Artificial Intelligence: Due Diligence 3.0

The promise of Artificial Intelligence in M&A — performing extensive and reliable due diligence searches rapidly — will change corporate law in ways we can’t yet predict.

[This is an excerpt from a paper written for the Osgoode Professional LLM Corporate Transactions course.]

GARRY KASPAROV WROTE,
“Playing chess, I learned the dramatic effect combining humans and machines. Humans have intuition, can recognise patterns and positions, and machines have brute-force of calculation and memory. By bringing these capabilities together in other walks of life, we can achieve incredible results.” One of these results is Artificial Intelligence (AI) in M&A due diligence.

Kasparov has credibility, or perhaps notoriety, in this regard. In his “Waking Up” podcast, philosopher and neuroscientist Sam Harris reminded Kasparov, “You will go down in history as the first person to be beaten by a machine in an intellectual pursuit where you were the most advanced member of our species.”1 Kasparov points out that he actually beat
IBM’s Deep Blue in 1996, though he lost to it a year later. Kasparov wanted a so-called “rubber match” to decide the matter forever, which did not happen. Just like in M&A: once the deal is completed, a rematch is nearly impossible.

Although not quite synonymous, the terms
AI and cognitive computing are used interchangeably herein if other authors did so. By either term, we are considering the following: “Modeled after human learning, smart machines process massive data, identifying patterns. These patterns are used to ‘create’ entirely new patterns, allowing machines to test hypotheses and find solutions unknown to the original programmers.”2

Voluminous due diligence was not always so influential in dealmaking. “The art and science of merger execution have made great strides since the late 1990s
a period when stock-market frenzy often led to a rush to judgment, and ultimately to buyer’s remorse. Since then, a more prudent, systematic approach to mergers and acquisitions has emerged,”3 and there is much discussion of strategic due diligence.

Add to the mix the volumes of correspondence data and other data generated within corporations. It can be said that
AI is a remedy to the build-up of this data. It’s a computerized response to a computerized problem. “AI is an overarching term that includes many branches and sub-sets of technologies. However, the most common forms of AI in the legal tech sector are Machine Learning, Deep Learning, and Natural Language Processing.”4

Deep Learning is used by a computer “to understand legal language, then compare the language with other contracts to identify boilerplate vs. custom, measure the complexity and readability of the language, and identify the responsibilities, rights, and terms of an agreement.”5

Natural Language Processing is “needed for data mining of external ‘big data’ sources and for addressing the legacy contract encoding problem. The latter is the biggest short-term problem, especially if [the law firm or department] has grown through acquisition or has not been rigorous in establishing a standardized clause library. But there are contract analytics providers (tools and/or services) that can help depending on your situation. It’s a thorny issue because of the dense ‘legalese’ written in various languages (which have multiple semantic and legal interpretations) that must be classified and mined to extract the atomic-level insights about obligations, rights and risks of interest. This
NLP of ‘natural’ commercial language can’t be scalably addressed by human coded, rules-based approaches.”6











SIDEBAR: Are AI tools actually being used in Canadian M&A due diligence?

Machine Learning “refers to computers that ‘learn’ from the data they process rather than relying on humans for rules-based procedural programming to act upon that data. It not only discovers patterns in data but also specifically helps correlate various data inputs and key data outputs, which helps enable predictive analytics.”7

But humans are still needed to take machine learning forward. “In a ‘supervised learning’ approach, human experts determine the outputs and the system ‘learns’ how to mimic the human experts, as well as uncover latent variables and interactions that humans wouldn’t have spotted on their own. ‘Unsupervised learning’ doesn’t rely on humans for direct training and stretches into the realm of deep learning.”8

For all of the promise of AI, one of things inhibiting it is the language barrier between lawyers and computers. This is changing as avant-garde lawyers decide to learn coding (while other lawyers dismiss it, much in the way that, in an earlier epoch, typing was discounted as “clerical” by some male lawyers).

Certain law schools, such as Georgetown, offer coding courses. Harvard Law explains its Programming course: “Law firms must understand what tasks can be most efficiently done by custom software and what are best left to human beings.”9

But when it comes to learning more about computer language, there is arguably a level of fear holding lawyers back, a worry about being replaced or at least eclipsed. For senior lawyers who do not themselves execute on due diligence anymore, they may be content to have juniors continue to do it as a learning exercise. But what about those students and associates who are executing on manual due diligence? What do they think about it?

Noah Waisberg left his early career in Big Law in the
US to establish an AI due diligence company. Zach Abramowitz, who also left BigLaw, has a blog-casting platform, ReplyAll. The two of them dialogue about the boredom, even resentment, they say that newer lawyers express. Abramowitz categorizes the scenarios that give rise to errors in human-conducted due diligence:

“Timing
Lots of people have stories about Friday afternoon partner staffing calls, trying to avoid picking up the phone, still getting caught, and having a ruined weekend. Better versions of this story include a partner knowing about the diligence earlier in the week, but forgetting to staff anyone till Friday.

