Article by Vyacheslav Polonski: “Unless you live under a rock, you probably have been inundated with recent news on machine learning and artificial intelligence (AI). With all the recent breakthroughs, it almost seems like AI can already predict the future. Police forces are using it to map when and where crime is likely to occur. Doctors can use it to predict when a patient is most likely to have a heart attack or stroke. Researchers are even trying to give AI imagination so it can plan for unexpected consequences.
Of course, many decisions in our lives require a good forecast, and AI agents are almost always better at forecasting than their human counterparts. Yet for all these technological advances, we still seem to deeply lack confidence in AI predictions. Recent cases show that people don’t like relying on AI and prefer to trust human experts, even if these experts are wrong.
If we want AI to really benefit people, we need to find a way to get people to trust it. To do that, we need to understand why people are so reluctant to trust AI in the first place….
Many people are also simply not familiar with many instances of AI actually working, because it often happens in the background. Instead, they are acutely aware of instances where AI goes terribly wrong:
- A Google algorithm that classifies people of colour as gorillas.
- A self-driving Uber that runs a red light in San Francisco.
- An automated YouTube ad campaign that displays ads next to anti-semitic and homophobic videos.
- An Amazon Alexa device that starts offering adult content to children.
- A Pokémon Go algorithm that replicates and amplifies racial segregation.
- A Microsoft chatbot that decides to become a white supremacist in less than a day.
- A Tesla car operating in autopilot mode that resulted in a fatal accident.
These unfortunate examples have received a disproportionate amount of media attention, emphasising the message that humans cannot always rely on technology. In the end, it all goes back to the simple truth that machine learning is not foolproof, in part because the humans who design it aren’t….
Fortunately we already have some ideas about how to improve trust in AI — there’s light at the end of the tunnel.
- Experience: One solution may be to provide more hands-on experiences with automation apps and other AI applications in everyday situations (like this robot that can get you a beer from the fridge). Thus, instead of presenting the Sony’s new robot dog Aibo as an exclusive product for the upper-class, we’d recommend making these kinds of innovations more accessible to the masses. Simply having previous experience with AI can significantly improve people’s attitudes towards the technology, as we found in our experimental study. And this is especially important for the general public that may not have a very sophisticated understanding of the technology. Similar evidence also suggests the more you use other technologies such as the Internet, the more you trust them.
- Insight: Another solution may be to open the “black-box” of machine learning algorithms and be slightly more transparent about how they work. Companies such as Google, Airbnb and Twitter already release transparency reports on a regular basis. These reports provide information about government requests and surveillance disclosures. A similar practice for AI systems could help people have a better understanding of how algorithmic decisions are made. Therefore, providing people with a top-level understanding of machine learning systems could go a long way towards alleviating algorithmic aversion.
- Control: Lastly, creating more of a collaborative decision-making process will help build trust and allow the AI to learn from human experience. In our work at Avantgarde Analytics, we have also found that involving people more in the AI decision-making process could improve trust and transparency. In a similar vein, a group of researchers at the University of Pennsylvania recently found that giving people control over algorithms can help create more trust in AI predictions. Volunteers in their study who were given the freedom to slightly modify an algorithm felt more satisfied with it, more likely to believe it was superior and more likely to use in in the future.
These guidelines (experience, insight and control) could help making AI systems more transparent and comprehensible to the individuals affected by their decisions….(More)”.