Most artificial intelligence (AI) are very specific. AlphaGO was developed to play chess and is able to beat all human in that area but it knew nothing about the real world. On to the other hand, artificial general intelligence (AGI) such as chatbots, have a relatively wider spread of common knowledge, even though they are still not perfect. We can show toddlers a picture of a cat and they will be able to identify a cat in the natural setting but AI is not able to learn in that manner. 

It also remains unknown whether AI could be turned into AGI or vice versa. Nevertheless, it is undeniable there is an increasing interest in developing AGI. Two weeks ago, Microsoft announced invested $1 billion in a partnership with OpenAI, a company co-founded by Elon Musk, to develop AGI or advanced software that has the intellectual capabilities comparable, if not, superior to the human being. 

The present AGI development challenges 

However, AI or AGI development is generally limited because of the absence of suitable hardware. Edge devices like mobile phones are not venues for AI development, explained by Luke Fleet, Physics Editor of Nature in a recent podcast, it has to be done in data centers. These data centers, often, are not environmentally friendly nor efficient. Their carbon emission is high and a large amount of water is required to cool down the CPUs (central processing units). Heating becomes a huge hurdle as it controls how fast CPUs could run to ensure they will not melt. 

Besides, AGI is developed either via the computer-science based or neuroscience-inspired artificial neural network, and each of the approaches runs on separate and incompatible platforms. A group of Chinese scientists from Tsinghua University’s Beijing Innovation Center for Future Chip had developed the Tianjic chip, a new hardware with a hybrid coding scheme to facilitate the running of different types of neural network at the same time on the same platform. Just like a mini data center but at one’s fingertip.  

In need of new hardware architecture 

The scientists presented their results in Nature. There was also an accompanying video to demonstrate how the Tianjic chip managed to perform simultaneous processing of diverse algorithms in an unmanned bicycle system. In the video, the bicycle was able to detect objects, take verbal instructions, track, balance and avoid obstacles without human interventions. 

Fleet noted comparing Tianjic chip with other chips out there, it is not clear if a chip capable of running multiple neural network would outperform other chips that just run on a standard type of neural network. Probably because of the unclear advantages such hybrid hardware will give, the industry is still not keen on playing around with different designs. 

Yet, the invention brought out an important idea: in order to develop new AI, there is a pressing need for new hardware architectures. The scientists themselves also noted in the paper, the Tianjic chip and unmanned bicycle system had provided an excellent experimental platform to study the evolution of AGI. They can now use it to experiment other critical issues such as whether the system is able to perform autonomous learning, to adapt to different environmental challenges like different road conditions and weather, as well as to generalize what it has learned to other areas. 

Author Bio
synthetic gene empathy chinese artificial intelligence data medicine healthcare ai

Hazel Tang

A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.