Self-Starting Artificial Intelligence

Artificial intelligence (AI) experts are both rare and valuable in today’s tech marketplace. Montreal-based Element AI reports fewer than 10,000 people in the world have the specialized skills required to build and teach complex, intelligent algorithms. According to the New York Times, Google, Facebook, and other massive firms offer unbeatable salaries, typically between $300,000 and $500,000, to those individuals, essentially cornering off the professional market from other emerging players in the AI space.

This talent shortage has led firms to look elsewhere to satiate the immediate demand for AI developers. Leading researchers have found automation may be the key to advancing AI capabilities despite the lack of human expertise: intelligent machines creating new intelligent machines.

In training machine learning programs, researchers flood an algorithm with relevant data to test, for example, how well it can recognize and identify certain objects or how accurately it can translate between languages. Based on the algorithm’s performance, researchers will alter the code until the desired level of accuracy is achieved. It now appears this task could soon be outsourced to machines.

Industry and academia’s most advanced AI efforts are creating algorithms capable of analyzing the development of other algorithms, learning how to recognize success and failure, and eventually making corrections to improve the software’s intuition. Google’s AutoML learning software has reportedly created object identification algorithms with higher rates of accuracy than those built by human experts. Researchers at MIT, UC Berkeley, nonprofit OpenAI, and Google’s DeepMind have also reported progress in the field.

If this “meta-learning” technology becomes more practical and accessible, AI could experience more rapid adoption in fields such as health care, transportation, and finance, where current capabilities are too expensive and unreliable.

Photo Credit: AILabs