ACTIVATING ENTERPRISE PROCESSES WITH HUMAN - AI TEAM
Alpana
Dubey
We have seen great advancements in the adoption of AI system
in the last couple of decades. Improvements in AI technology has changed our perception
about what can be achieved by computing machinery. Even though AI has been
largely impacting the way we work, one of the common relationships between
human and AI system, that has been leveraged till date, is using AI system as a
“tool” to perform certain task not much as a “partner” to work together to
solve a common goal by augmenting each other.
This view has largely influenced AI research to push towards achieving
highest level of automation through AI system by enhancing AI systems with more
data, better algorithms, and better computing power. These efforts have brought
forth several new approaches and better systems. However, one important aspect
that has not been well leveraged in this journey is the critical role human
could play to boost the performance of Human-AI system even with the current
set of technologies [2, 3].
A symbiosis between Human and AI systems can result in much efficient
and effective system than the human or the AI alone. One of the common reasons
for this is that human and AI bring unique and complementary strengths. This
phenomenon has been also expressed in the detail explanation of Moravec’s paradox
[4, 5]. For instance, Human are good at reasoning with smaller number of
observations, but AI system needs a large amount of data to provide the same
level of reasoning. Human cannot process large amount of data at speed for
interpretation whereas AI system are good at generating interpretation from
large data. Humans exhibit conformation bias while interpreting data whereas AI
system can be designed to be un-biased by appropriate un-biasing algorithms.
Humans are good at dealing with novel unfamiliar and dynamically changing
situations whereas AI system are not good at handling new situations or
adopting to a new situation. Humans are good at setting goals and hypothesis.
AI systems are good at executing the goals once the rules of goal planning are
established. A set of interesting examples of human and AI complementing each
other can be seen in the Hollywood movie Upgrade released in 2018 where humans
are transplanted with AI chips to complement their strength.
Considering
the orthogonal strengths human and AI brings, this is high time to view and
model business processes as Human-AI team. Paul et al. in their book
“Human+Machine: Reimagining Work in the Age of AI” [1] have touched upon
various relationships between Human and AI and how they can be leveraged by the
organizations for AI based transformation. In this article, we discuss the design
principles that one must take in account while designing a Human-AI teaming
application.
Example
of Human-AI Team in Business Process
We draw a small example from service industry as shown in above Figure 1 to understand a Human-AI team. Consider a service engineer
troubleshooting a machine. We can imagine several AI systems augmenting the service engineer while (s)he is
troubleshooting the machine. For instance, an AI system, namely “Knowledge
retrieval Assistant” which is based on information retrieval and natural
language processing (NLP) can retrieve the troubleshooting instructions and
related help documentation about the machine. As these documents may be huge
and large in number, fetching the relevant sections from the documents is going
to be very time consuming for the service engineer whereas computer algorithms
can do this at much faster rate. At the same time how to use all the
information together to really fix the machine is something where service
engineer could apply his creativity. There may be some situations where the
service engineer may not be able to fix the machine with current set of
instructions. Here, another system, namely “Expert Assistant”, helps in finding
the experts who can provide support to the service engineer while he is
troubleshooting and fixing the machine. Here, we see that two AI assistant are
amplifying the service engineer capabilities by providing relevant details and
connecting with experts and human is finally solving the issue.
Properties of Effective Teaming
Imagine that the AI assistants in the previous examples
are not only acting as mere tools; but they are intelligent entities which have
visibility of what actions other AI assistants are performing, what actions the
service engineer has taken, the type of machine (s)he is fixing, what (s)he is
trying to fix, what kind of fixes (s)he has made in the past, and what are
his/her observations from the machine. In addition, consider the assistants
also understand the intention of the service engineer and status of the
machine. All this information would make AI assistants smarter and intelligent
because with this information they can proactively support the engineer based
on the current context and the goal. This kind of teaming intelligence is of
utmost importance for effective Human-AI teaming where individual entities are
“context aware”, “goal aware”, “self-aware”, and “proactive”. In addition, for
effective teaming, interactions among the entities need to be fluid either with
a proper communication language and or with appropriate GUIs. We have developed
a taxonomy of Human-AI teaming concepts by combining concepts available in literature
as shown in Figure 2. The taxonomy is
broadly developed around 4 key dimensions and 15 sub-dimensions. While
designing Human-AI teaming application, one needs to consider these dimensions and
assess whether the essential characteristics are met in the design.
Steps to realize Human-AI Team in Business Process
Successfully realizing Human-AI team in a business process requires
a thorough analysis of the process. The key steps that one needs to pursue are
Analyze, Model and Develop / Deploy as shown in the side bar.
Once these steps are executed, one need to use his/her favorite
implementation platform. However, the current platforms do not support all the
design elements that we have discussed so far. These concepts have been
discussed at an abstract level in the literature and lack a proper guidance on
how these can be realized concretely.
HACO
Platform
Through our research efforts at Accenture Labs, we are
advancing approaches for software application development to transform enterprise
processes with more teaming intelligence. Our platform namely Human AI Collaboration
Platform (HACO) helps in realizing the essential properties a Human-AI teaming
application must have. The key features and the workflow involved in the
implementing Human-AI teaming application is shown in Figure 3. A short video
about the platform can be access at https://youtu.be/lNyrrk8dMqU
A research paper, based on this work, will be published in the proceedings of 13th Innovations in Software Engineering Conference ISEC 2020, to be held on Feb 27th – 29th, 2020.
References:
[1] Human + Machine: Reimagining Work in the Age of AI. Paul R. Daugherty, H. James Wilson.
[2] https://beta.techcrunch.com/2016/04/12/the-era-of-ai-human-hybrid-intelligence/?_ga=2.184699881.13417694.1526480637-1368337914.1526480637
[3] https://techcrunch.com/2016/11/01/how-combined-human-and-computer-intelligence-will-redefine-jobs/
[4] Minsky, Marvin (1986), The Society of Mind, Simon and Schuster, p. 29
[5] Hans Moravec, Mind Children: The Future of Robot and Human Intelligence, p. 15.
[6] HACO: A Framework for Developing Human-AI Teaming. Accepted at 13th Innovations in Software Engineering Conference (ISEC 2020)
About
the Author
Alpana is a senior researcher at Accenture Labs,
Bangalore. She leads Digital Experiences R&D group at Bangalore. Alpana has
close to 14 years of experience working in industrial research labs. During her
research career, she started new research areas and drove company-wide adoption
of software engineering best practices. She has several patents, has published
over two dozen research papers in scientific journals, and peer reviewed
technical conferences of repute in the area of software engineering, crowdsourcing,
static analysis, reverse engineering, and pattern mining. Before joining
Accenture, she worked with research labs of ABB, Siemens, and Philips. She is a
senior member of IEEE and received her PhD in Computer Science and Engineering
from IIT Kanpur (India) in 2006.