“Misses
Pretty much every Biglaw midlevel M&A associate through junior partner has a time when they realized their juniors systematically missed finding things they were supposed to. Better (most?) versions of this story involve realizing the diligence is not accurate very late night or right before the weekend, and having to redo it all.

“New Categories
The team finishes a multi-week diligence project, then the deal structure changes. It turns out the original review missed a now-critical category.”

Abramowitz then asks, “Okay, but aside from lazy, inexperienced or depressed first years, what are the structural or hierarchical problems with how firms are doing due diligence? Waisberg replies: “The junior lawyers who do most of the review don’t actually know what they’re looking for. As smart and well-educated as they may be, and even though their firms may have fancy training programs, it takes practice to spot change of control clauses/exclusivity/non-compete
/MFN. This is the systematic component of how mistakes get made in due diligence. The random component of errors is that diligence is often done by tired, distracted, rushed lawyers. Surprisingly, doing diligence at 2 am night after night does not improve quality.”

Many associates are in a certain negative mood about the efficacy of manual due diligence. Lawyers, being human, get tired and cranky, with unfortunate implications for voluminous due diligence in
M&A. Slaughter & May’s Steve Cooke says AI in M&A due diligence “will give junior lawyers … their lives back.”

But what will lawyers do with that newfound time? A Forbes magazine testimonial on the Ross website has it that “… lawyers will be able to avoid some of the mundane tasks…”10 “By automating labor-intensive, low-value tasks, artificial intelligence systems free up lawyers and other legal professionals to concentrate on complex, high-value projects. But when between 13 and 23 percent of the average lawyer’s time could be automated, the financial impact on the industry is an open question.”11


What is AI good at? AI is useful in M&A due diligence for that which (1) is routine, (2) voluminous, (3) needs customization, or (4) is nearly invisible and possibly fraudulent.

Learn the reasons why hire a lawyer for mergers & acquisitions due diligence with this article.

As Waisberg said, “Since lawyers were looking for the same provisions so frequently (e.g., change of control, assignment, exclusivity), I thought it might be possible to build software to help find this info.”12 This is the “quick win” of AI: it can simply take in more volume on routine functions than a human can. It is the voluminous work of reviewing hundreds or, indeed, thousands of documents of the same type that holds out the most accepted potential for AI.

At the next level,
AI is highly susceptible to customization. For example, take Brexit: “Brexit Contract Review Solution is a new collaboration between NextLaw Labs (Dentons law firm’s legal tech investment vehicle) and RAVN Systems. The solution lever­ages RAVN’s AI technology and a bespoke algorithm co-developed with Dentons’ subject matter experts to enable high-volume contract review to pinpoint provisions that the UK secession may impact.”13

In discussing the development of their company NexLP, which uses
AI to analyze data and identify trends, co-founder Jay Leib describes the questions he and his partner, Dan Roth, started to consider in the late 1990s:

“What is the future of the industry? We thought about it,” he says. “There were whistleblowers in their companies who knew what was going on, and the unstructured data contained the stories. Companies could detect potential problems early on, provide alternatives to counsel and the C-suite, and understand their exposure. It would prevent unnecessary legal spend and mitigate risk, thus protecting the company’s brand and shareholder value.”14

“Nearly 80 percent of a company’s data is unstructured,” Leib says. “While unstructured data represents the lion’s share of a company’s data, for years lawyers have been stuck with antiquated tools that focus primarily or solely on Boolean search. Better tools are needed to truly understand data, infer meaning, classify the various types of ideas present, and help you get to the result fast
even if that result didn’t involve the keywords you used.”15

Document reviewers have always tried to detect what a target company might be concealing, but now vendors of
AI tools are developing “technology that can turn information into stories. Story Engine is a program that can read through unstructured data and summarize conversations, including the ideas discussed, the frequency of the communications and the mood of the speakers.”16

This is not to say that computers can feel the mood of speakers as humans to varying degrees can, but computers can ingest more unstructured data than we can and, within that, recognize patterns. Then enter the humans: corporate leaders will ultimately still make the decision to pursue deals are not. And they will continue to be greatly assisted by internal and external counsel when it comes to strategy. It takes wisdom to discern, “What does it all mean?”

