AI Passes Year 8 & 12 Science Tests with a Score of 80 Percent

Over four years ago, in San Francisco, a competition was held amongst 700 computer scientists to see who could develop an artificially intelligent program capable of passing a year 8 test with the winner taking home $80,000 in prize money. However, none of the 700 computer scientists could successfully code an AI program capable of passing the test. The artificially intelligent programs showcased were incapable of understanding and interpreting language and logical reasoning, a skill required of students entering high school.

Fast forward four years later and researchers at the Allen Institute for AI managed to make a breakthrough, they successfully engineered an artificially intelligent program capable of passing a year 8 science examination. Not only was the AI capable of passing the test but it scored an average of 90% on the test. This is significantly higher than the most advanced AI at the SF competition which scored a measly 60%. The newly built AI also managed to complete a year 12 science examination with an average score of 80%.

The AI, dubbed ‘Aristo’ proves to us that research and understanding in the field of artificial intelligence is growing at an exponentially high rate. AI is breaking barriers every other week, with news that something not possible a few years back has just been accomplished. It just goes to show that AI itself is not the limitation, it is our knowledge and understanding of how to manipulate data in order for AI to understand and interpret it.

Aristo was built for multiple-choice science tests only.

Aristo was built for multiple-choice science tests only.

A science test doesn’t just require the ability to memorise something and just regurgitate it onto a sheet of paper. Science requires logical reasoning and a good sense of mathematical ability in order to approach the problem at hand. Connecting the dots by using the information at hand and understanding which rules are applicable help guide you to solving a problem. This is exactly how Aristo was trained to approach science exam questions. Although this is a breakthrough in our understanding of AI, we are still a long shot off AI mastering natural language and being able to fully interpret it.

This has significant business consequences. What I can say, with complete confidence, is you are going to see a whole new generation of products, some from start-ups, some from the big companies.
— Oren Etzioni - Former University of Washington Professor, currently overseeing the Allen Institute

Although advancements in this area of AI could lead to AI systems capable of carrying out conversations, it could also lead to the spread of fake news. Computer scientists must be careful when considering the problems at hand that come with designing an AI capable of recognising and understanding speech/text. We need to ensure that false information is not going to be spread at a faster rate with the introduction of new technology like this.

Sources: https://www.nytimes.com/2019/09/04/technology/artificial-intelligence-aristo-passed-test.html

Zacharia Sharif