![](https://csdnimg.cn/release/download_crawler_static/87444635/bg9.jpg)
9
2. Operations to strategy:
the growth of AI everywhere
A significant driver of AI is the growth of data gathered
from machines and devices, i.e. IoT data. IoT data has many
immediate operational benefits, driven simply by allowing
two machines to ‘talk’ to each other and use the interaction
to drive some logic (e.g. a truck remains in the yard; don’t
close the gate).
Collected together, however, large sets of IoT data
can provide unprecedented situational awareness for
organizations, and this can drive simulations, produce
insights, and enable more accurate predictions. AI models
can help in this regard by quickly generating useful
decisions, ideas and forecasts, deriving value from the vast
volumes of IoT data available.
As such systems are built, we can contrast their efficacy with
the incumbent decision-makers: experienced humans. While
people are not able to consider every data point, we can be
remarkably good at making good decisions quite fast (often
using shortcuts such as rules of thumb and intuitions we
can’t quite explain). When organizations have both of these
potential decision-makers available, when would we choose
to go with AI?
When would you give an algorithm control?
Let’s imagine you lead an organization that needs to
adjust a range of operational settings each day to optimize
production.
Every morning a group of senior colleagues agree these
settings, based on their assessment of how the day will
unfold. Your performance data shows that these adjustments
are sufficiently useful half (50%) of the time.
If AI could match this performance, would you use this to
automate control of the settings? Or would the AI need to
do better than the employee group? We asked respondents
to imagine this scenario and asked them what level of
performance would lead their organization to switch to AI.
On a purely rational level, switching at equivalent
performance makes sense because you save the time and
effort of the group of employees every morning
2
. However,
only 10% of respondents would make the switch at this level.
Close to half (48%) wanted to see AI achieve 60%, while two
in five respondents would require 70% or higher.
We may be tempted to permanently use both AI and the
employee group at the same time. But what if they don’t
agree? This would introduce yet another decision, potentially
stalling the process. It is likely that one decision-maker
will become dominant over time, in which case the
other becomes redundant, and so efficient organizations
will choose one or the other. (Also note that the same
redundancy would occur if the two always agreed. With
identical performance, it would be become potentially
wasteful to keep both systems in place.)
In reality, we would expect many organizations to let the AI
“shadow” the employee group, until everybody is convinced
that it is safe and effective. Hopefully this consensus would
be easy to establish. If it is not, organizations will need to
pioneer new ways to resolve the issues that emerge when
impressive AI model clash with experienced employees.
Understanding trust
The requirement for outperformance which our respondents
wanted shows how our thinking around AI is not always as
rational as we may expect and that we need to understand
the role of trust and feelings in these decisions.
“AI models are becoming very complex and so we need to find new ways
to make them easier to interpret and understand. This is very important to
building trust in these systems.”
Organizations are using AI to make sense of vast internet of things datasets, while edge
computing is helping to embed automated intelligence into the world. These twin trends will
drive greater predictive power, more autonomous machines and richer simulations.
2
In this scenario the AI system is already in place, so capital investment is not part of the equation, only operating costs (e.g. energy costs, employee remuneration)
and the opportunity cost (i.e. of having employees working on something else). Hence, respondents could reasonably deduce that the AI system would cost less
than the employees time, but this was not explicitly stated (in order to limit any bias).