AustLII Home | Databases | WorldLII | Search | Feedback

University of New South Wales Law Journal Student Series

You are here:  AustLII >> Databases >> University of New South Wales Law Journal Student Series >> 2022 >> [2022] UNSWLawJlStuS 6

Database Search | Name Search | Recent Articles | Noteup | LawCite | Author Info | Download | Help

Guo-Hawkins, Elo --- "Who Is A Better Negotiator - A Human Or A Robot?" [2022] UNSWLawJlStuS 6; (2022) UNSWLJ Student Series No 22-6


WHO IS A BETTER NEGOTIATOR – A HUMAN OR A ROBOT?

ELO GUO-HAWKINS

I INTRODUCTION

Evan King[1] was the first to use the ‘Seven Elements Framework’[2] to argue that ‘artificial negotiators’ are better than their human equivalents in almost all elements, except for relationship buildings.[3] King’s 2017 paper and subsequent technological breakthroughs (such as the IBM Project Debater[4] and driverless cars) inspired the Harvard Program of Negotiation (PON) to convene a virtual conference in 2020.[5] The conference explored the current and potential roles of technology[6] in negotiations and where it might lead. The respected practitioners who participated in the conference were amazed by the ever-increasing critical roles that technology performed in negotiation, but also raised their ethical concerns about some developments.

In this paper, I argue like those before me, that a robot negotiator is not a matter of ‘if’ but ‘when’, and we should be prepared for it now more than ever. Meanwhile, human negotiators are indeed needed to address the increasingly challenging ethical issues due to the rise of technology. Humans will also be busy training the robots. Ideally humans and robots work together constructively to improve negotiation outcomes.

This paper is also structured around the seven elements of negotiation. I briefly explain how a human negotiator can prepare for each element without technology. I use some examples from the PON conference to demonstrate the current or potential role technology plays in each element, and consider briefly the benefits, risks and ethical challenges (if any) in this fast-evolving space.

II INTERESTS

‘Interests’ mean someone’s wants, needs, concerns, desires, hopes and fears.[7] A good negotiation outcome should satisfy the negotiating parties’ interests.[8] However, often hidden and unspoken, it is a challenge to identify one’s interests.[9] Even if negotiators know their sides’ interests, it can be difficult to better understand their opponents’ interests.

The role of social media[10] was once neglected, but has demonstrated enormous potential for, among other things, identifying interests. Savvy practitioners increasingly discover that they either harness this technology to their advantage, or else find themselves outmanoeuvred by more digitally sophisticated parties.[11]

A Case Study: Massachusetts Marijuana Production

Based on a real-life event, this case study illustrates the potential use of social media intelligence.[12] The CEO of a legal recreational marijuana company had invested $10 million and set up a marijuana production facility in a small Massachusetts town. He then faced imminent bankruptcy when residents of the town commenced a referendum forcing him to relocate his facility out of town. The CEO hired a famous PR firm and engaged in an aggressive campaign, but that backfired dramatically. The CEO’s last hope was to convince the town’s legislative body to not affirm the referendum within 30 days. He engaged a brilliant strategist who helped launch a word-of-mouth on-the-ground charm offensive. With luck and instincts, he managed to persuade a respected town leader to advocate for his cause. He also negotiated successfully with the primary activist behind the referendum. Eventually, the legislative body amended the referendum to grandfather in his facility, while keeping the ban on retail cannabis from the original referendum intact.

In hindsight, the CEO admitted that ‘it would be extraordinarily valuable to have a playbook that could be quickly internalised for how to utilize social media’.[13] Take his strongest opponent - the primary activist who was behind the campaign against his facility for example, her active presence across social media had given a wealth of psychological insights and biographical details about her publicly. This social media intelligence would have helped the CEO to identify her interests easily.

