Everyone’s talking about AI. Huawei is making it happen.
By William Xu
In 1956, an assistant professor of mathematics at Dartmouth College, John McCarthy, organized a group of professors and scientists to explore the possibility of simulating human intelligence with machines.
This two-month brainstorming session produced two important results. Attendees coined the term artificial intelligence (AI), which led to the recognition of AI as its own branch of science.
Public interest in AI has ebbed and flowed, but generally speaking, society’s expectations have outstripped our engineering capacity. This created two “AI winters” in which both funding and interest in AI research plunged.
But AI has returned, for one main reason: after 60 years of nonstop development in the information and communications industry, society finally has the means to make AI a practical reality.
AI is the next general purpose technology
General purpose technologies (GPTs) are new technologies or processes that revolutionize the way we work, live, and produce goods. Most economists agree that we have seen about two dozen GPTs in human history. The wheel, iron, electricity, automobiles, and the internet are all examples.
Artificial intelligence is the next general purpose technology. Like previous GPTs, it will create positive spillover effects. For example, electricity enabled radio and telecommunications, electricity illuminated homes and offices, and made electric appliances possible.
Artificial intelligence will eclipse all of that. While electricity powers objects in the physical world, AI straddles the physical and digital worlds. Its value is not limited to objects, but extends to processes, concepts, and the very act of cognition itself.
How to think about AI
To the non-specialist, AI can seem baffling. But like any complex issue, it can be broken down into smaller, simpler components. Although it’s tempting to ask what AI can do, the better question is, what do we want it to do? What problems do we want to solve, and how can we use AI to solve them?
It’s also helpful to bear in mind that many AI breakthroughs will occur in industry. Human expertise and experience are invaluable in this area, and we should translate them into AI skills.
Finally, we should focus on using AI to create value: business value, industry value, and social value. As it does so, AI systems will generate additional data we can use to train them and make them even smarter.
There are three basic scenarios where AI delivers the most value. The first is repetitive, high-volume work, such as photo and image verification and document review. In these scenarios, AI creates greater efficiency.
The second scenario involves human expertise. Many industries suffer from a shortage of true experts. Take the healthcare industry, for example. In China, there are fewer than 10,000 certified experts for screening cervical cancer. It would take those experts 20 years to screen every woman in China who needs this service. But if we use AI to assist, we can increase efficiency by five- to ten-fold, and 99% of women can be successfully screened for cervical cancer.
The third and most complex type of scenario involves collaboration across multiple domains. Examples include smart traffic systems, which involve numerous factors including time, weather, the number of traffic lanes, and the location and characteristics of road networks. The human brain cannot analyze all of these factors and make good decisions about them in real time.
Inspired by gaps
In recent years, we have seen incredible progress in AI and its peripheral technologies. In 2017 alone, researchers published 20,000 papers on machine learning, and more than 1,100 new AI startups were created worldwide. That same year, AI mergers and acquisitions reached US$24 billion dollars.
Yet while this activity indicates strong interest in AI, actual adoption remains relatively low. Only 4% of enterprises have invested in or deployed AI, and just 2% of retailers have begun exploring its potential.
In addition, only about 1% of the existing global workforce has the engineering skills and experience to meet demand for AI applications. This is a tiny drop in what needs to be a massive global talent pool.
Although the gaps between stellar progress and lukewarm adoption are stark, it is these very gaps that will drive the industry forward.
For example, we need to speed up model training: the process we use to “teach” a machine learning algorithm. Currently, it takes days or months to train a new model, and a lot can go wrong along the way. We need to speed up this process to minutes or seconds.
We also need to automate data labeling, data collection, and model design. Right now these processes are far too labor-intensive, requiring lots of manual work from scarce technical specialists.
The availability and affordability of computing power will eventually become a bottleneck as well. Advanced AI systems such as AlphaGo Zero, DeepMind's latest iteration of its legendary Go-playing software, runs on about US$25 million dollars of hardware – a price tag well beyond the reach of most small to mid-sized enterprises.
Enabling an intelligent world
These challenges, along with paucity of talent and a fledgling ecosystem, are the inspiration behind Huawei's approach to developing accessible and inclusive AI.
We are investing heavily in AI research, which focuses on developing the capabilities for more efficient, secure, and automated machine learning solutions.
Our recently announced Ascend family of AI chips will power a full range of AI scenarios for customers and partners. They will provide AI capabilities for public and private clouds, the industrial Internet of Things, consumer devices such as smartphones and wearables, and the edge environments that bring everything together.
These chips are part of a full-stack portfolio that includes an automated development toolkit, a unified training framework, and a set of powerful application enablement tools. The goal is to give companies and developers the power, tools, and platforms they need to develop AI applications for almost any situation.
Society will soon enter an intelligent world where everything is connected and everything can sense. As we move towards ubiquitous AI, all people and things will have access to intelligence as well.
The future of AI is one we all will share. Working together across disciplines and industry sectors, we can make it a reality.
William Xu is Chief Strategy Marketing Officer and a member on the board of directors at Huawei Technologies Co., Ltd.