Engelmore Award Article
B usin ess a n d M a n ufacturin g
Banking and Finance
Transp ortatio n
M ilit a r y
US G overn m ent
E n erg y
A uto m obiles an d Trucks
Teleco m m u nications
Leisure and Entertain m ent
M e dicine
M e dia
Figure 3. Top 15 Industry Topics in IAAI Articles, 1989–2016.
Ex pert Syste m s
M achine Learnin g
N atural La n g u a g e
Planning and Scheduling
Statistical Learnin g
U n c ertain ty
C o nstraint-B ase d R e as o nin g
V isio n
Figure 4. Top 15 Technology Topics in IAAI Articles, 1989–2016.
data by transformation into structured data. The platform automatically tags documents with accurate
and consistent metadata, guided and enriched by
subject matter expertise. The figures show data from
deployed applications only — 316 of them. The figures include only 15 of a long tail of more than 100
industry and technology topics that have been covered in IAAI. Figure 3 also excludes information technology applications, that is, AI applied to our own
business, which would otherwise be number one.
Another way of analyzing IAAI topics over the
years is shown in figure 5. The figure shows that the
technology mix has evolved. Expert systems clearly
dominated the early days of IAAI. Machine learning
is notably absent early on. Over time, however, the
mixture has become more diverse, with no topic
clearly dominating in recent conferences. We note
that it is not the case that expert systems died.
Rather, after a few years, they became more standard
practice than innovation. Fewer papers were pub-
lished about novel applications of expert systems.
They disappeared into the fabric, now applied every-
where, from the high-end emulation of rare human
experts, to the embedding and application of rule
books and procedure manuals. However, we will like-
ly see more hybrid machine-learning technologies
that can automatically update their reasoning
engines as the application data change over time.
Some technologies have not been represented
much at IAAI, like speech understanding and robots.
They do appear, just not in the top 15. In the recent
2016 deployed papers track, four technologies are
applied: spatial reasoning, crowdsourcing, machine
learning, and ontologies. We also note that there have
been two papers on deep learning, one in 2015 and
one in 2016, neither documenting a deployed appli-
cation. Some of this may be due to self-selection in
that our data are limited to IAAI conferences, which
may not accurately reflect how often these technolo-
gies are utilized in the overall application world.
In our final analysis of IAAI articles, figure 6 shows
a quick overview of the top concepts mentioned over
the years. The analysis was done with a modified
form of the C-value/NC-value method (Frantzi, Ananiadou, and Mima 2000), which extracts significant
concept names found in text, as opposed to just the
most frequently used phrases. Note that there may be
some temporal bias in this analysis due to the data
set reflecting the past decades of IAAI papers, versus
trends in the most recent papers.
High-Impact AI Applications
Many of the past IAAI program chairs and cochairs
and AAAI Fellows kindly responded to a request for
their views on what have been the high-impact applications, including some that opened up a new area,
presented at IAAI conferences over the years.
Because we have selected high-impact applications
and it takes time to establish whether an application
has had high impact, some of the examples may look
a bit dated. Note, however, that in several cases, a
recent update has been presented at IAAI.
A few of the applications that were singled out by
several respondents as being high impact are summarized in the following.
1983: Process Diagnosis System (PDS)
The Process Diagnosis System (Fox, Lowenfeld, and
Kleinosky 1983) started out as an expert system shell.
It has been in active use and continuous development
since 1985. 1985! Though the origin of PDS predates
IAAI, it serves as an early example of deployed AI. It
started with a presentation by Mark Fox at Westinghouse. Over the 30-year period, Westinghouse sold
the business to Siemens, where it is now at the heart
of their Power Diagnostics Center that performs centralized rule-based monitoring of over 1200 gas tur-