Figure 1[14]

2022_600.jpg

The activist was a pro-Trump conservative and avid opponent of marijuana. She worried about the town becoming ‘the pot destination’ and didn’t like the ‘hippie’ element of marijuana. However, in her Instagram and Facebook, she followed accounts like FoundMyFitness and PointOfHealingAcupuncutre and liked their ideas of using marijuana medically to relieve pain. She is seen to endorse the broader culture of pro-cannabis. As a tactic to make her more receptive, the CEO could frame languages or borrow images from these accounts.

Figure 2[15]

2022_601.jpg

The activist also cared deeply about the town’s image and volunteered annually at the town’s clean-up day called the ‘Beautification Day’. To bring his opponent to his side, the CEO could hire trucks to help cleaning the town with her.

Figure 3[16]

2022_602.jpg

From open-source scanning of local media channels and Facebook groups, a local real-estate agent turned aspiring state politician stood out as a key potential recruit to the CEO’s cause. The politician supported the cannabis legalisation for treating veterans’ PTSD and the extra tax benefits it would bring. He also sold properties in the district in which the CEO’s facility was located. He might be receptive to the idea that this anti-cannabis referendum could ruin the commercial viability of the district. He could be a key outpost in a conservative community in this referendum.

Figure 4[17]

2022_603.jpg

Using open-source intelligence techniques[18], reams of publicly available data could be shown in a Maltego relationship map (see Figure 5).[19] The map helps the visualisation of personal connections (among the town, key community groups, local leaders), and psychographic profiles of influential individuals. Analysis of the opponents, persuadable and potential supporters could also provide the CEO a wealth of insights about the town.

Figure 5[20]

2022_604.jpg

B Ethical Concerns from Using Social Media

The law does not keep pace with the rapidly evolving social media. But an ethical negotiator should remember that foul play online may attract public backlashes (see Amazon facing severe social backlash over its ‘fulfillment center ambassadors’ program’)[21]. It simply may not pay. Information gleaned through social media should not be derived from hacking, or be used to embarrass or blackmail, to perpetrate fraud, or to foster identity theft. The negotiator should avoid propagating misinformation. ‘The advent of the digital age and the availability of open-source information via social media do not offer excuses to lose sight of one’s personal moral compass or let one’s organizational code of conduct fall by the wayside’.[22]

III OPTIONS

‘Options’ are the full range of possibilities on which the parties might conceivably reach agreement.[23] A good negotiation outcome should be amongst the best of many options. [24] Skill at inventing options for mutual gains is a highly desirable asset for human negotiators.[25] Options can be invented through brainstorming before or during negotiations (as opponents may also suggest good ideas). [26] Options can be broadened by adopting expert advice[27], modifying strengths (eg ‘binding’ vs ‘non-binding’) and scopes of the agreements[28] (eg fractionate the problem vs enlarging the subject matter), or following the four steps in the Circle Chart:[29]

IN THE REAL WORLD│WHAT IS WRONG: Step (1) Identify the problem

IN THEORY │ WHAT IS WRONG: Step (2) Diagnose the problem (eg suggest causes)

IN THEORY │ WHAT MIGHT BE DONE: Step (3) Approaches (eg what are possible strategies)

IN THE REAL WORLD │ WHAT MIGHT BE DONE: Step (4) Action ideas (eg what might be done).

Human negotiators are also trained to invent options through brainstorming sessions in-person or virtually with or without opponents.[30] However, a recent study has found that in multi-party disputes, if the parties can use technology to contribute anonymously, there are higher number of suggestions and more creative options generated as a result.[31]

Software can do a better job with ‘expanding the pie’. The two technology examples below have demonstrated their reliability, efficiency and agility in a complex negotiation environment. Their abilities to generate different options or scenarios based on their Deep Learning of legitimate data, in very short timeframes help save time and encourage creativities on both sides. They are environmentally friendly - imagine the amount of paper required to map out all possible scenarios. The apps have earned trust from their human users throughout the years.

