Artificial Intelligence in Text Searching

One product feature often emphasized by vendors of applicant tracking software is text searching enhanced by artificial intelligence. The terminology varies, but generally it is called "concept searching," "advanced searching," "neural search" or sometimes "expert system search".

The systems are intended to outperform basic keyword searching via the use of inference rules, comparison engines, learning algorithms, synonym tracking and other methods designed to generate more "accurate" search results. In marketing the feature, the claim of superior search results is usually paired with an assertion that standard keyword searching is ineffective and outmoded for high-volume text searches.

The critical issue Main Sequence Technologies sees with artificial intelligence in selection is the currently unsolvable complexity arising from the unfolding of the social context and human will in staffing. To expect current artificial intelligence technology to make those connections and then act on them without human verification would be optimistic, to say the least.

A carefully evolved, trained, and designed system may approach an impressive percentage of accuracy, yet, like a clock that tells the incorrect time on every tenth reading, it will require cross-checking each time it is used. The timesavings from its use are substantially reduced and, in the real world, users of both smart and smartly enabled keyword searching seem to spend about the same time to achieve very similar results.

In addition, some of the functionality of the artificial intelligence involves comparison engines. A common query presented to these systems gives the user the option to " find resumes like this one". While this may seem like a useful tool if you have a resume that fits a particular need in mind, it might be recalled that asking for sameness is the literal antithesis of diversity. If such systems are inadvertently (or intentionally) allowed to "type-form" recruiting for given positions (by geography, salary range, education, grammar, idiom, syntax, etc.) or other elements not related to bona fide occupational qualifications, liability could arise. Even the appearance or potential of disparate impact might negate any timesavings gained or “quality” of talent obtained.

Main Sequence has one of the most diverse client bases in the industry: across all staffing models and types of organizations; this helps us to understand what people want from many dimensions. Our engineering model is based on providing a cost controlled, self-contained deployment with Microsoft technology. Main Sequence recognizes the need to constantly explore and understand emerging technologies as we contribute to our partnerships with our clients. We are constantly evaluating vendor and supplier offerings, development in other industries, and macro technology trends in our ongoing efforts to enhance PCRecruiter's design. Agile software development ruthlessly reveals the costs and benefits of various features. Successful "agile" software, by definition, implements someone's best methods for a given process.

Main Sequence has not had sufficient enough customer demand to implement artificial intelligence text search, and we do not find the currently available performance, costs and knowledge base to make a compelling case for inclusion into PCRecruiter.

Main Sequence believes that in the arena of employee selection, inference should usually be left to human agents. Automated decision support tools can help professionals to make rapid, informed judgments where required, and most importantly, to help them take effective action on those judgments. But the tools should be objective in use and effect, and in today’s legal environment, it can only be a defensive asset that any tools used be understandable and demonstrably objective.

PCRecruiter's unique searching competencies have focused on enhancing the ability of users to make the essential inferences and decisions that lead to search success, achievement of diversity goals, and quick identification of potentially high-performers.

>> Staffing, and recruitment in particular, is aimed at discovery and exploration of the attributes of an individual person, a role, and the requirements that put them together. When a user executes a 'basic" keyword search, they have command over all search variables. Often, subtle adjustments of the search terms can reveal results that users are seeking. However, when artificial intelligence is invoked, the user no longer has direct control of the search variables, and is sometimes unable to determine precisely what result changes occur with each alteration of search variables. Yes, you could turn the AI on and off, but why carry the costs and risks to de-select half the time?

>> Context aids decision making and accurate contextual decisions support better outcomes. Artificial intelligence has difficulty with context. The whole idea of a resume is to provide controlled context- a limited universe of information bound in particular arrays. For instance, the name of an organization appearing on a resume (if it's not a school or training institution) is most often a former employer, or a customer that the individual worked with; important information in multiple contexts that skilled recruiters connect to make things happen.

Its existence on a resume provides some of that context. Current artificial intelligence technology has a difficult time in discerning the actual relationships of individuals and organizations, which is of critical importance in staffing.

