One of the latest wrinkles (among many) in artificial intelligence (AI) research is AI programs that can learn by watching. They don’t have to be painstakingly coded with instructions for performing a task. Nor, like the old numerically controlled machine tools, do they have to be guided step-by-step through a task until they can reproduce it flawlessly. Instead…
NVIDIA has an AI that can learn to do a task just by watching a human do it.
Game developers have to develop scenery such as urban streets for their games. The process is time-consuming and expensive. But NVIDIA has an AI that can watch dash-cam videos and then generate city scenes on demand.
Google’s DeepMind “has developed an AI that teaches itself to recognise a range of visual and audio concepts just by watching tiny snippets of video. This AI can grasp the concept of lawn mowing or tickling, for example, but it hasn’t been taught the words to describe what it’s hearing or seeing.”
At the University of California, Berkeley, an AI learns acrobatics by watching YouTube videos.
At MIT, another AI watched TV and learned to predict what would happen next. That might not seem terribly difficult, given the caliber of the shows, but it’s a major step toward AI being able to interact with people much as people do.
Now consider the Internet of Things (IoT). Broadly it means everything connected to the Internet—but not just your phone, tablet, laptop, and PC. It encompasses “smart” fitness trackers, hospital and personal medical devices, household assistants (think of Alexa), baby-monitors, nanny-cams, dash-cams, ordinary cameras, cars, hairbrushes, rectal thermometers, thermostats, mirrors, refrigerators, toasters, toilets, lightbulbs, light switches, and more—anything that can collect data and deliver it to your phone, your cloud repository, your doctor, or… They also talk to each other, so your Internet-connected refrigerator can tell your car to remind you to pick up milk on the way home from work (or call the grocery so the milk is waiting on your doorstep). Or perhaps the toilet can call your car to inform you that you need more fiber or you need to leave work early because you now have a doctor’s appointment. Or you can have the car tell the robomower it’s time to cut the lawn, if the sensors in the irrigation system haven’t already done so.
There are real benefits to such a thoroughly interconnected world. In industrial settings, it can increase efficiency and reduce waste. Perhaps it can also reduce the need for human workers.
But at home? It would drive me stark, raving mad! Like most of us, I am accustomed to paying a certain amount of attention to what goes on in my life. I know what the fridge’s inventory looks like, when the grass is getting long, when my hair is falling out, and how I’m feeling. There’s even a whole self-improvement Thing dedicated to helping people pay attention—mindfulness. I’m sure it’s only a matter of time before fitness trackers such as Fitbit start telling you not only how many steps you’ve walked today, how you’ve slept, and your heart rate and weight but also how mindful you are.
Okay. Your mileage may vary, and nothing could possibly go wrong. Right?
Well, it’s all pretty new and the techies are still working out the bugs. Security is a serious issue, as the discovery that smart lightbulbs are storing Wi-Fi passwords in plain text shows. Then there’s the baby monitor that got hacked, with kidnapping threats. IoT devices have been used for a botnet. Will Knight at Technology Review notes that the Chinese company Huawei, which makes much of the gadgetry underlying the IoT and has been accused of stealing trade secrets and being connected to the surveillance-happy Chinese government and its military, is facing restrictions by many countries because of national security concerns. The company does not seem to have much regard for privacy, either, so if they get their ears into your personal devices… And they aren’t the only ones watching you. In 2017, a New Hampshire judge ordered Amazon to hand over an Echo device’s recordings of what was going on in a home during a double murder.
But all this is hardly the worst of it. Look back at that business of AIs learning by watching. As the AIs behind Amazon Echo’s Alexa and other IoT devices get upgraded, they seem quite likely to gain that capability. At that point, everything in your house with a camera will be watching you. Everything with a mike will be listening. What you type—including passwords—will not be secret. What you take out of the fridge or liquor cabinet or medicine cabinet will not be secret. What you whisper in your lover’s ear will not be secret. What you do when you look at xxx materials on the TV or computer won’t be secret. You will, in other words, be a completely open book.
Now—what will the AIs reading that book be learning? Those owned by corporations will be learning how much money you have, where it’s stashed, what you spend it on, and what you are thinking of spending it on. They already know a good deal of that, to be sure, for your thoughts and wishes are often expressed by what you Google, which promptly shows up as ads in your Facebook feed. But they don’t—yet—know what you talk about with your spouse. They will.
AIs owned by police forces may well know when you commit a crime, or even talk about committing one.
And AIs owned by hackers… That brings me to the title of this essay. They’ll be learning to be you. Given what’s already happening, it may be less than a decade before an AI can tell when your teenaged daughter has gone out to visit her friend, Molly, call her from your phone, and say in your own voice, “I’ll pick you up at Molly’s house. The car had to go the shop, so I’m driving a loaner. Look for a blue van.”
That is a particularly nasty form of AIdentity theft.
An AI “you” could also do other things. “Swatting” is calling 911 to report a crime in progress at someone else’s address. Innocents have been killed as a result. The AIdentity theft version could involve calling 911 to say, “’I’m shooting everybody!! They’re all plotting against me!! BANG! BANG!” (Bomb threats are so passé.) Again, innocents could die. Less life-threatening would be calls to the Internal Revenue Service reporting folks for tax-evasion. Or to Immigration and Customs Enforcement (ICE), reporting people for having forged citizenship papers.
What else could a malicious soul with such an AI do? Cancel credit cards, insurance policies, or leases, spoil relationships, close bank accounts, quit jobs… The one certainty is that those malicious souls will think up new nasties.
There is of course nothing restricting these souls to being individual human beings. What could a nation do, as an act of AIdentity war? What chaos would result if every employee in a country sent the boss an “I quit, you SOB!” email? If every boss sent every employee a “You’re fired!” email? Better—remember the bit about your daughter—what if those emails were phone calls?
If such thoughts stoke your paranoia, what can you do? With AIs watching you from everywhere in your house, it’s not enough to protect every IoT device with strong passwords (different for every device, of course—you know you should). They’ll be learning by watching, remember.
It is probably best to consider very carefully whether you really need
everything in the house to be connected to the Internet. And if the day comes
when you can’t buy non-IoT appliances, whether their IoT functions really need
to be turned on.
 I’m not kidding. https://www.usatoday.com/story/tech/2018/01/11/do-we-really-need-alexa-powered-toilet/1024646001/