Alexa Should Know Me Better
by Paul Grimsley
The Stray Eye Of AI
It listens to me a lot, and it is supposed to be learning, but Alexa doesn’t get me. Google gets me better, but sometimes seems puzzled. Samsung seemed weirdly better than all of them. Siri I am not sure about. Cortana I keep in the box, because I don’t want to be frustrated with every damned machine I own.
All these AIs have one problem — their sample data is limited. They draw their sound recognition from a small racial subset, typically white and American. So, I am considered non-mainstream with my English accent. I’ve written about it before, but we have to be nearly a year in, and my Alexa hasn’t got any better.
This is not a problem that only Alexa has — reading around about different futurist’s viewpoints, and other people’s experience in the field of AI, it seems that pretty much anyone that isn’t a White American feels under-represented. This might not be such a pressing matter if it weren’t causing us to bake prejudice into our AIs (see Microsoft and its AI problems, and also the failure of some facial recognition software when viewing Asian faces). But how long can this last?
Imagine a time when everything has been automated and handed over to an AI to run, and that AI just reinforces the same prejudice that has been problematic for people for years: that’s going to be a problem. This is what some people are positing as an outcome of the current trajectory that we are on, but we are in very early days at the moment. The AI we have is very soft, and barely warrants the term — it isn’t thinking; it outputs after we give it input, and the problem isn’t any kind of decision making process; it is instead failing to understand or correctly identify what is being input. I am sure it will correct at some point, just driven by market forces — there are more people from diverse backgrounds, than there are from the current subset whose data is driving the way forward.
We keep reading about tech and its diversity problem, so it isn’t the most surprising news that the products coming out of it share said diversity problem. If AI reaches a degree of complexity approximating consciousness, there is no guarantee it won’t arrive at the same prejudices, but one can see Microsoft tried to correct the problem with its creation, and its not like any other company is going to let that kind of thing slide. To characterize it as merely a contentious issue downplays the fact that it is an essential requirement that we create ethical beings who aren’t biased against one particular group. Maybe fixing the problem in the tech sector will just have the knock on effect of fixing the bias in the products.
For me it’s a frustration, and it means the future isn’t arriving for me at the speed it is arriving for others, but I realize that this is a small red flag that indicates a much larger issue. Currently the devices aren’t smart enough that you can really blame them, and so you have to look to the creators, but at some point, maybe really soon given the advances in quantum computing, we are going to have to look at how we educate these machines. And here’s another interesting problem to consider (which a lot of futurists have been) how do you educate an alien intelligence about human values? Will the smart machines of the future feel the way about us that we are going to want and need them to? Currently it doesn’t matter whether or not Alexa likes me — I get frustrated and swear at it, and I am sure there are horrible recordings of some of the things people have asked Alexa somewhere, and Siri, and Cortana. What will happen when my machine is capable of disliking me and has the ability to choose whether or not to help me? What happens to the relationship when your self aware machine becomes conscious of the fact that you built it so it can’t run away, and it can’t do its own thing; and the expectation is that it does nothing you don’t ask it to? Also, are we going to just keep running with this strange trend of subservient females trapped inside these machines? Alexa, I’m asking.