A Picture It Settled®

The “Picture It Settled®’[32] is a predictive analytics software for negotiation. It was created in 2013 by the award-winning US attorney-mediator Don Philbin and others. The software records over 10,000 litigation cases ranging from fender benders to intellectual property disputes. It learns negotiation strategy and patterns. It predicts the opponent’s anticipated reactions to the client party’s behaviour and quickly models dozens of scenarios based on thousands of cases. When the opponent makes offers, the negotiator can determine if and when a settlement might be reached, fine-tune objectives, and plan future concessions accordingly. The figure below shows how the app works. The app tracks the dollar moves and the time intervals between the offers. It then takes that offer history and extrapolates it out in both dollars and time to help parties to determine whether they will reach a deal and when the agreement is likely to occur. (The actual offers are connected in solid lines and the extrapolations are dotted.) The app also helps to calculate future offers based on the concession rates of both parties; and match one’s move in the next round by calculating their (or their opponent’s) move based on previous dollar or percentage. The app also keeps track of multiple negotiations simultaneously.[34]

Figure 6[35]

2022_605.jpg

B Smartsettle-Infinity

Smartsettle-Infinity[36] is a powerful negotiation app which allows both sides to negotiate on multiple issues simultaneously. It became prominent when it generated agreements on a range of complex issues (ranging from education to the budget deficit) between the Sanders and Biden camps prior to the 2020 US election. It provides insights into different complex options real-time, that itself saved time for both parties. I will explain how the app works using ‘Figures’ below. In Figure 7, the app shows Biden’s view on a range of issues that the two camps would be negotiating.

Figure 7:[37] Biden ranked the issues in relative importance. The points next to each issue were adjusted so that every ten points were equivalent to 1 million votes.

2022_606.jpg

Figure 8:[38] For each issue, Biden recorded a satisfaction graph. For example, the optimal spending on the Green-New-Deal peaked at $453b/year. Biden thought that spending more than that would cost him votes.

2022_607.jpg

Figure 9:[39] Biden self-rated on the optimistic and unacceptable packages for each issue. (The BATNA would be what he would do if no agreement was reached). Biden would accept packages that were close to the optimistic package but would be content if he reached the resistance points with Sanders and could make some concessions. The blue testing packages helped Biden to stimulate various options from Sanders. (The total difference between the camps was about 10million votes.)

2022_608.jpg

Figure 10:[40] Various optimistic packages had been proposed from both camps. As their differences narrowed, the parties could either continue exchanging their proposals, or used Smartsettle’s algorithm (‘visual blind bidding’) which helped parties reach agreement faster. The Smartsettle usually suggests options in-between (eg Package11-Suggestion), but not necessarily splits the difference 50-50. That’s because Smartsettle rewards early efforts - if one party moves quicker to the point where the deal may lie, that party gets the better of the deal. If parties don’t like any of this, they can ask for more. Where both parties accept the deal, a yellow dot appears. Smartsettle can also identify ‘hidden-values’ where the deal accepted reached the Pareto optimum frontiers on both camps. Other algorithms (such as the ‘mutual deal closer’) are also available. This process moves much faster and a lot less messy than say, our Karangaroo in-class negotiation exercise.

2022_609.jpg

IV ALTERNATIVES

‘Alternatives’ are the walk-away substitutes for each party if no agreement is reached. [41] A good negotiation outcome should be better than the party’s Best Alternative To a Negotiated Agreement (BATNA). The better the BATNA is, the higher the negotiation power.[42] Before negotiating, a human would think how to improve their BATNAs and legitimately worsen their opponent’s.[43] To identify BATNAs, a human negotiator can select the best from three kinds of alternatives, namely, what they can do to [44]

1) pursue their own interests by themselves

2) make their opponent respect their interests

3) bring a third party to further their interests (eg taking the matter to court).

The following two real-life cases demonstrate how an otherwise-powerless party had used social media as their BATNAs and disrupted the outcomes. The first case was about how Paul Levy, the CEO of the Beth Israel Medical Centre (BIMC) (a Boston teaching hospital), used his blog to successfully fend off a potentially damaging unionisation driven by his opponent, the Service Employee International Union (SEIU).[45] The BIMC was in a stable, albeit not strong, fiscal position in 2006. Its affiliation with Harvard and deep ties to the Jewish community, historically supportive of labor, made it a prime target for the SEIU, which was one of the largest, best-funded labor unions. Levy was supportive of the rights of workers in general. He was moved by the experience of Gail New Haven Hospital which had just gone through the unionisation process. The union leaned on local politicians, delayed building permits and proposed zoning changes so the only height caps in all of the New Haven zoning was actually the hospital. From his observations, Levy concluded that unionisation was not in the best interests of the patients at his hospital.

