Glasgow – While models of artificial intelligence (AI) They still have difficulty reading analog watches and calculating dates with precision, a robot developed by scientists from the University of Edinburgh has managed to prepare coffee in dynamic environments, marking a new advance in the integration of AI in daily tasks.
A study by the University of Edinburgh released on Wednesday shows that multimodal AI systems, designed to process images and text, only succeed in the position of the clock needles in less than one in four occasions, with errors greater than 75 % in several cases.
The researchers tested several watches designs, including some with Roman numbers, with and without a second, and with different colors of sphere.
Errors were more frequent in watches with Roman numbers or stylized hands.
In addition, the elimination of the seconder did not improve the performance of the models, which suggests problems in the detection of the hands and the interpretation of angles.
In calculations related to dates in calendars, such as the identification of holidays or the determination of past and future dates, the most advanced models failed in 20 % of the cases.
“Our findings highlight a significant gap in the AI ability to perform skills that are quite basic for humans.
Despite these failures in simple tasks, the same institution has developed a robot that successfully executes a complex task in a real environment: prepare coffee in a busy kitchen.
This system, which combines advanced IA with high sensitivity sensors and precision motor skills, interacts with its environment more similar to the human, according to researchers.
The device – a robotic arm with seven mobile joints – interprets verbal instructions, analyzes its environment and can, for example, locate a cup in a drawer that has never opened before, measure and mix ground coffee with hot water and adapt to unexpected changes, such as someone moves the cup in full preparation.
The study, published in the magazine ‘Nature Machine Intelligence’, was supported by the Engineering and Physical Sciences Council of the United Kingdom (EPSRC).
“We are glimpsed a future where robots with increasingly advanced intelligence will be common. of the University of Edinburgh and project leader.