So is this what humans do better? Discern meaning in a deal, using judgment? In Algorithms to Live By: The Computer Science of Human Decisions, Brian Christian and Tom Griffiths explore the ways in which humans can combine “computer algorithms” with human qualities in order to make decisions. They offer the oft-told anecdote about Charles Darwin composing a “pro and con” list to answer the question, for himself, as to whether or not he should marry his cousin Emma Wedgwood. Based on a “narrow margin of victory,” Darwin concluded, “Marry … Q.E.D.17

Christian and Griffiths continue, explaining that it was Benjamin Franklin, before Darwin, who devised and praised “Moral or Prudential Algebra,” in which the more factors considered the better.18 Not so in the present age, according to Christian and Griffiths: “The question of how hard to think, and how many factors to consider, is at the heart of a knotty problem that statisticians and machine-learning researchers call ‘overfitting.’ And dealing with that problem reveals that there’s a wisdom to deliberately thinking less. Being aware of overfitting changes how we should approach the market …”19

But Darwin proved his decision, didn’t he? And in this paper, we have been praising all this data that computers can process for the benefit of
M&A. Why are we now worrying about overfitting? First, Darwin likely approached his mathematical calculation predisposed to marriage. So too the leaders of an acquiring company, generally speaking, want to acquire the target, or a target, and therefore, want the due diligence to pan out. And secondly, as Christian and Griffiths write, “Every decision is a kind of prediction … and every prediction, crucially, involves thinking about two distinct things: what you know and what you don’t. … A good theory, of course, will do both. But the fact that every prediction must in effect pull double duty creates a certain unavoidable tension.”20

That tension sounds very much like the M&A context, in which corporate leaders are asking, ‘Knowing what we know based on due diligence, if we marry, will it be a happy, qua prosperous, corporate marriage? Will the pros outweigh the cons in the future?’ After all, cons revealed in due diligence, namely defects, could be remedied by new corporate leadership. And pros can be eclipsed. Judgment still needs to be exercised. Since computers are highly susceptible to overfitting, human lawyers would do well to take advantage of AI, then reduce and curate the data into forward-looking strategic advice.

“No two deals are alike and each merger or acquisition depends on a multitude of factors. Some of these factors will include obvious elements such as company balance sheets, employee compensation, share allocation as well as specific factors related to the IP, or assets in question. The entire process is extremely detailed, laborious and can run from months to years depending on the size and complexity of a deal.”21

Let us return to Garry Kasparov as we draw our conclusion: “Whilst machines are taking over more parts of our lives, and people say this is killing many jobs, we have to realise this has been happening for thousands of years. Machines replaced farm animals, then manual labour, and now they’re taking over jobs from people with college degrees and twitter accounts
and everyone is making a big noise. Replacing manual labour allowed humanity to concentrate on developing our minds, and now, perhaps by taking over more menial aspects of our cognition, machines will help us to look for greater creativity, curiosity and happiness.”22

ENDNOTES


1 Frederic Friedel, http://en.chessbase.com/post/kasparov-on-the-future-of-artificial-intelligence
2 Julie Sobowale, http://www.abajournal.com/magazine/article/how_artificial_intelligence_is_transforming_the_legal_profession, p. 1.
3 Adolph, Gerald and Simon Gillies and Joerg Krings, “Strategic Due Diligence: A Foundation for M&A Success” at https://www.strategy-business.com/article/enews092806?gko=21dd3, para. 1.
4 Outsell Company Analysis, “Advancing the Business of Data & Information”, April 27, 2017 AI Legal Tech conference paper.
5 https://www.legalrobot.com/
6 Pierre Mitchell, http://spendmatters.com/2017/01/10/artificial-intelligence-contract-management-part-4-natural-language-processing-machine-learning/
7 Ibid.
8 Ibid.
9 http://hls.harvard.edu/academics/curriculum/catalog/default.aspx?o=71516
10 http://www.rossintelligence.com/
11 Alston Ghafourifar https://venturebeat.com/2017/04/12/will-ai-powered-robot-lawyers-still-use-cheesy-billboard-ads/
12 Zach Abramowitz, http://abovethelaw.com/2014/09/why-diligence-sucks-why-you-suck-at-it-and-why-robots-want-your-job/
13 For more on the voluminous contracts that need changing with Brexit, see: https://www.ft.com/content/f1435a8e-372b-11e7-bce4-9023f8c0fd2e.
14 Sobowale, Note 2, p. 1.
15 Sobowale, Note 2, p. 2.
16 Sobowale, Note 2, p. 2.
17 Christian, Brian and Tom Griffiths. Algorithms to Live By: The Computer Science of Human Decisions (New York: Penguin, 2017), p. 150
18 Christian and Griffiths, Note 16, p. 151.
19 Christian and Griffiths, Note 16, p. 151.
20 Christian and Griffiths, Note 16, p. 152.
21 http://mergertechnology.com/security/artificial-intelligence-merger-acquisition-due-diligence-3662.
22 Vikas Shah, at https://thoughteconomics.com/garry-kasparov-interview/, para. 17.