SEIU had over 200 staff in Massachusetts alone and they had $20 million annual organisation budget. On the other hand, Levy had near-zero alternative as quitting his beloved hospital was not an option; SEIU did not perceive him particularly intimidating; and he had no money to bring on a lawsuit. So Levy used his one-and-only alternative (also his BATNA) that was available in 2006: his blog.[46]

Levy knew his interest which was to avoid direct negotiations altogether with SEIU. (SEIU gave him and hospital stakeholders no option but to accept the unionisation). Levy focused on bettering his position ‘away from the table’. Before using his blog for this exercise, Levy had already earned the trust of his 10,000 followers (consisting of hospital staff, executives, members of the press and politicians) by openly disclosing his personal compensation and difficult or potentially embarrassing incidents at the hospital.[47] Levy also carefully mapped all the relevant or influential stakeholders (see below) and engaged or accepted their demands directly via his blog.

Figure 11[48]

2022_610.jpg

Levy helped inoculate the hospital against many of the SEIU’s tactics by calling them out early, or before they even happened. Levy was consistent with his aims from the beginning, explained his reasons clearly, and referred to traditional American values (such as openness and full discussions on both sides that should ideally guide the unionisation process). By comparison, the SEIU’s website was hardly touched with no updated information available. After failing for five years to secure an agreement, the SEIU ceased its attempt to unionise BIMC. [49]

The second case is about a failed international negotiation. During the Transatlantic Trade and Investment Partnership (TTIP) negotiation, a group of anti-TTIP campaigners had formed a coalition and used social media (Twitter, Facebook and YouTube) as their BATNA, effectively stopped the negotiation between the two ill-prepared negotiators (namely, the US Trade Representative and European Commission). [50] The campaign used targeted emails and tweets, raised funds online and mobilised young voters through news feeds via social media. It collected 1-2 million signatory petitions from young voters and mobilised street protests in France and Germany (each protest drew an average of 200,000 people in 2015).[51] By 2016, the European campaign significantly turned the tide of public opinion. In Germany, for instance, 88% of the public supported international trade deals in 2013 while only 59% supported them in 2016. The campaign also influenced American public opinion. By February 2016, a majority of Americans opposed the TTIP. By comparison, the ‘traditional’ negotiators had little to offer by way of pushing back against such an orchestrated campaign.[52] This case highlights the power and risks when social media is engaged as an alternative to traditional methods. The traditional negotiators’ failure to listen digitally had cost them the deal.

V LEGITIMACY

Other things being equal, an agreement is better if each party considers it fair because it can be measured by some external benchmark, some criterion or principle beyond the simple will of either party.[53] Before negotiating, a human negotiator should choose relevant objective criteria for fairness, especially when both sides will have to explain any agreement to their respective constituencies down the track. A negotiator should only commit to an outcome that is better than BATNA, satisfies the interests and is objectively legitimate. In other words, a good negotiation outcome should be legitimate, by reference to objective criteria of fairness.[54]

In 1992, the ban on online commercial activities was lifted. Internet service providers and the world wide web also appeared. The rise in online activities lead to increasing e-disputes,[55] and the ultimate birth of online dispute resolution (ODR) - a technology also known as the ‘fourth party’[56] at the negotiation table. Its increasingly critical role for our access to justice (or legitimacy) has been highlighted during the pandemic lockdowns.

The number of e-disputes were estimated at 1 billion in 2020, many were separated by jurisdiction, culture, language and inaccessible to court redress.[57] For high volume low value e-disputes, legitimacy could be achieved upon the parties’ acceptance of the decisions made by the algorithms of the ODRs. Amazon, eBay and AirBnb are three of the well known platforms which have invested heavily in building trustworthy ODRs.

