We’re all used to proving that we’re not a robot online by clicking on all the squares that contain a traffic light or a bicycle or a fire hydrant.
It might not be the best way to prove your humanity soon though, as an Alibaba AI algorithm has just outperformed humans in answering image-related questions such as counting the number of children in a room full of people.
The company’s AliceMind algorithm was taking part in the annual VQA (visual question answering) challenge, an annual event that pits AIs from the likes of Facebook and Microsoft against each other.
This is the first time that an AI algorithm from the Chinese tech giant has come first in the contest.
The challenge featured more than 250,000 images and over a million questions.
AliceMind scored an accuracy rate of 81.26% when answering questions related to the image, which was slightly higher than the 80.83% averaged by humans.
Alibaba has taken great strides forward in this area thanks to a number of technologies, including multimodal pretrained language models and adaptive cross-modal semantic fusion.
The combination of proprietary technologies has allowed the company to improve its performance in analysing the images and interpreting the questions, as well as reasoning out the answers and expressing them in a natural, conversational style.
VQA technology has a number of potential uses
Si Luo, head of natural language processing (NLP) at Alibaba’s Damo Academy, said that the technology’s success marked a significant milestone in the field of machine intelligence.
He added that the intention was to assist people rather than replace them, however, leaving humans to concentrate on the creative tasks they still excel at.
As well as triumphing at the challenge, Alibaba has already deployed elements of its VQA technology in a number of different applications, including its smart chatbot Alime.
Si said that the technology could also be used to search e-commerce sites for particular items, to provide medical diagnoses by analysing images, and to interpret images captured by the cameras in smart cars.
Alibaba has made a number of AI breakthroughs over the past couple of years.
In 2018, it outperformed the human benchmark in the popular reading-comprehension challenge the Stanford Question Answering Dataset.
It has also beaten the human average in the Microsoft Machine Reading Comprehension dataset.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
