We've all been talking about AI over the last couple of months, debating the pro's and con's of ChatGPT, art generators and other AI software.
But there is one company which has been using AI extensively for at least the last three or four years; Amazon.
Image by Juan Agustín Correa Torrealba from Pixabay
The main fields where Amazon uses AI are in Seller Support, and in managing their catalogue of product data.
If you are a marketplace seller on Amazon, it is clear that initial replies to support tickets are created by a fairly basic AI that looks for keywords in your description of the problem, and then sends back a reply from a limited selection of pre-scripted responses. About a third of the time, the response is vaguely helpful.
The real issue is they way they handle product compliance. Personally, I feel they should talk to the manufacturers of a product (who will often be selling on Amazon themselves, and will find it far easier to submit the required documentation) but instead, they handle it on a seller-by-seller basis. This leads to inconsistency, with some sellers banned from selling a product while for others it is fine.
The first clue that there's a problem will normally be a strongly-worded standard email from Amazon, and waking up to a screen like this;
Screenshot by me, edited to remove my account info
Going in to look at the violations in more detail will get you to a screen like this;
Viewing details says they are "Restricted Professional Use Chemicals". The reality is that the products in question are water-based acrylic paints in small plastic bottles, widely available to the public, non-toxic and non-hazardous. They've been flagged up because the AI has spotted something (probably a word in the description or item title) that it doesn't like.
The sad reality is that initiating an appeal, even when we include all the requested information from the manufacturer, has so far resulted in a 100% rejection rate, with the rejection email normally arriving within 60 seconds of submitting the appeal.
This is almost certainly a bad combination of a poorly programmed AI combined with a policy that sellers are considered as a disposable part of the system. Just based on the way it behaves, it is clear that there is no human involvement at any stage of the process. With Amazon's notoriously insane standards they hold marketplace sellers to, it's likely that only a few more of these falsely-identified policy violations will result in my account being suspended and a corresponding loss of income.
This is just one example of how a poorly designed AI can have a significantly adverse effect on someone's life (not to mention stress levels - sorry if this post is sounding like a rant !).
What will happen when they use AI without human input to deal with routine pieces of bureaucracy, like parking tickets and social credit scoring ?