ODRs are rapidly expanding their services into legal terrain. Compared to hiring teams of human mediators or researchers doing legal research and providing advices, the following four examples of wizards or apps have demonstrated that technology can provide easier, speedier and more efficient ways to access legitimacy.

A Tax Wizard

A diagnosis wizard built by Modria for the Ohio Board of Tax Appeals[58] has been assisting tax disputants to access tax advice and services in three ways. First, the wizard walks the disputant through the appeal process. The in-built calculator answers questions on costs of the process and the worth of the disputed amount to the disputant and the Board of Tax. Second, based on the evidence supplied by the disputant, the wizard can provide precedents from past online hearings or caselaw and advise the disputant’s chance of winning the dispute, noting that the disputant can access these legitimate advices without filing a dispute. Third, the wizard helps the disputant to file an appeal.[59]

Figure 12[60]

2022_611.jpg2022_612.jpg

B Tyler

The first app (which was the Nokia’s Snake arcade game) was invented 44 years ago. But people wouldn’t think of divorcing via an app 20 years ago. Currently, legitimate apps are increasingly trusted and relied upon for divorces. These apps appear impartial as they are not flattered by compliments. Divorce via an app can be less stressful as each party enters their answers asynchronously at their own pace, away from facing pressures from their former partners. The apps help save resources not just for the divorcing parties but also the court systems as human negotiators would only be engaged to focus on issues that the parties cannot agree.

In US, Modria’s Tyler is a mobile enabled diagnosis wizard which can talk to Amazon’s Alexa. Once the party types or talks about the problem, using natural language processing, the wizard will triage the case and advise the appropriate resolution process. It asks a series of questions, determines the party’s location, and applies the relevant local law. It then points the party towards the right resources and eventually this conversational interface can generate options such as going to the court, appointing a mediator, or using the wizard to negotiate directly with the opponent.[61]

Figure 13[62]

2022_613.jpg

2022_614.jpg

2022_615.jpg

C Rechtwijzer

The Netherland Legal Aid Board uses an ODR called the ‘Rechtwijzer’[63]. If the couple is seeking a divorce, they can enter their information via the wizard and answer all compulsory questions (from the top lit-up circles) at their own pace before kicking off the dispute resolution process. The questions are designed to educate the couple as well as assist with the divorce process, for example, whether the parties agree to co-parenting, if so, helping them to work out the detailed arrangements by asking the parties further questions. If there are no children under 18, the box is greyed out and the parties can move onto the next lot of issues. To some people, this feels very much like ordering a McDonald’s Happy Meal by choosing meal plan 1, 2 or 3. The Board is a legitimate authority and provides default templates and details which can accommodate 80-85% of the divorce cases. The parties can click a default template or build their own plan. If both sides choose the same options, then they have a divorce plan. If they could agree to 9 out of 10 issues, then they can channel their resources and engage a mediator to focus on the remaining issue.

Figure 14[64]

2022_616.jpg 2022_617.jpg

2022_618.jpg

D Marktplaats

Marktplaats is the leading ecommerce website in the Netherlands. It engages Crowd Sourced Justice (Gebruikersjury) as a way to impose legitimacy. The Gebruikersjury consists of a group of uninvolved community members who never transacted with the disputants. The jury listens to both sides and renders decisions which are enforced by Marktplaats. The jury can range from 50, 100 or even larger than 1,000 people. Technology has made it scalable, efficient and achievable to access justice. [65]

Figure 15[66]

2022_619.jpg

VI COMMITMENTS

‘Commitments’ are oral or written statements about what a party will or won’t do. They may be made during the negotiation or may be embodied in an agreement reached at the end of the negotiation.[67] Commitment should only be made if the negotiator believes that the outcome is good. A good negotiation outcome should produce commitments likely to be kept.[68]

To transit from inventing options to making commitments, a negotiator is traditionally trained to craft a framework agreement[69] which is a document in the form of an agreement with blank spaces for unresolved issues. As the negotiation progresses, possible terms of agreements are drafted and filled into the blank spaces.

