IEEE Systems, Man and Cybernetics Magazine - October 2019 - 28

An effective or optimal teaming relationship needs to
machines must be trusted in psychological and sociological
be thought out that pairs human strengths for normal
aspects. There needs to be an overall understanding relaoperations with appropriate degrees of machine autontive to a machine's specific state and intent and of how all
omy. The suitable forms of human-autonomy interaction
information is properly accounted for in terms of its deciand communication need to be determined. It is important
sion-making processes. This is also valid reciprocally in
to minimize the acceptable burden on humans and optimany cases. Machines must have trust in their human partmize support from machines in accordance with their
ners as well. In a true partnership, a team trying to accomdegree of autonomy. For example, in aviation, standards
plish a specific mission requires a keen sense of each
for the respective roles of the pilots and aircraft can be
member's intentions [13], [16], [24].
created. Similarly, the Society of Automotive Engineers
The question to be answered is how a human-machine
(SAE) has established a standard of levels for ranking the
trust is going to be created, calibrated, indicated, transmitdegree of autonomy available in driverless cars and autoted, and understood. The autonomy development communimated vehicles [38]. A base model
ty must continue to use the lessons
car currently includes cruise conit has learned to inform and evolve
trol as a standard option, and that
the ethics and procedures to handle
An effective or
type of technology is part of the
the use of AUSs and their operationSAE's Level 1, while cars from
al employment. While the autonomy
optimal teaming
Tesla are already operating at
is supposed to be used as intended,
relationship needs to
Level 4.
some people might over-rely on it.
Other proposed means of speciIn cases where autonomy may not
be thought out that
fying the degrees of autonomy
be well suited, such as merging
pairs human strengths
exist and can be helpful. Examples
self-driving cars into congested
include the Sheridan scales, which
city traffic, integrating it would be
for normal operations
offer four- and 10-level categories
ill-advised. It is a challenge for
with appropriate
from manual to fully automatic
society to embrace autonomy espedegrees of machine
machine operation [39]. According
cially when it comes to attributing
to studies by the U.S. Defense Sciresponsibility when something
autonomy.
ence Board, autonomy levels are
goes wrong due to its employment
not very useful for military autonoin real-world scenarios.
mous systems; it is, instead, recommended
that
levels
should
be abandoned in favor of an
Quantifying and Ranking the Level of OA
autonomy
framework
[40].
In
any case, as autonomy
In terms of the rank or degree of OA, a continuous specmatures, we must determine how these technologies work
trum involves a tradeoff between the AUS and the operatogether synergistically to perform with the degrees of
tor/user. At one end of the spectrum are systems that are
independence that are required for various operations and
exclusively remotely operated and rely entirely on people
what those degrees imply regarding system capabilities.
in a very manual way to provide direct input to achieve
some goal. At the other end, systems will exist that are
capable of full autonomy that enables operations in a comThe Role of the T&E Community in Facilitating OA
plex environment with the specific ability to perform
Historically, the T&E community has been fortunate to
almost all of the activities that humans could. Along the
rely on analytic models for system performance using
spectrum, human and machine collaboration and cooperaassociated data that are fully quantitatively based. A crittion is required to achieve operational capabilities [37]. In
ical factor is that this does not apply well in terms of
considering the degree of autonomy required for a given
autonomy-based technology. The traditional T&E comoperation, the following questions and considerations
munity is accustomed to more determinism than the curoften apply:
rent autonomy technologies afford; however, emerging
technologies are demanding the creation and adoption of
◆ Which decision-making tasks should be assigned to
a completely new philosophy. For autonomy, the tradihumans and which to machines?
tional T&E community will have to accept that there is a
◆ How should the associated risks and disadvantages of
different standard that needs to be developed based on a
human and machine decision-making be mitigated?
new science. Such a standard may specify that an autono◆ People are great at perceiving and understanding commous system will do what it is supposed to do most of the
plex environments that are very challenging for
time (for example, 80%) and that what it does at the other
machines.
times (20%) is not predictable or cannot be well defined
◆ Machines are great at certain dull, dirty, dangerous,
by precise models and quantitative measures. Eventually,
difficult, or computational tasks and may be able to
it must be understood that this will suffice, and it is proboperate continuously (without the breaks that humans
ably the best that can be expected. Therefore, a part of
require) as long as they have power.
28

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE O ctober 2019

IEEE Systems, Man and Cybernetics Magazine - October 2019

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - October 2019