There has been a lot of discussion around Artificial Intelligence
(A.I.) now that it is becoming more ubiquitous than ever due to three major developments
that took off in 2017 and into 2018: better algorithms, increases in networked
computing power and the ability to capture, store and mine massive amounts of
data.
The launch of voice-activated virtual assistants like Alexa,
Siri, Google Home, Cortana, Apple HomePod, and others has propelled A.I. into
mainstream thinking and the consumer market.
The fact of the matter is that visionaries, researchers,
scientists, and developers have been
working with Artificial Intelligence for more than sixty years. Only in the
past few years are we seeing an explosion of uses and devices from chatbots to
home assistants to medical diagnosis to robotic devices that vacuum our homes, mow
our lawns or manufacture goods.
Will A.I. be mankind’s final invention? Will it ultimately
destroy civilization and eventually all human life? Will it eliminate jobs?
Will it enhance human productivity beyond our wildest dreams? Will it be
mankind’s best invention?
Maybe, yes, maybe no, but I see three major areas of concern
that need to be addressed right away: algorithm security, algorithmic bias and algorithm interactivity.
Algorithms power our technology and pretty much how we view
and participate in the world. They are a complex web of if/then scenarios and a
set of instructions for the device. Every time you go to Amazon or Netflix or
any site for that matter your activity is being tracked and that is why when
you go to another page or another site, an ad or a suggestion may pop up of
what you looked at previously. Netflix’s powerful algorithms learn your
entertainment preferences and suggest similar movies or shows. Algorithms are
really the ancestors of A.I. These are just the tip of the iceberg considering
the algorithms used in A.I.
A.I. algorithms have become so complex with machine learning
and neural networks that in May this year the
European Union’s Data Protection Regulation goes into effect after decades in the
making. The regulation sheds light on the “black box” notion of the algorithms
and gives E.U. citizens the right to know how the algorithms work when machines
make decisions that affect their lives.
|
Stephen Hawking Elon Musk Bill Gates |
This is a step in the right direction,
but a herculean challenge for the likes of Facebook, Google, Microsoft and
other techno giants that are entrenched
in A.I. simply because no one is clear on exactly how the algorithms work or either
they’re too complicated to understand, or they’re proprietary algorithms that
companies want to keep secret.
In addition, the
AI
Now Institute at New York University, a research institute examining the
social implications of artificial intelligence, recently applauded New York
City in becoming the first city in the nation to take up the issue of
algorithmic accountability when it set up its Automated Decision Making Task
Force.
“The task force is required to present the Mayor and
ultimately the public with recommendations on identifying automated-decision
systems in New York City government, developing procedures identifying and
remedying harm, developing a process for public review, and assessing the
feasibility of archiving automated decision systems and relevant data,”
according to the
letter sent
to Mayor de Blasio by AI Now outlining the mission of the task force.
As we get more secure and comfortable with devices like
Siri, Alexa, Google Home, Cortana and the Internet of Things, can we implicitly
rely on information from these devices? Suppose these devices and others to
come were hacked and used for nefarious agendas. What if personal bias were
inadvertently or intentionally programmed into the algorithms?
In his book,
Future Crimes, in Chapter 8, In
Screen We Trust, Marc Goodman writes that every screen is hackable and “whether
or not you realize it, your entire experience in the online world and displayed
on digital screens is being curated for you.”
We recently experienced a similar catastrophic event when our
social media was altered falsely to sway public opinion away from one candidate
to another by Russian interlopers during the 2016 presidential campaign.
Imagine if many of the devices that we
depend on, especially our mobile phones, were hacked and the algorithms changed
to spawn a different result? The ensuring scenarios are unthinkable and
ultimately uncontrollable.
Algorithmic bias has recently come under scrutiny by various
researchers. The AI Now Institute is
working with the ACLU because of the high stakes decisions that impact criminal
justice, law enforcement, housing, hiring, and education to list a few.
Probably, the most challenging and catastrophic real-world
problem facing A.I. and its algorithms today is what I call Algorithm
Interoperability. It is described by organizational theorist Charles Perrow in his
seminal book, Normal Accidents: Living with High-Risk
Technologies - when complex systems are tightly coupled and designed to
immediately interact with each other.
The catastrophe started when a frightened trader ordered the
immediate sale of $4.1 billion futures contracts and ETFs (exchange-traded funds) related to Europe, wrote
Barrat. At the time, Greece was having trouble financing its national debt and
the debt crisis had weakened the European and US economies.
“After the sale, the price of the futures contracts (E-Mini
S&P 500) fell 4 percent in four minutes. High-frequency trade algorithms
(HFTs) detected the price drop. To lock in profits, they automatically
triggered a sell-off, which occurred in milliseconds (the fastest buy or sell
order is currently three milliseconds—three one-thousandths of a second). The
lower price automatically triggered other HFTs to buy E-Mini S&P 500, and
to sell other equities to get the cash to do so. Faster than humans could intervene,
a cascading chain reaction drove the Dow down 1,000 points. It all happened in
twenty minutes,” Barrat wrote.
Perrow called the problem “incomprehensibility,” according
to Barrat where an incident is not expected and incomprehensible for a critical
period. No one anticipated how the Wall Street algorithms would interact with
each other, and so the event was incomprehensible and unstoppable.
How do we solve these problems and not build A.I.
machines that cause more harm than good? With the EU algorithm law going into effect
and AI Now’s algorithm accountability initiatives as well as a slew of others
that will come to be, they will shed some light on these black boxes and make
their creators accountable. But will it make algorithms more vulnerable to hacking or copying as organizations will be required to
publicly reveal how they work?
What’s apparent is that A.I. algorithms need to be highly secured
with blockchain-based architectures, Distributed Ledger Technology and other
technologies to come, so they cannot be hacked and changed by bad actors.
Algorithm accountability means the black box aspect of this
technology needs to be explained and made public so that it is clearly
understood but not made vulnerable to hackers because of its transparency.
Accountability is paramount to ensure
that machine-made decisions are not made with human bias’ that were
inadvertently or intentionally programmed into the systems.
Interactivity may be the biggest challenge yet in A.I. to
prevent the buy/sell frenzy experienced by Wall Street. Perhaps, we need to
create an A.I. system that can test algorithm interactivity with multiple
scenarios in real time as new data flows into the system.
In any event, A.I. is a powerful new technology that should
be created with safeguards to ensure it works for the good of all of us.