When artificial intelligence takes over ...
Have you ever wondered why your GPS navigation system gave you an impossible route?
Have you ever raised your eyebrows when logging on to your online shopping site and finding obscure products in your personalized recommendations?
Have you ever discovered you are charged more when buying tickets on a site you registered as a user and visited often?
I haven’t, until I was hit by the recent trending news of delivery riders, in which it was reported that the computer-calculated target times were mission impossible in some cases, while the optimized routes provided could violate traffic rules in others.
For a long time, I had taken so-called algorithms, big data and optimized computer calculations for granted, despite my little understanding of these words, or my occasional wondering about strange routes, recommended products or over-priced tickets as a returning customer.
I took it for granted that the algorithms are objective and trustworthy, as they imply no human intervention. After all, it is always humans who make mistakes, isn’t it?
Now looking again, I’m more suspicious.
What’s that baby milk doing in my personalized recommendations? I don’t have a child! Oh, it may be that one time I bought diapers for my cousin.
There were a few times when my cab drivers, old Shanghai uncles who say they know every corner of the city’s downtown area, frown at the GPS navigation. Sometimes they were right — the system pointed to a much longer route, in one case, a new road that had not been finished.
In other cases, they were wrong. The system pointed to a different route because the shorter one had traffic jams or accidents and took more time — the uncles didn’t know what the system knew — the real-time situation on the road.
AI engineer and angel investor Bell, whom I interviewed for the delivery riders story, told me about this idea of bias of algorithms and ethics behind algorithms.
“The initial data set could have errors, or those who created data had a bias or an agenda,” he said.
“And even good data may be constrained to cause artificial intelligence to learn the wrong way and lead to incorrect decisions. Then the key issue, you have intentional or accidental bias in the process of designing the algorithm — either the designer had a different agenda than announced or he designed a set of wrong elements in the equation.”
He gave me the example of prices for returning customers, saying it is no secret in the industry that prices are calculated differently for different types of customers. Against the common sense of a discount for returning customers, they are considered high-frequency users who don’t look at prices. So they are often given a higher price.
It reminded me of a conservation with a car-hailing service driver last year, who asked if I ever compared my prices with friends for the same route.
“You are probably charged more if you are using the service every day,” the driver told me.
I do use it every day, and out of curiosity, I did compare with my cousin who usually drives her own car, for the same route from my place to hers. She paid two bucks less. I couldn’t prove whether it was one-time happening or algorithm bias as Bell had told me.
I called customer services and was assured there was nothing called algorithm bias.
Then, I saw on the news this week that the Ministry of Culture and Tourism released a regulation prohibiting online tourism operators from abusing technology such as big data — not to set unfair trade conditions based on consumer’s past purchases and preferences. It will take effect from October 1.
I have long read and watched violent takeovers by artificial intelligence in sci-fi novels or movies. It had never occurred to me that it may have already be happening somehow, in a more gradual and calmer way without me realizing it.
Now I have come to recognize the takeover, I missed the times when I lost or cracked my mobile phones during international travel — worrying in the first two or three days, but it turned out to be fine not paying with cell, checking tourist sites for recommended spots or restaurants, following the digital map, or getting an Uber.
Oh, wait, my traveling companions did all those.