Anniversary Edition.
Machine learning techniques are still rudimentary, but
they’re improving all the time. Google’s Eric Schmidt predicted
in 2016 that they will underpin “every successful, huge” ini-
tial public offering during the next few years, and he might
be right. The limits of the different techniques—which go by
names like neural nets, ensemble learning, and support vector
machines—have yet to be reached. But in general, additional
data mean better results. The more a system knows about you,
the more it helps.
In a few years, a virtual assistant may place and receive phone
calls on your behalf, send emails, drive your car, and diagnose
medical conditions. This is not general-purpose artificial intel-
ligence like we see in science fiction, but it can still be useful.
And your future assistant will do a better job if you let it access
personal information about you.
Every major technology company I’m familiar with uses
machine learning techniques. Amazon uses them for product
recommendations, personalized ads, Alexa, and its Amazon
Go physical store. Twitter uses them to create personalized
timelines. Netflix uses them to improve streaming quality and
to personalize not only movies but also the on-screen artwork
displayed for each movie. Facebook uses them to recognize your
friends’ faces and generate street addresses from satellite imag-
ery. Yelp uses them to improve image classification (so restau-
rant photographs uploaded by users are categorized properly).
Thanks to off-the-shelf services like Google Cloud and Ama-
zon Web Services, machine learning employing large data sets
already has become the foundation of personalized computing.
Some libertarians may like the trade-off involved in shar-
ing personal data in exchange for recommendations or virtual
assistants. Other libertarians may not. But the choice should
be yours: Your idiosyncratic dislike of someone else’s personal
preferences does not qualify as a compelling reason to demand
government intervention.
None of this should be taken as a defense of companies that
lie about what they’re doing, conceal important details, or fail
to adequately protect their users’ information. Laws prohibiting
fraud remain libertarian-compatible, and the tort bar will be
happy to pounce on misfeasance. An example from 2005: Sony
BMG failed to disclose that its CD copy protection contained a
so-called rootkit, which introduced vulnerabilities and leaked
user data. It was a braindead corporate decision, made worse by
management’s initial response, which ended with Sony writing
settlement checks for up to $50 million.
By now, astute readers will have realized that there is a poten-
tial privacy problem separate from corporate blundering: If
large quantities of your data are remotely stored on servers, law
enforcement and intelligence agencies will surely demand the
ability to gain access. Worse, the privacy threat increases with
the volume and sensitivity of the data. The very information that
allows a virtual assistant to operate efficiently—your spending
habits, political and religious views, and minute-by-minute
location—is a target for a legal, or perhaps even extralegal, fish-
ing expedition.
One response would be to enact a law curbing what informa-
tion third parties can collect. But that makes as much sense as
preventing companies from manufacturing binoculars simply
because police can use them for spying.
The more sensible approach is to curb government sur-
veillance. That means taking steps such as updating the Pri-
vacy Act of 1974 to limit government access to outsourced
databases; increasing the authority of inspectors general at
federal agencies to monitor abuses; boosting criminal penal-
ties for lawbreaking officials; and perhaps most important,
rethinking the drug laws that continue to invite snooping into
our personal lives. (In 2017, over half of the 3,813 federal and
state wiretaps reported to the Administrative Office of the U.S.
Courts were for drug-related offenses. Investigations into vio-
lent crimes, including homicide, robbery, assault, and kidnap-
ping, amounted to less than 10 percent of the total, though
these figures do not include wiretaps done under the Foreign
Intelligence Surveillance Act.)
Another crucial fix is to provide a broader legal shield pro-
tecting personal data held by third parties. This is happen-
ing, albeit slowly. In June, the Supreme Court ruled that police
needed a warrant supported by probable cause before they
could obtain cell-site records (which revealed location infor-
mation) from a man’s wireless carrier. And since the 6th Circuit
Court of Appeals’ 2010 ruling in U.S. v. Warshak, technology28 OCTOBER 2018
“Private data collection
done with the user’s
consent isn’t spying. It’s a
way of figuring out what
individual customers
want and need to serve
them better.”