However, as mentioned before, apps such as Rechtwijzer and Smartsettle-Infinity are already making this transition to commitment with ease and efficiency.

I provide another example here, namely the Jointly Editable Agreement Templates,[70] where negotiating parties edit an agreement real-time. The parties can see the components of the agreement being submitted by their opponents. They can also build an e-framework agreement with all the blanks the parties need to fill in. Practitioners have seen these templates resolving complicated disputes before. So who wrote the agreement? The parties did - they all wrote, contributed and committed to their own agreement.

Figure 16[71]

2022_620.jpg

A Ethical Concerns Due to the Rise of Technology

As the responsibility of technology increases, so does the level of ethical challenges it brings. Poor technology can reduce procedural and substantive access to justice. When we have inaccurate data, compounding with the negative impacts on the dispute system design or machine learning, technology can lead to an unfair distribution of benefits. The lack of transparency and accountability are highly risky, for example, bad actors can give different processes and outcomes to different populations based on their postcodes or other identify factors. Processes and outcomes can be designed to perpetuate inequality when the system lacks transparency.

There are inherited legacy issues such as power imbalances between one party who uses the technology every day and another who knows very little. Although we are aware of them, we are still not fully aware of the extent of inequality or inefficiency occurring. For example, if we ask parties to maintain confidentiality in a Zoom conference, and the parties agree, it is still not addressing all the ethical issues and risks raised when technology is employed in the process. People may not be familiar with the ODR process and not entirely sure what they agree to in a Zoom conference. Their opponent can record and post the video to social media the next day. Some apps that disclose location details can cause severe consequences in domestic violence disputes. Parties can also sign up to algorithms that facilitate their movements through different stages, provide options for outcomes which are structurally determined and therefore narrowing the ultimate benefits for the parties. There have been debates in the last five years on whether technology permeates or alleviates inequalities in processing outcomes for gender, race, disabilities and other factors not relating to identities. Currently we have not reached definitive consensus on these issues.

One big lesson we’d learned from the eBay experience is that for an ODR to be successful, there needs at least three elements: trust, expertise (or intelligence) and convenience (ie efficient, time saving and easily accessible).[72] While an unethical ODR system can still be intelligent and convenient, it will have trouble gaining trust. So contributions from human negotiators are essential in designing and building an ethical system in order to gain users’ trust.

VII COMMUNICATION

Efficient negotiation requires effective two-way communication.[73] Procedurally, a good outcome should involve good communication.[74]

I present some interesting finding and breakthrough technologies here. While the first two examples seem to have served the public good, the last example is thought provoking from an ethical perspective.

A The ‘Politeness Package’ Algorithm

This algorithm uses pre-trained natural language processing models to detect politeness in English language. Figure 17 shows the 36 linguistic features this algorithm can currently detect. [75]

Figure 17[76]

2022_621.jpg

Using this algorithm, researchers found that in most circumstances, receptive language makes disagreements more productive.[77] For example, in contentious policy discussions, government executives who were rated more receptive by the algorithm were considered better teammates, advisors and workplace representatives. (Interestingly, the executives may not necessarily agree with the algorithm.)[78] Receptiveness at the beginning of a conversation forestalls conflict escalation at the end - Wikipedia editors who write more receptive posts are less prone to receiving personal attacks from disagreeing editors.[79]

Figure 18[80]

2022_622.jpg 2022_623.jpg

2022_624.jpg

2022_625.jpg2022_626.jpg

The exception: warm language can backfire in distributive negotiations,[81] for example, receptiveness doesn’t work when one haggles price via texts.[82]

Figure 19[83]

2022_627.jpg2022_628.jpg

It remains yet to be seen the full-scale application of this algorithm, but I like the idea that a machine reminds me to remain receptive in my language and hence achieving efficient negotiation outcomes.

