reports of large cash transactions, the FinCEN Artificial Intelligence System (FAIS) used link diagrams to
support detection of money laundering (Senator et
al. 1995). ADS: NASD Regulation Advanced-Detection System (Kirkland et al. 1999) used temporal
sequences to support detection of securities fraud.
Their different domains of use dictated different
FAIS links and evaluates reports of large cash transactions. To give an idea of the money involved, fin-
cen.gov reported suspicious transactions totaling
approximately $28 billion in October 2015. The FAIS
key idea is “connecting the dots” — thus link diagrams, now commonplace in social network analysis, was an appropriate choice. The appropriate representation choice in FAIS enabled a reporting app
based on the original detection system. This was an
ADS monitors trades and quotations in the Nasdaq
Stock Market, to identify suspicious patterns and
practices. In this application, temporal sequences are
key — not so much links as in FAIS — so a representation that supports them was a good choice.
2005 and 2014: Engineering Works
Scheduling for Hong Kong’s Rail Network
The Hong Kong rail network moves 5 million passengers a day through the city’s rapid transit subway,
airport express, and commuter rail lines. The AI
application streamlines the planning, scheduling,
and rescheduling process and provides automatic
detection of potential conflicts as work requests are
entered; verification that no conflicts exist in any
approved work schedules before execution; generation and optimization of weekly operational schedules; automatic update to repair schedules after
changes; and generation of quarterly schedules for
planning (Chun et al. 2005, Chun and Suen 2014).
To be successful, the system must coordinate with
the staff members who carry out the scheduled work.
To this end, the developers found that the system
must be able to explain the schedules it creates. As a
result, they veered away from the original genetic
algorithms approach toward heuristic search. A
recent report about this system appeared in New Scientist (Hodson 2014).
1994 and 2004: Plastics
Color Formulation Tool
Since 1994, GE Plastics (later SABIC) has employed a
case-based reasoning (CBR) tool that determines color formulas that match requested colors (Cheetham
2004). Form Tool has saved millions of dollars in productivity and material (that is, colorant) costs. It is
the basis for the online color-selection service called
Determining the colorants and loading levels that
can be added so the plastic matches a given color is a
difficult problem for multiple reasons. For example,
there is no accurate method to predict the color pro-
duced when a set of colorants is added to plastic.
Unlike paint, where light primarily reflects off the
surface, in plastics a significant percentage of light
penetrates the surface and reacts with the internal
structure to produce a color that depends on both the
internal structure and the lighting conditions (natur-
al sunlight versus fluorescent lighting).
The AI system used case-based reasoning to replace
programs that used prohibitively expensive exhaustive search to determine the colorant-loading proportions for a color formula that matches a customer’s desired color.
1995: Scheduling of
Port of Singapore Authority
This expert system (Weng et al. 1995) is responsible
for assisting with planning and management of all
operations of the Port of Singapore Authority. With
hundreds of vessels calling at Singapore every day, a
fast and efficient allocation of marine resources to
assist the vessels in navigating in the port waters is
essential. Manual planning using pen and paper was
erroneous, uncoordinated, and slow in coping with
the rapid increase in the vessel traffic. Included in the
purview of the application is scheduling the movement of vessels through channels to terminals,
deploying pilots to tugs and launches, allocating
berths and anchorages to ships, and planning
stowage of containers.
To generate accurate, executable deployment
schedules, the automated scheduler requires real-time feedback from the resources on their job status,
any estimated delays, and end times of their jobs.
This is achieved by integrating the system with the
port’s mobile radio data terminal system.
2006: Expressive Commerce
and Its Application to Sourcing
This application has produced one of the largest ROI
figures of any system thus far reported at IAAI. Originally CombineNet, later renamed SciQuest, it
improves procurement decisions for spend categories
that are typically beyond the capabilities of traditional eSourcing software. Even in the early days of
2006, it had already handled $35 billion in auctions
and delivered $4.4 billion in savings to customers
through lower sourcing costs (Sandholm 2007).
The challenge in developing an expressive commerce system is handling the combinatorial explosion of possible allocations of businesses to suppliers.
Their key development is a sophisticated tree search
algorithm. Much has been written about this algorithm (refer to Sandholm  for a list of articles),
though some of its details are kept proprietary.
CiteSeerX (Wu et al. 2014) is a database and search
engine for more than 4 million research articles from