PCRecruiter goes farther than most systems in establishing context for users. The PCRecruiter search engine will reveal the location of the search term in the person's record (e.g. in a recruiter's notes, resume, profile documents, and elsewhere). When a recruiter sees a keyword in the notes section of a candidate's record, they can be informed that the keyword applies to that individual merely on the context of the search hit; likewise for digital job applications, data sheets, questionnaires, and documents of all kinds.

>> PCRecruiter offers keyword-comparisons to help users expand their searches. The system can examine the results of a user's query and provide a raw count of additional words commonly found in records matching the original search criteria. This tool helps users to identify terms that they did not ask for, but which regularly appear in records containing the terms that they did ask for. This feature makes no determination or inferences on its own, but rather, supplies the user with additional information that might help them to expand their result set.

>> PCRecruiter has a very important feature in Rollup Lists. This feature allows users to group and ungroup records in any combination that they wish, for as long as required. By filtering the PCRecruiter search engine to certain lists, tremendous operational flexibility can be developed and the search engine can be thought of as extending to past results.

>> One excellent use of artificial intelligence is in the extraction of demographic and contact information about individuals from their resume documents. These decisions carry no legal weight, but do save considerable administrative resources in creating accurate database records. Some implementations of artificial intelligence and applicant tracking technology really aim no higher than accurate extraction of this kind of data. The PCRecruiter Resume Inhaler offers tremendous performance in this area and is one of the most accurate automated resume processors on the market. The Inhaler is more configurable than any competing product, allowing users to manually enter new filters and rules at will for improved resume format recognition.

>> This leads to another practical consideration for artificial intelligence text search and applicant tracking software; it tends to increase the cost basis of the vendor, complicates installation, reduces integration and development speed, and provides potential technical, updating, and portability issues due to the integration of the technology. In the case of third-party vendors, it creates the attendant complications of multiple or dependent licensing and potential variations in upgrade paths and lifecycles.

>> PCRecruiter maintains its searching index directly within its database structure. This enables much easier portability, integration, maintenance and control, which in turn lowers costs and increases user satisfaction. Because PCRecruiter stores resumes and other documents directly within the database as well as the search index, many artificial intelligence text search options remain open to users who want to explore their data with specialized tools. PCRecruiter has a high quality engine, but there are many focused text engines and approaches being developed that may offer value for some users.

>> Nearly all of the leading search products support Microsoft database formats directly. In addition, linking records within PCRecruiter to a Rollup List is relatively easy, even for lightly experienced data analysts. Self- integrating any number of 'best of breed' search tools over the lifecycle of PCRecruiter has proven to be practical, and it will extend the value of a PCRecruiter investment in the unlikely case that Main Sequence would ignore the establishment of artificial intelligence text searching as a leading practice in the staffing world.

The performance advantage of current artificial intelligence text search is not anecdotally better than PCRecruiter's combined feature set and no peer-reviewed sciences or standards groups have yet asserted that artificial intelligence text searching is more effective. Our own testing and experience with clients, demonstrations, and comparisons has shown that PCRecruiter easily and reliably produces effective result sets.

Forward thinking legal analysts have identified potential hidden costs in the use of Artificial Intelligence for both software vendors and customer/ users, as these may arise from establishment of liability for the decisions and recommendations made by computer systems across the economy.

There have been lawsuits and legal controversies in such areas as banking (redlining, equal opportunity lending), insurance (rating), and medicine (predicting death rates for resource allocation), where machines decisions have been factored into actionable outcomes. In these cases, third parties were sometimes involved.

Of course you can be sued for getting out of bed in the morning, but product liability and employment practice litigation is an everyday occurrence; companies like Microsoft are sued regularly under a variety of legal theories, along with many other household name vendors, and this action may travel down through the legal ecosystem someday.

Main Sequence is constantly evaluating costs and benefits of supplying features to customers, and we think the state-of-the-art today supports this measured approach to the question of artificial intelligence in recruitment selection.