Amazon, Google, Apple, and Facebook are rushing to build artificially intelligent smart systems to serve us, connect us, help us, and ... increase their own profits.
(Of course, not necessarily in that order.)
But there are at least eight massive challenges to artificial intelligence -- and most of them are not technological. Amazon's director of cognitive interfaces for Amazon Web Services AI Pilar Manchón talked about them today at TechBeach Retreat
Not surprisingly, many of them are about data and its uses.
"We all like having a real nice personal assistant that helps us," Manchón said. "To do that we have to share huge amounts of private information."
The eight core challenges are:
- Cloud versus edge compute
- Personalization versus privacy
- Trust and delegation
- Data bias
The first problem is where you do artificial intelligence.
Close to the individual or end points -- edge computing -- can be great for privacy and immediacy, but the cloud is where massive computing resources and bigger datasets can be applied to solve problems.
That's related to the second challenge: personalization versus privacy.
We want personalized experiences, but the cost of acquiring them is giving up massive amounts of personal data to Google, to Amazon, to Facebook, and, perhaps to a lesser degree, Apple.
Agency is an interesting one.
When Alexa does something for us, it is acting as an agent, doing our bidding. But because we don't own the agent, using personal assistants prompts difficult questions.
"On behalf of whom are you acting?" Manchón said referencing Alexa, Siri, or other personal assistants. "Are you acting in my best interests ... or in the best interests of the company you're from?"
(Interestingly, ordering batteries via Amazon's Alexa defaults to Amazon Essentials
batteries ... and you can't order other brands via voice. That has led some to say that Amazon is privileging its private label brands and taking over market share unfairly.)
Security is always an issue, of course, whenever your personal data is at stake -- and even more so for artificial intelligence and personal assistants. To be helpful, assistants need access to your calendar, your email, and, potentially down the road, the most personal social networks that you belong to.
Trust and delegation is an interesting one.
"What happens to the data?" Manchón asked. "What can it be used for? And what happens to that data when you travel around the world?"
A even more difficult problem, perhaps, is policy.
Governments are not notoriously quick at adjusting legislation to new technological capability. Even worse, government officials tend to be older, less technologically-savvy individuals who may not understand core issues, or what kinds of policies are needed or beneficial for safe and smart future development.
"Do you think that governments around the world are aware of these issues?" Manchón asked.
Data bias is another interesting challenge.
AI systems are only as smart as the data we feed them, and if that data is wrong, or biased, or incomplete, we see problems. Garbage in, garbage out, in old-fashioned computer parlance.
Manchón referenced Microsoft's famously racist chatbot, Tay, which was taught by people on Twitter to be intolerant and nasty. But there are less immediately obvious problems:
"Water sensors are light-based, so if you're black, they don't work very well," another speaker at the same conference, Uber's head of International Design Research, Nancy Douyon said yesterday. "And, Apple Watches can't pick up the pulse of people with darker skin."
The eight core challenge is fear.
There's plenty of fear of artificial intelligence and robots, Manchón said, particularly with regard to jobs. That will go away, she said, as people realize that AI will not do everything, but will complement work that humans do. The key is the relationship between people and smart machines.
Which makes Manchón's last statement so key:
"As Spiderman and Voltaire said... with great power comes great responsibility."