Jun 11, 2021

Chinese AI Model Wu Dao 2.0 Unnerves Europe

BAAI
China
AI
EU
3 min
Through its Wu Dao 2.0 model, China takes the lead in the race to dominate artificial intelligence and natural language processing

Unveiled by the Beijing Academy of Artificial Intelligence (BAAI), China’s latest artificial intelligence model can process natural language text, write essays and poems, and recognise images. Not only that, it can create alt text from static images and generate realistic images from text. Wu Dao 2.0 is multimodal, and it’s light years ahead of anything the U.S. or Europe has built. 

 

According to Wu Dao 2.0’s developers, the model is 10 times more powerful than OpenAI’s GPT-3, formerly lauded as state-of-the-art in AI. Trained on 1.75 trillion parameters, Wu Dao can sort through massive amounts of data and generate text that’s virtually indistinguishable from human prose. “The way to artificial general intelligence is big models and big computers”, said Dr Zhang Hongjiang, chairman of BAAI. “What we are building is a power plant for the future of AI. With mega data, mega computing power, and mega models, we can transform data to fuel the AI applications of the future”. 

 

Monopolising NLP 

What makes Europe nervous is that once China pulls ahead, other nations may never catch up. Language models are often a symbol of how advanced a certain country is in its artificial intelligence capabilities. Once a clearly superior model emerges, such as Wu Dao 2.0, companies will adopt it in order to compete in the global market. “We’re about to lose digital sovereignty in the AI space if we don’t act right now”, said Jörg Bienert, chairman of Germany’s Federal AI Association

 

Just as Google runs the playing field for search engines, Wu Dao 2.0 could grow to monopolise the language modelling world. As a result, some critics of the new technology warn that massive NLP models will enhance China’s propaganda mission. And as consumers turn to Alexa and Siri searches as a means for rapid answers, the U.S. and Europe worry that China’s AI could spread misinformation and skew global data. 

 

After all, NLP models often perpetuate the worldviews of the countries that create them—and Western countries are determined not to let China control the narrative. Tensions rose this month between western nations and China as the U.S. passed its Innovation and Competition Act through the Senate. In addition, European leaders have raised the alarm. “[China] beat the U.S. to it, while the EU was not even close in the race”, tweeted German MEP Axel Voss

 

More Money, More Power 

To accelerate the EU’s AI capabilities, seven European countries—Germany, Austria, Sweden, Croatia, Slovenia, the Netherlands, France, and Belgium recently sent a letter to the EU’s Internal Market Commissioner asking him to increase funding. France, for instance, will launch its Large European AI Models initiative to create Euro-based NLP models, but other countries are still far behind and need funding to catch up. 

 

Compared to the U.S. and China, which control 80% of blockchain and AI investment, Europe invests only 7% or €1.75bn. Overall, the EU faces a €10bn gap in AI research that it now intends to make up—if that’s possible. The truth is, European countries will have to work together and share research to stand a chance against the world’s leading AI superpowers. 

 

Although Europe has started to throw money at AI with programmes such as Horizon 2020, and recently developed new supercomputing and data-sharing abilities, many don’t think that the bloc can “counter the Chinese threat” alone. According to Bienert of Germany’s AI Association, the West must act before it’s too late: “[W]e need a collaboration between the U.S. and Europe”. 

Share article