AI Trustworthiness Challenges and Opportunities Related to IIoT
Figure 1
“The agent (orange blob in upper left part
of screen) learns to commit suicide to kill its
enemy (purple spring) and collects enough
points to get another life. The whole cycle is
repeated over and over again.” (Excerpt
from ‘Back to Basics: Benchmarking
Canonical Evolution Strategies for Playing
Atari’)
The main message here is that without any
specialized knowledge, the AI system was
able to find and exploit critical design flaws
that had eluded hundreds of thousands of
players for decades.
Today, AI is applied to complex problems
that are difficult for machines to address.
Enabling the AI capabilities to fill this void
presents the set of Use Cases that benefit
most from these new technologies, all of
which overlap the IoT verticals. The top AI
Use Cases involve serious side effects when
the AI fails to come to the correct conclusion.
Therefore, proper consideration of the
trustworthiness of the AI-related IoT
technologies, such as finding potentially
catastrophic flaws before they manifest
themselves, can save time, money, materials
and even lives.
Why is this novel? The AI was able to
discover previously unknown exploits,
hidden for decades, that allowed it to break
high score records. These exploits were later
verified by human operators on original Atari
hardware.
More than just fun and games
It is not hard to imagine different
applications for AI beyond video games. The
authors could have easily explored flight
control systems in aircraft, manufacturing or
medical equipment.
IIC Journal of Innovation
- 78 -