An artificial intelligence system (AI) has passed a third-level science test, correctly answering 90 percent of the questions. The result was obtained by Aristo, an AI made by the Allen Institute for Artificial Intelligence in the United States, demonstrating once again the rapid progress in the field of artificial intelligence. In the future, Aristo could be adapted to other contexts, for example to produce automatic response systems and autonomously support conversations with human beings.
In 2015, a competition offered a prize of 80,000 dollars for those who were able to build an AI capable of passing a third-level science test („8th grade“ in the United States). More than 700 IT experts participated in the initiative, but none of them had succeeded in achieving the goal: the most promising AI had stopped at 60 percent of correct answers. Just four years later, Aristo has achieved surprising results, demonstrating the acceleration of the AI sector in recent times, with increasingly refined and efficient systems.
At the Allen Institute – a research organization founded by Microsoft co-founder Paul Allen – they started working on Aristo in 2013, wondering if it was possible to emancipate AI from the classic tests they are subjected to to measure their abilities, usually board games and video games. They thought that tests in science schools were a good alternative, because – in addition to requiring the learning of specific notions – they make it necessary to apply the logic to solve problems.
The development of Aristo is started by BERT, a system developed by Google researchers (and therefore made available to all) to read and interpret the texts, in order to use the information collected for other purposes. By sending us thousands of entries from Wikipedia and novels, for example, BERT learns to guess words deleted from a sentence, making them meaningful again.
In a short time, the Google system has gathered a huge amount of information on how the language works, a capacity that can be exploited by others – like the Allen Institute researchers – to build their artificial intelligences.
Aristo was provided with thousands of questions and science test answers, which allowed him to learn the characteristics and ultimately to answer autonomously to other similar questions. In a short time, Aristo has shown that he can correctly answer both simpler and notional questions, and others in which it was necessary to use logic.
Hundreds of research institutes around the world work on systems such as Aristo and, in a short time, have achieved results that only a decade ago seemed unattainable. The increased computing power of computers and the refinement of cataloging and data learning systems have made it possible to make great progress, reducing programming times.