ChinAI #161: IBM Watson Leaves China in Defeat
The Inside Story
Greetings from a world where…
“it's no fun being the last fucking eunuch in the forbidden city” — Succession, Season 3, Episode 3
…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).
Feature Translation: IBM Watson in China
CONTEXT: Watson for Oncology, a clinical decision-support system for cancer treatments, has ceased sales in mainland China. Watson Health first entered China in 2016 and experienced a “honeymoon period.” By 2018, nearly 80 hospitals across 22 Chinese provinces and 43 Chinese cities used Watson for Oncology. At that time, according to one source, the Greater China region accounted for half of Watson Health’s revenue in a quarter.
Three years later, why did IBM Watson “leave China in defeat” (败走中国)?1 The fourth article in a very interesting Leiphone series — “The Medical Weak Points of the Giants” — tackles this question. The author, Yuchen Li, is currently drafting the last article of the series, titled “Why BAT Can’t Sell Medical Cloud.” *I welcome contributors to translate other articles in this series, and I always compensate translation work, thanks to those who financially support ChinAI.
KEY TAKEAWAYS:
The role of policy uncertainty and the geopolitical environment: changing U.S.-China relations made hospitals worry about the introduction of Watson, since its home base is in the states. The article references “invisible” restrictions on Watson introducing other services like imaging products, which further enhanced the difficulty of commercialization.
A range of other factors: With cases like these, it’s always hard to parse out the impact of government policies/pressure vs. alternative explanations. The article outlines many other factors, including stubbornness when working with local partners, leadership issues, and workplace culture.
Let’s unpack three additional factors further:
Public relations mismanagement: Many complaints about IBM not defending Watson Health against rumors and negative press about their products. The article states, “Unable to fight the fire of public opinion, the IBM Watson Health’s China marketing department is very passive, and most of their work only involves supporting exhibitions, which makes employees somewhat frustrated. On one occasion, Baiyang Intelligent, the agent of Watson for Oncology, went to the upper level management, hoping to speak on behalf of Watson for Oncology, but the response was still ‘no’.” Kang Ming (pseudonym), a former employee of Watson Health, attributes this to IBM’s roots in selling to big business and government clients, which makes them not value consumer-facing PR and self-media.
Unstandardized nature of cancer treatment in China: This was one of the drivers for Watson’s initial entry into China. IBM commissioned a survey by a consulting company. They found that in some areas of northeast China, doctors’ treatment plans for cancer patients were only consistent with standard guidelines for 30% of the time. The implication is that there should be a lot of potential for decision-support systems like Watson for Oncology. Two issues: 1) It’s a steep cost to create more standardized hospital information systems. One current project in a top hospital in southern China runs hundreds of millions of dollars. 2) Chinese doctors opposed Watson’s entry. Some doctors have linked interests with pharmaceutical companies and propose treatments outside of the guidelines (pharma rebates paid to doctors for oncology drugs are not insignificant).
Inherent issues with applying natural language processing-based solutions to clinical applications like cancer treatment: Cancer treatment is a highly individualized process that requires a lot of complex reasoning based on many indicators. A Chinese scholar who has studied NLP for many years said to Leiphone: “Watson’s problem is an inherent problem in the natural language processing industry—there is no structure. The structure is the connection between two things, it’s the problem of knowledge expression. No. There is no reasoning with structure, and there is no intelligence without reasoning. It is troublesome to solve this problem. The BERT model has no structure. It can only say that it has seen a lot of data, and it can guess. As for the scheme recommended by Watson, it is also based on calculation and probability. And the result of this probability has no clinical significance.”
ChinAI Links (Four to Forward)
Must-read: Harnessed Lightning — How the Chinese Military is Adopting Artificial Intelligence
By Ryan Fedasiuk, Jennifer Melot, and Ben Murphy, this CSET report analyzes 343 AI-related equipment contracts, selected from a broader set of procurement records published by PLA units and state-owned defense enterprises in 2020. This is one of the most detailed assessments of how the Chinese military is applying AI in practice. There’s so much to digest here, including examples of Chinese suppliers that “make a business out of sourcing foreign data or components and reselling them to sanctioned Chinese defense companies or PLA units.” One funny detail in the appendices: The authors also used an AI research assistant to check their manual coding of the 343 contracts into seven different application categories. On the first run through, this AI research assistant disagreed with the authors’ initial coding half of the time. Still, speaks to the extent to which the report tried to double-check results.
Should-read: GovAI Relaunching as a Nonprofit
Includes details about reorganization of GovAI, hiring opportunities, fellowship opportunities, an inaugural GovAI conference, and continuing seminar series!
Should-read: China’s Hypersonic Weapons Tests Don’t Have to be a Sputnik Moment
In an excellent piece for War on the Rocks, Sanne Verschuren argues:
China’s recent tests with hypersonic weapons systems — and the added layer of fractional orbital bombardment systems — are not a Sputnik moment. The technology is far less dangerous than it is often portrayed. However, these hypersonic tests fit in a broader pattern of the nuclear powers advancing their nuclear arsenals in ways that make the world less safe. Rather than trying to outbid China in a costly arms race, U.S. policymakers should start a conversation around the strategic implications of missile defense and rein in the ever-expanding U.S. missile defense mission.
Should-read: Gettysburg
This was the piece that made me think the most this week.
Thank you for reading and engaging.
These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is a postdoctoral fellow at Stanford's Center for International Security and Cooperation, sponsored by Stanford's Institute for Human-Centered Artificial Intelligence.
Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99
Other Watson products continue to be promoted in the Chinese market, including those related to clinical trials and drug discovery knowledge bases.