Artificial Intelligence enters into the field of construction agreements.
As any astute bidmeister who plunges into the murky waters of specialty contract review will attest, the world of construction agreements is enough to make one’s head spin like a dervish dancer on crack.
Granted, much of the structured terms in the body of the contract are translated into legalese by the general contractor’s team. There, the “professionals” dwell for hours on lofty subjects like indemnification, waiver of subrogation, additional insured endorsements and a myriad of other vaguely defined conditions and clauses that only a weasel-wording pettifogger (or a word fanatic, such as myself) could possibly fathom. But the meat and potatoes of the agreement—actual values, sources, scope of work, schedules, inclusions, exclusions and clarifications—are more frequently left for the foot soldiers in the trenches, estimators and project managers, to wrestle with.
Unfortunately, even the composition of a mere bid abstract, which is a condensed version of the subcontract itself, is a ponderous endeavor for even the most astute estimator/project manager, often requiring multiple pages of development and many hours of careful review—precious hours that could otherwise be spent honing the calculations of the estimate proper.
Fortunately, a development has emerged that can trim those manifold pages and transform those hours into minutes. Unsurprisingly, that development has come with the advent of artificial intelligence (AI). Recently developed contract review programs enjoy optimal efficiency using natural language processing (NLP). Among other benefits like speed and accuracy, NLP recognizes risk clauses and unfavorable terms in contract language. NLP is a subfield of AI and linguistics that allows computers to “understand,” interpret and even generate human text and speech.
It combines computational linguistics—i.e., rule-based modeling of language—with statistical machine learning. Machine learning is the engine driving most NLP programs. These systems use algorithms to analyze language structure, context and sentiment. The upshot of all this techno-jargon is that NLP converts unstructured human language into data that machines can process into tasks—like document search, which is what contract review essentially consists of.
Given the advances that AI has generated in the field of language recognition, it is not inconceivable that a “smart” program might be able to confidently review a subcontract, but also act as a “co-pilot” in composing a comprehensive bid proposal, with an astute estimator filling in the blanks with values, sources, scope of work, etc.
In short, a well-crafted program could remove a lot of the daily drudgery and finally allow bidmeisters to focus on their main task of estimating.
However, as is true with most miraculous solutions, there are some serious obstacles to this remedy. Not surprisingly, the first is cost. While the long-term benefits of these new AI generated programs are becoming apparent, the prohibitive costs of initial experimentation are better suited for general contractors and design professionals who enjoy the efficiencies more readily. Clearly, the paybacks for subcontractors will be less direct and more evident as they develop.
But cost is actually a secondary apprehension for the field of contract evaluation with regards to AI. A more troubling concern has to do with basic reliability. As programs become data-driven (and thereby more efficient), they inevitably generate operational flaws, which in turn produce contractual and legal risks. These risks are latent by nature and subject to exposure.
Such concerns begin with data reliability. Machine-learning programs using NLP for contract authorship and review are completely dependent on the data they reflect. But generative data can be outdated or incomplete, resulting in false or misleading information. Moreover, critical context is frequently missed by AI-driven programs, which are limited in their ability to interpret subtleties in contract language, such as inference or implication.
Programs can therefore make assumptions that are not clearly stated or supported. Then, to further complicate things, AI programs are instilled with a positive confidence sort of instinct that tends to make inaccurate information difficult to detect.
My personal feelings on this specific subject are mixed. While I am fascinated with the advances and the potential that AI in general can present to the construction industry, I remain skeptical that the developments are complete and completely positive. Regarding contract review, the concerns about accuracy are troubling, and if the information cannot be relied upon, then a redundant second review by a human would render any gains in efficiency null and void. So, what’s the point?
But what contributes more deeply to my ambivalence is the underlying apprehension that such tasks in our field will eventually render our vocations obsolete. In other words, is contract review just the beginning of the end of estimating? It’s a question that cannot be easily dismissed.
Vince Bailey is an estimator/project manager in the Phoenix area.
