by Klara Melbinger
From the invention of the simple Spinning Jenny to Amazon’s launch of the Echo, humankind has oscillated between the virtue and the vice of machines. The Second Machine Age created “smart technology” consisting of iPhones, 3D printers and smart sleeping systems. AI, the art of engineering intelligent rather than smart machines, could shift the paradigm of machine use and their place in everyday life. The “Third Machine Age” promises to be an era in which technology no longer assists humans but acts alongside them. But some worry that intelligent machines may develop cognitive capabilities beyond human control.
In October 2015, AlphaGo, a program powered by the Google DeepMind Project, defeated European Go champion Fan Hui. The ancient Chinese board game Go is considered the most difficult strategy game for computers to win. Most famously, Deep Blue, a chess-playing computer developed by IBM, became the first computer to defeat a reigning chess champion during the Deep Blue vs. Garri Kasparow chess matches in 1996. But what sets the accomplishment of AlphaGo and Deep Blue apart is the fact that the number of possible moves in Go is far greater than in any other strategy game. For example, there are 10761 possibilities in Go compared to 10120 in chess.
Senior Research Scientist and Tech Lead at Google, Greg Corrado, believes that the construction of intelligent computers is not about creating machines that can learn, but rather creating machines “that can learn to be intelligent.” Research into the design of a combined AI, conducted by Google and Stanford, aims to merge the abilities of different “super-smart specialists.” Such systems could have impressive features, including straight-up object-recognition systems with the ability to recognize a cat or a fish in a photograph. Hybrid systems could go a step further and perhaps identify the significance of a scene depicting a cat reaching into a fish bowl, thereby displaying some understanding of the predator-prey relationship extracted from a pre-acquired knowledge base.
It is, in fact, the human brain that has prompted developments in AI. Companies such as Google, Facebook, Microsoft and Baidu, along with a handful of prominent researchers, have begun to take a new approach toward AI – Deep Learning. Machines should be able to learn and combine information by themselves, which entails processing masses of unlabeled data in contrast to feeding them with labeled, pre-identified data. Inspired by the human brain, Deep Learning systems use artificial neural network algorithms to process data.
By allowing computers to possess cognitive abilities, however, some fear a Terminator-like invasion or a society of machines mirroring humans emotionally, as depicted in Spike Jonze’s Her. Irving Wladawsky-Berger, Strategic Adviser on Digital Strategy and Innovation, gives an interesting account of how to perceive AI technology. AI can be considered the “next generation of sophisticated tools enhancing human capabilities,” which can be compared to the impact that electricity, the internet or cars have had on human productivity, and as a technology that is “radically different, because they [intelligent machines] embody something as fundamentally human as intelligence.”
In general, pundits are on-board with Wladawsky-Berger’s assessment. They believe that AI will assist humankind rather than eradicate it and when used as a task-specific tool, will increase human efficiency. As long as AI systems are specialized in a single field, they will hardly pose any threat. The danger, however, lies in the development of an Artificial General Intelligence (AGI), a computer able to process un-labeled data and draw conclusions beyond the scope of human control. The development of AGI was the initial goal of all AI-development programs, but since cerebral activity still remains to some extent a mystery, the creation of specialized AI remains dominant. Nevertheless, since 2008, specialists have gathered once a year at the “Conference Series on Artificial General Intelligence” to discuss the possibility of increasing the scope of AI technology beyond the level of general intelligence.
Already, AI has been put to use in many areas. Applications range from medicine to tackling cyber-fraud. The online payment system PayPal, for instance, has started to apply Deep Learning techniques that use non-linear algorithms. These algorithms are capable of analyzing a vast amount of latent features, including time signals, actors and geographic locations that may result in a specific type of fraudulent behavior and can even track fraud schemes. The startup Tera Deep incorporates Deep Learning systems with the ability to process inputs through deep neural-network algorithms. They can turn a simple webcam into something comparable to a Dropcam with voice or face recognition systems. Smart applications, such as Amazon Echo and Jibo, have already found their way into everyday households.
Neural networks that incorporate Deep Learning modules have already mastered the recognition of particular patterns and features of data, which has allowed for great progress in computer vision, speech recognition, text analysis and machine listening. In summer 2014, the Chatbot Eugene Goostman convinced 33% of humans participating in Turing Tests, a test intended to measure a machine’s ability to exhibit intelligent behavior, that the machine was indeed a thirteen-year-old from the Ukraine. Tests involved the Siri-like machine conversing with a human via an instant messaging platform. This success may significantly further the development of Google RankBrain, an AI program intended to help process search results. Google RankBrain will revolutionize search engines, enabling the use of colloquial language and implied questions.
The efficiency of AI systems and their data-storage capacities are based on the ability to make correlations. George Miller, an American psychologist, first discovered that the human brain’s capacity for short-term memory did not depend on the amount of information, but on its structure. A human brain can simultaneously store seven pieces of information in its short-term memory, whether seven letters or seven words. If the combination of words makes sense, the brain can remember all seven sentences. Thus, an AI that can make sense of a mass of information will be much more efficient, a project currently being tackled by the Google Deep Mind Project.
Progress in the field of AI is likely to go a long way before posing a serious threat to humankind. But some, such as technological pioneer Elon Musk, founder of Tesla Motors, SpaceX and PayPal, are demanding strict ethical guidelines to ensure lasting safety. And indeed, technological development grows at an exponential rate. As the use of AI becomes an increasingly ingrained part of our daily lives, the question of who will control those systems in mass use will become of paramount importance.