People & Strategy Summer 2017 Vol. 40 Issue 3 - 43
Misconception 1: AI Cannot Be Understood
by Those Lacking a Technical Background
Technical terms like "deep learning" and "neural network"
appear in any explanation of AI, creating the impression that
those without technical expertise will have a limited grasp
of AI concepts. In fact, AI itself is simple and fully within
the grasp of all functions including HR specialists from any
industry. HR leaders can disregard the technological aspects
of AI, and instead grasp the essence of what it is trying to
achieve. This understanding will precede a very important
time for HR.
while, the means of realizing these targets will change with
continuous flexible adaptation to the existing situation. This
is how adaptation to changing data and insights is achieved.
Rather than working through set rules, as prior systems have
There is a huge difference between
the AI we see in the movies
and the real AI.
In the 20th century, the standardization and institutionalization of work started by Frederick Taylor became widely adopted. Matched with the age of large-scale production, there was
a growing conviction that "it is best to set rules and maintain
their repetition." Such rule-direction came to influence the
design of business systems, structures, methodologies. Nearly
all legacy HR systems were developed within this context.
The computer was the defining tool used to make this
happen at scale, and the aforementioned institutionalization of work became widely utilized. Using rules, computers
can perform descriptive tasks quickly and at a low cost. This
rule-directed approach fit well within the 20th century,
the age of roads, railroads, telecommunications, and other
societal infrastructure expansion. These expansions required
controlled, closed systems and these rules helped to support
their growth. Many of these rules-based systems are still
As organizations evolved, we saw demand diversify and
change cycles shorten. The reasons for these shifts often
weren't predictable, and writing rules for these changes was a
complex, time consuming process. These rule changes were
nearly always reactive and slowed organizations considerably,
leaving them behind on productivity and the ability to scale.
The outcome was that we saw often prohibitively complex
combinations of how work could be done. In response,
companies began to standardize and institutionalize their operations, and manage employees through standardized policy
and procedures. These control mechanisms fit for the period
of time at the turn of the century, but they certainly do not fit
with a millennial population and the current organizational
focus on innovation, agility, and growth.
Creating Outcome-Direction Through AI
Outcome-directed thinking breaks the barriers imposed by
rule-directed boundaries, and the means to achieve it is AI.
An "outcome" is defined as a numerical value representing
the desired achievement level or target. For example, in
corporations, these values relate to business results, and in
medicine, these values represent recovery efficacy or other
outcomes. At Hitachi, one great aspect of the AI that we
have been working on over the last several years is that it can
be applied to all business environments. In fact, it can be
applied anywhere that we can define the outcome.
For outcome-direction, we must first set a target, followed
by the sort of numbers in which it can be measured. Mean-
done, outcome-direction AI promotes this flexible adaptation, up-ending the concept of rule-direction by adapting
on-the-fly to real-time analysis and results.
The next challenge: How can this be approach be put to
use systematically? Today, daily data is collected by information systems. For example, records related to past purchases
can be used in procurement systems, individual customer
purchase histories can be used in sales systems, and manager
careers can be recorded in HR systems. These records show
differences in, for example, changes between last week and
this week, customer A and customer B, or manager X and
manager Y. By closely analyzing large amounts of data, and
utilizing this non-rules based approach, flexibility can be
achieved in response to changes and diversifications that we
see on a day-by-day or minute-by-minute basis. The reason
for this is because, whilst the pattern of changes is often seen
as random or unorganized, these unperceivable changes can
actually be seen by AI as having a clear influence on specific
activity, actions or other phenomena, and critically on the
overall outcomes themselves.
We can now review past records that, when observed from
all angles through AI, in any given condition and with any
given action, produces a single outcome or result with a with
a very high value or correlation. Taking this bank of data and
applying AI reveals any prior, until now invisible relationship.
In other words, any data that can be reviewed by AI can link
the desired outcomes we seek and show us what factors impact the desired outcome. This is a hugely important consideration for all business.
Such newly discovered patterns are not theoretical or hypothetical as they show us the truth of the past that we were
unable to see at that time. Applying this truth, and matching
fluctuations to future control conditions, allows for flexible
adaptation to change, increasing the likelihood of the desired outcome. Such is the basic thinking behind outcome-direction or outcome-based thinking with AI. Previously, we
would have used judgment or experience in determining
which actions impacted outcomes. Often these assumptions
were inaccurate. Now, AI gives us the ability to see what actions really impact desired outcomes, whatever those may be.
This outcome-direction was too costly to uncover before,
and/or could not be implemented due to limitations across
systems of record and the general capability of business systems. However, these infrastructure technologies that already
contain accumulated data (including sales, performance,
VOLUME 40 | ISSUE 3 | SUMMER 2017