KamChAI is one signal of the take-off of Artificial Intelligence and robotics in the tea industry. It is the first robot to be launched in Hong Kong, to make its complex and specialized milk tea. It competed against local and VIP milk tea masters in the annual KamCha – “Golden Cup” – competition in 2018. The contest aimed to test if a robot could match the concentration and skill of humans; in the background is the broader question of can it replace them. The answer is: not yet.
HK-style milk tea is an adaptation of traditional British fare, using various blends of finely-ground black tea filtered through a cloth net that looks like a silk stocking. Rather than fresh milk, it is cooled with imported evaporated condensed milk. This switch evolved in the local cha chaang teng no-frills tea restaurants whose customers couldn’t afford fresh milk and to give it a strong, sweet and creamy taste. The tea must be constantly agitated and steeped up to seven times to balance the thick and strong canned evaporated milk. There are many claims of secret recipes and of it being an art that takes years to learn.
The robot weighs 300 kilograms. It includes many sensors that handle the brewing motions and measurement of ingredients. Capturing these in software provides two levels of capability: (1) rule-based application of the procedures for handling all the steps, generally obtained by interviewing and observing experts, and (2) machine-learning neural nets that allow the robot to improve performance through the equivalent of trial-and-error learning, via “training” by masses of examples.
The competition version of KamChAI is at the start of this AI learning; it seems very much rule-based. It did not match the experts’ performance. For example, after making eight cups of tea, it poured too much into the last cup. “KamChAI is a little bit messy, but it’s like watching a child learn.”
It did not equal the speed of the human expert, who makes 12-14 cups per minute versus KamChAI’s 9. It’s unclear how and if machine learning is built into the robot’s current and future capabilities. Its developer recognizes that a key next step is to add face recognition that will identify regular customers and automatically make their individual brew. The hand and arm movements and sensors need training through examples to be able to adjust to misplaced cups, spills, variations in leaf, etc. There are many misconceptions about what the “Intelligence” in AI really amounts to. For example, a typical neural model implementation may distinguish coffee cups but not have even a tiny clue of that they are used for. KamChAI is at the low end of the I.
There is a secondary agenda in the plans of The Hong Kong Productivity Council (HKPC) which has championed it to “help expedite reindustrialisation, encourage smart manufacturing and promote technology transfer as steps to transforming Hong Kong into an international innovation center.” Hong Kong is short of skilled labor. In the service sector, estimates are that there will be a shortage of 80,000 workers for 300,000 jobs. Articles on KamChAI highlight that it may be the only way to preserve the skilled traditions and art of milk tea making.
As the machines and their software learn, they will evolve and even build new levels of expertise. Add mobility and they will begin to offer table and street service. In Hong Kong’s scarce physical space, they will meet a primary goal of HKPC: smaller and smaller stores with fewer and fewer staff. They will add more and more personalized service; also in Hong Kong today, HUNG+ kiosks combine face recognition, access to customer food and beverage purchase history and even current weather to make personalized snack and beverage recommendations.
AI is core to the future of labor-intensive industries and, yes, the job impacts will be far-reaching. So, too, will the service opportunities. And the management challenges are part of the package.
SOURCES: Hong Kong Productivity Council, HK Press