How AI startled the World
The Revolution
The artificial intelligence revolution is sweeping the globe, and it’s gaining momentum. The rate of progress in artificial intelligence is one of the most disputed aspects of the constant increase in teaching computers and robots how to observe the world, make sense of it, and eventually execute complex tasks in both the physical and virtual environments. And not just real product breakthroughs and research milestones are used to gauge how rapidly the industry is advancing, but also the predictions and expressed worries of AI executives, futurists, academics, economists, and legislators. The world will be changed by AI, but how and when are still unknown.
In a continuous effort to assist address those questions, the conclusions of a group of specialists were published today. The experts were assembled as part of the second annual AI Index and include members of Harvard, MIT, Stanford, the organisation OpenAI, and the Partnership on AI industrial consortium, among others. The idea is to use concrete data to track the field’s progress and make sense of it in relation to problematic issues like workplace automation and the overall quest for artificial general intelligence, or intelligence that would allow a computer to perform any activity that a human could.
Commercial and scientific activity in AI, as well as financing, is booming pretty much everywhere on the earth in this spirit of global analysis. In Europe and Asia, there is a particularly high concentration, with China, Japan, and South Korea leading the East in AI research paper publishing, university enrolment, and patent applications. In reality, Europe is the most prolific publisher of AI papers, accounting for 28% of all AI-related publications in 2017. China comes in second with 25% of the vote, followed by North America with 17%.
The Power Shift
Machine learning and so-called probabilistic reasoning — or the type of cognition-related performance that allows a game-playing AI to outperform a human opponent — is far and away from the dominating study category, according to the report.
Work on computer vision, a foundational sub-discipline of AI that’s assisting in the development of self-driving cars and powering augmented reality and object recognition, and neural networks, which, like machine learning, are crucial in training those algorithms to improve over time, aren’t far behind. Natural language processing, which allows your voice assistant to comprehend what you’re saying and reply properly, and the general strategic planning process, which will be required by robots as automated devices become more interwoven into daily life, are less relevant for now.
AI continues to improve in terms of performance, particularly in disciplines like computer vision. The time it takes to spin up a model that can identify photographs with state-of-the-art accuracy plummeted “from roughly one hour to around four minutes” in just 18 months, according to the paper, which measured benchmark performance for the widely-used image training database ImageNet. This translates to a 16-fold increase in training speed. Other areas, such as object segmentation, which allows the software to distinguish between an image’s backdrop and its subject, has seen a 72 per cent boost in precision in just three years.
Breaking down certain key 2018 milestones in domains like game-playing and medical diagnostics, where progress is advancing at unexpected rates, AI has dominated the endgame. These include milestone performances versus amateur and later former professional players in the online battle arena game Dota 2, as well as improvements from Google-owned DeepMind in playing the classic first-person shooter Quake in objective-oriented game variants like capture the flag.
Accuracy and proficiency in areas like machine translation and parsing, which allows the software to comprehend syntactic patterns and more easily answer queries, is improving, but with diminishing returns as algorithms approach closer to the human-level understanding of language.
The Evolution
All of this hard data is crucial in determining where the AI field is right now, as well as how it has evolved over time and is expected to evolve in the future. Yet, when it comes to more difficult questions about automation and how AI could be used in areas like criminal justice, border patrol screenings, combat, and other areas where performance is less essential than underlying governmental policy, we’re still in the dark. AI will only become more advanced, but there are a number of technological, as well as bias and safety, challenges to overcome before such software can be used without error in hospitals, police departments, and airports.
We’ve accepted that mass unemployment is unlikely to occur anytime soon as a result of automation, and the greater question is whether we as a society are prepared for the nature of labour to shift toward less stable, lower-paying jobs without safety nets like health insurance.
Not everyone will lose their job immediately. Rather, certain positions will be automated while others will be phased out over time. And some occupations will always necessitate the presence of a human. Workers’ fates will be determined by employer limits, labour laws and regulations, and whether or not a good enough system exists to help them transfer into new positions or industries. According to a McKinsey Global Institute analysis from November of last year, global automation might cost 800 million jobs by 2030, yet just approximately 6% of all jobs are at risk of being completely automated. How that transition from a human-only work to one aided by AI or robots is handled could be the difference between a full-fledged disaster and a historical paradigm shift.
The Aftermath
There isn’t always dread and gloom. Asking the proper questions and ensuring that policymakers, the public, and AI industry leaders get the data they need to make educated decisions is part of the AI Index report’s premise. It may be too soon to accurately assess AI’s impact on society — the industry is still in its infancy — but preparing ourselves for what it all means and how it will affect daily life, work, and public institutions such as health care, education, and law enforcement is perhaps just as important as the research and product development itself. We can only avoid the risk of developing technologies that change the world for the worse if we invest in both law enforcement and R&D.
