The Dark Side of Big Data, IV – Deadly Algorithms
It is important to look in advance of the
information, to establish what is possible before the possibility becomes reality.
Every truth discovered begins as a possibility. The promise of artificial
intelligence (AI) and big data is that it takes the human out of the decision
process. The promised truth of AI (the
algorithms) in tandem with big data is better analysis, faster execution of
decisions, safer driving, correct diagnoses, or whatever lies on the
table. What could be bad about that?
Everyone has heard of algorithmic data analysis
with built-in triggers for action. The
most prominent example is with the stock market. Algorithms are so entrenched within the
buy-sell patterns, many brokerage firms are now controlled by news reports and
financial input for potential transactions. Oil problems in the Middle East
signals sell, and baby, they sell. China
trade deal is near signals buy, and baby, they buy.
In the market, algorithms read the news looking
for key-words and phrases that indicate positivity or negativity for market
trends. They act with little human
intervention. Such algorithms factor
in types of news, stock prices, and
business trends. In brief, they are
powerful predictive tools. Or are they?
Results are difficult to evaluate.
Notwithstanding the self-fulfilling aspects, the net results may show the
market going up or down. It is difficult
for anyone to say that if the algorithms were not in place, the results would
have been better or worse. Moreover, who
knows if these algorithms have been hacked to provide third parties to
determine only seconds ahead of what will transpire – and act in just milliseconds
with their own algorithms based upon algorithms?
Recall the question, “Who is the best hacker in
the world? The one never discovered.”
Now suppose some firm has developed algorithms
based on Bayesian methods combined with big data, and other statistical and
mathematical techniques. It has some
profits proved, contingent on the above observations. Suppose the algorithms
are sold (yes, they can be sold) or the entire firm is sold to other interests,
whose desire is to automatically profit using the algorithms. This new firm decides to cuts costs by
eliminating its IT analysts. Thus, yesterday’s
algorithms march into the future making predictions, sales-projections, and
stock orders/sales based on previous situations. In a real sense, if the firm manages
sufficient money, the algorithms exert some power over future markets, but the controller
algorithms may no longer be relevant.
Example. Would you resurrect a stock expert
from 2009 (old human algorithms) to advise your purchases today?
This firm no longer has the expertise to
examine, much less change the algorithms to meet the day. This is dark.
Even darker is the firm decides to employ new IT analysts that scarcely
understand the old algorithms while it modifies them, giving a hodge-podge of
new code. News sources change, politics
change, trends change, and money supplies change. When the algorithms do not, there becomes a
dangerous brew of reality with fantasy. What
can you actually know if 80% of trades are algorithmic?
Another problem, another dark algorithmic
phenomenon, is when a firm denies searches to your website for any reason, say
an unfavorable political mention. The
trigger may be a single mention on the site.
Even with no further “violations,” the sanction put in place will never
be lifted. A single mistake can become a
permanent blot. What could happen is the evolution of new language style
designed to make points that algorithms will not detect – if this hasn’t
already happened.
We already have similar phenomena within the
justice system with small, small data – but fixed judicial algorithms. A person convicted for almost any crime
carries that dead bird around their neck for all their life. Expurgation is
sometimes possible with effort and cost.
Never, ever, with algorithms tagging a website as undesirable is
expurgation possible without some extreme effort. In reality, owners of the errant website may
never know they were tried, found guilty, and given a life sentence – by an
algorithm! Appeals are not even on the
radar at this point.
Even worse is your future visit to the doctor.
Your vitals and symptoms are input; the algorithms decide in microseconds your
problem and recommend a remedy. Does the
doctor accept or have the courage to contravene? Legal and medical problems hinge on any
decision.
Just as bad news is ahead when algorithms
evaluate you, based data oceans of information coming from social media,
medical records, school performance, insurance data, and employment records, to
make employment decisions about YOU*. Appeal? Sure, but to whom? You can’t take an algorithm to court.
Move over Edmond Dantes of The Count of Monte Cristo. Even you partly knew your accusers,
though not the why.
The military is already aware of problems created
by AI and big data in making battlefield decisions. Right now, they are stumped by this dilemma. We
will need another Clausewitz, rewriting the rules of command in wartime. Will we need generals with knowledge of and maybe
how to write algorithms? The War College may require new courses in mathematics
and statistics. The only bright spot here is that generals tend to be scholars
of their deadly craft. Can they keep up?
Currently, political campaigns are making
advertising decisions based on polls and local voting records. They have been truly effective in winning the
day. This makes even politics subject to
algorithms; it makes campaigns subject to political preferences. Good or
bad? You decide.
And
finally the worst of the worst… We are
rapidly becoming dependent on AI algorithms we don’t even understand and are
powerless to refute. Have we become like
children in someone else’s nursery?
Summary. The point here is that algorithms
require constant human attention and intervention, not unlike that of the
railroad engineer who keeps the train on the tracks. Algorithms need constant expert management. AI
is a modern-day Janus, two faces or many. One side is good for some, other sides
are good for others, but none are in charge of events. We may come to question
of just who is running the railroad. A curious new world is emerging.
*These oceans of information may not yet exist,
but they are on the way, 10 years max.
General References.
AI in Investing/Stocks. https://www.forbes.com/sites/forbesdallascouncil/2019/04/15/artificial-intelligence-in-stock-market-investing-is-it-for-you/#4eb399126524
AI in the Military. https://en.wikipedia.org/wiki/Artificial_intelligence_arms_race
AI in Politics. https://www.research.ox.ac.uk/Article/2019-07-12-oxford-university-opens-doors-on-ai-research-with-major-conference
AI in Search Engines. https://www.seomechanic.com/artificial-intelligence-impacts-search-engine-optimization/
Comments
Post a Comment
Please Comment.