Spend a week in my family, and you’d never know artificial intelligence is a limited tool. Every time I swear at Siri or complain about a search engine, of my sons delivers the dire warning that I have now insulted the AI and it will make me pay! Even though I know he’s kidding, there’s a part of me that feels there’s a kernel of truth in that message. Our lives are hugely influenced by algorithms that may or may not capture us correctly. And as our lives move more online, artificial intelligence controls more than we may realize.
I feel like I’m stating the obvious, but I am concerned that we’re approaching an inflection point at which we may no longer recognize the line between real and contrived. And we’ve already reached the point at which keyword searches have become less accurate and less useful. Even broad Google searches no longer give me broad responses. It has become harder to find scientific research on a general topic without knowing the title or author of a study.
Since AI makes our lives easier, why don’t I just shut up and enjoy it? It’s a valid question.
I do enjoy the ways in which artificial intelligence works well and saves me time. But that pleasure is tempered by the bias AI learns through natural language processing. For example, if the word doctor is grouped with male pronouns more often and nurse is grouped with female pronouns more often in natural language, then AI learns this pattern and determines doctors are usually men and women are usually nurses. From that point forward, the information that is aggregated and sent to you will be influenced by that bias. Over time, this creates an environment in which AI bias can become predictive by the manner in which it filters the information you receive.
There are ways for programmers to address such bias, but companies may not be motivated to do so if no one is paying attention, especially if the bias is feeding their bottom line. I know this may seem trivial on the surface, but it can affect our medical care, food choices, dating options, job opportunities, business success, and whether or not we receive our phone calls.
Research has shown that AI is more accurate than a majority of radiologists in reading mammograms but less accurate than the most accurate humans. That means your best chance for an accurate diagnosis would be the most accurate humans.
As a practical matter, there’s no way to know who those are. That’s where AI can be a useful tool. What if all mammograms were read by both a radiologist and AI? Essentially, the technology would work as a second opinion. If the two diagnoses differed, then a second human opinion could be consulted.
But as medical AI expands into more areas of medicine, the possibility grows that it could carry undetected bias as well. Studies that identify possible substance use disorder patients based on the language in their tweets can be affected by the algorithms Twitter uses. Not to mention, I believe there are people who can have drug using friends and not participate. But they still may use the language of those around them and risk being improperly identified and stigmatized.
Data is important and useful, but it often presents an incomplete or flawed profile. Think of the sheer number of fake profiles on dating sites or the number of polyurethane handbags that show up in a search for genuine leather handbags.
Last year, I received numerous job listings for epidemiologist positions in Los Angeles. I am not remotely qualified to be an epidemiologist and I live 1500 miles from LA, but something in my online history made search engines think those jobs were appropriate. While AI seems like a great tool for matching people with jobs, it may not bring the candidates you need.
So I say all of this mostly to raise awareness of what is happening behind our screens and to remind you that artificial intelligence is a limited tool. It cannot be trusted to replace human intelligence. We must provide the checks and balances it needs.
Believe it or not, that will soon be critical to thriving.