinking occurs. The software learns, in a very real sense, to recognize patterns in digital representations of sounds, images, and other data.”85
6.2 Deep data and deep learning: the good, bad, and ugly
Arguably, when IBM’s Deep Blue computer defeated chess champion Garry Kasparov in game one of a six-game match on 10 February 1996, a threshold in AI was crossed. While Kasparov won the 6-game match on this occasion, he lost to IBM’s Deep Blue supercomputer on 12 May 1997 in a 6-game rematch. Since then AI machines have been beating human champions in games left, right, and center: They beat human champions in Jeopardy (February 2011), the Chinese game Go (March 2016), Poker (January 2017), once again in Go by AlphaGo Zero (October 2017; it learnt on its own from a blank slate), again in chess (December 2017; the machine taught itself in four hours), etc. Of these, the most significant is AlphaGo Zero which learnt purely by playing against itself millions of times over. It began by placing stones on the Go board at random but swiftly improved as it discovered winning strategies. It is a big step towards building versatile learning algorithms.
The technologies related to deep data and deep learning have tremendous scope for good and bad use. For example, it is a boon for diagnosis and treatment of patients, but much science still needs to be done.86 There are related non-medical issues too: “What’s the best way to win the confidence of public and regulators?” “Is academia training enough mathematicians and medical-data scientists, to harness the potential embedded in the data?” “Genomic data sets alone have already shown their value. The presence or absence of a particular gene variant can put people in high- or low-risk groups for various diseases and identify in some cases which people with cancer are likely to respond to certain drugs.”87 But there are other equally voluminous diverse sources of data which must be integrated in the context of a patient’s physiology, behavior and health. And all this data must be held in the strict, non-negotiable, privacy regulations of medical data.88 Big data offers the opportunity to allow clinical trials to be conducted partly in silico.
Invasion of privacy is a rapidly increasing concern. Deep learning has the capability to do deep profiling of humans, situations, events, buying habits, etc. and integrate them into a larger picture and invade personal privacy. An innocuous example is cited in an editorial in Nature89. The mere purchasing of scientific books can escalate into learning much more about an individual, his network, lifestyle, political leanings, and what not. Beware of doing anything that gets recorded in a database, e.g., using your mobile. You are being watched, profiled, and analyzed whether you like it or not. Internet of Things, automation, cognitive computing (e.g., face recognition) have already advanced well beyond the average human’s capacity. Facial-recognition technology can measure key aspects of a face, e.g., skin tone, and cross-reference them against huge databases of photographs collected by government agencies and businesses and shared on social media. The technology obviously provides extremely powerful tools for surveillance. Such technologies allow governments to watch your every move.
Massive, malicious gathering and use of personal data by authoritarian governments is no longer a possibility but a reality.90 By 2020, China hopes to implement a national “social credit” system that would assign every citizen a rating based on how they behave at work, in public venues and in their financial dealings. Its power to monitor people would be awesome. Such data collection can always be justified, as in this case, in the name of scientific decision-making that promotes poverty alleviation, better management, and social stability and it can be easily done by offering free health care program, especially to the poor and the displaced, who are often seen as potential law-and-order risk.
The Economist expressed deep concern about China:
In parts of the province [western region of Xinjiang] streets have poles bristling with CCTV cameras every 100-200 meters. They record each passing driver’s face and the car’s number plate. Uighurs’ mobile phones must run government-issued spyware. The data associated with their ID cards include not just name, sex and occupation, but can contain relatives’ details, fingerprints, blood type, DNA information, detention record and reliability status.91
While surveillance on this scale may be limited to pockets, the fact is repressing minorities by such means is easy in autocracies. especially in countries like China where Xi Jinping wants a one-party state to reign supreme. Such measures threaten open societies, much more so than, say, the rising atavistic ideological movement of populist nationalism.92 On 24 January 2019, George Soros addressing the World Economic Forum in Davos said, “I want to call attention to the mortal danger facing open societies from the instruments of control that machine learning and artificial intelligence can put in the hands of repressive regimes,” citing China as a pioneering example. In public perception such existential AI threat is yet to sink in because the threat is new, and its workings are hidden in black boxes bearing cutting-edge technology. This gives repressive regimes an inherent advantage in controlling the population; in open societies blanket surveillance creates mortal danger for every individual. The future can be dystopian.
Opportunities for digitizing and automating tasks are far from being over. The more cognitive tasks are automated and embellished with language processing and pattern matching, and enhanced with physical dexterity, mobility, and sensory perception, the bigger will be its impact on depriving human workers of jobs performing these tasks and on division of labor in a society. Eventually, inter alia, assembly line workers, taxi drivers and long-haul truckers (Google/Waymo, Tesla, nuTonomy, Uber, and many others have already invested in self-driving vehicles) will have to seek other forms of employment. Machines capable of perceptual tasks, e.g., language translation, speech recognition, text reading, computer vision will begin to replace human specialists in pathology, radiology, security, language translation, paralegal work, and many others.
Hardware for implementing AI software continues to progress rapidly as are energy sources that power the hardware. Beyond mere speed up and energy efficiency, other important technologies are advancing too. These include mobile Internet, IoT, cloud computing and storage, AI, autonomous vehicles, robotics, virtual and augmented reality, virtual personal assistants, fitness trackers, everything cloud, 3D printing, hyperloop, drone services, renewable energy, and machine learning. And finally, there is bionics leading to a future population of super-humanoids capable of forming a society and running the world.
Presently, some humans (the vast majority is excluded) are more effective than computers at tasks, e.g., creative reasoning, non-routine dexterity, and interpersonal empathy. But to be mentally agile and efficient, they need AI enabled computers. AI will perhaps take some time to take on an Einstein or a Newton, but perhaps not too long a time. Accidents can and do happen. Chance may favor a prepared robot. Vitiating the scene further are man-made problems: global climate change, a geriatric world, starvation amid plenty, exploding need for education, skilling, health care, channelizing immigration, etc.
The enhanced role of business and of leading business executives in shaping the 20th century has been phenomenal. The growth of the military-industrial complex and its role in times of war made the U.S. a formidable superpower. They showed that they could convert commercial operations to defence production with speed and efficiency. They showed uncanny ability to harness vast resources and achieve high peaks in productivity in support of war efforts and national emergencies. The wide-ranging consumer products and service industries permeate our lives from the cradle to the grave. This phenomenal aspect of division of labor is the creator, sustainer, and employer of the middle class. That established division of labor is now under attack by AI machines in manufacturing and services because of breakthroughs in computing, communications, 3D manufacturing, and mobility made in the 20th century.
When AI creates massive unemployment, the Devil’s workshop will flourish. Mental illness, ill-feelings against the world, jealousy against the prosperous, the craving to destroy, etc. will grip people on a massive scale. Invasion of privacy with criminal intent e.g., through identity theft, hacking and stealing confidential personal data will become commonplace.93 The most vulnerable will be those born into the middle-class, brought up in cocooned security and the promise of a comfortable future if they did well in their rote education.
[ 85 ] Hof (2013).
[ 86 ] Nature-Editorial (20161124). “The power of big data must be harnessed for medical progress. But how?”
[ 87 ] Nature-Editorial (20161124).
[ 88 ] See, e.g., Lancet-Editorial (20180714).
[ 89 ] Nature-Editorial (20170406).
[ 90 ] Griffiths (2017). See also: Richardson (2017).
[ 91 ] Economist (20180531).
[ 92 ] Soros (2019). See also: Dawson (2019). It contains Soros’ speech at Davos, 24 January 2019.
[ 93 ] See, e.g., Revell (2015).