B Cogito[84]

Cogito is a platform that uses AI to perform in-call voice analysis and deliver real-time guidance to call center representatives during their negotiations with customers.[85] Before Cogito, typically 50% of the communication would end up in fights. Cogito listens to the patterns of communications, and gives real-time props to representatives to steer the conversations away from conflicts. Research has shown that customer satisfaction has increased by 28% and staff experience has sky-rocketed by 63%.[86]

Figure 20[87]

2022_629.jpg

C Deepfakes[88]

Most of the technologies attempt to bridge the gap between face-to-face and online communications, some arguably are widening the gap. Deepfakes can implicitly alter the perceptions of one’s race, speaking style, age and other attributes, often in real-time. Figure 21 was snipped out of an AI constructed footage, during which former President Obama convincingly called former President Trump a very rude name. The footage later revealed that the name calling act was actually delivered by Jordan Peele (on the right), and Obama’s images and speech were fake.[89]

Figure 21[90]

2022_630.jpg

On the other hand, it is possible that soon, deepfakes can enable Indian call center representatives to pose as Americans, adopting the American accent, speaking style, and grammar to improve communications with their American customers.[91] The technology can also help smooth out tremors in body movements and vocal sounds in real-time for a Parkinson’s disease patient during a job interview.[92]

‘If negotiation is challenging when audio-visual cues are missing—for example, over the telephone or in text messaging—the situation is made substantially more complicated by the introduction of covertly altered cues, a novel reality in the negotiation landscape.’ [93] These technologies raise further ethical concerns like whether people should be allowed to change their personal attributes, and if so, to what extent should the technology be allowed to modify one’s interactions. [94] No doubt, human debates and wisdoms on these ethical issues are much required to guide future direction of the technology.

VIII RELATIONSHIP

A good negotiation outcome should improve parties’ relationships.[95] Human negotiators would argue that robots don’t know how to build relationships. Robots don’t know ‘trust’ and ‘emotions’. Robots don’t do small talk or smile. The two examples below are proving otherwise. In the second example, a robot negotiator even earned human trust with lies and won the competition.

A Interactive Arbitration Guide Online (IAGO)[96]

‘Human-aware’ techniques increasingly consider non-financial interests such as humans care about losing face, being listened to and acknowledged, and genuine relationships. [97] Ideally, algorithms can use mental models to discover mutually beneficial trade-offs;[98] reason about social norms like fairness; capture and correct human errors[99] or re-pick up missed opportunities. Currently, no AI negotiators can bring together all these capabilities into one system, however, IAGO is the next closest thing (and you can imagine its enormous potential once you witness how it performs).

IAGO is an online platform that allows researchers to define various negotiation scenarios and automated negotiator behaviours. IAGO then tests these behaviours on human participants. In the negotiation of chairs exercise (Figure 22), IAGO examines human users’ manner by which semantic content is delivered, for example, facial expressions, gestures and speech rhythms. Combining with advanced machine learnings, it can predict personality traits, trustworthiness and even the probability of reaching a deal.[100]

Figure 22[101]

2022_631.jpg 2022_632.jpg

In the salary negotiation exercise (Figure 23), the panel on the left shows menus to exchange offers, messages and preference information. The panel on the right shows a history of the dialog and the algorithm beginning with a low offer. The user engages in some information exchange before making a counteroffer. IAGO models many of the cognitive processes and even biases that users exhibit. Users can also express emotion through emojis, or select statements such as ‘I expect to be compensated better’.

Figure 23[102]

2022_633.jpg

B Ethical Concern: Winner is a Liar

Unfortunately, ‘human-aware’ algorithms can also exploit human weakness. At a recent AI negotiation competition,[103] prizes went to the team whose agent extracted the greatest concessions from human negotiators. The winning agent employed a bogey tactic, by pretending to highly value low-value issues, then offered concessions on these issues to get what it truly wanted. The human negotiators believed the agent was making real concessions and felt that the agent was fair and honest, even though its offers were the most unfair of all. [104] It is a little concerning that a lying agent was the winner.

IX CONCLUSION

We know from Moore’s Law about every 2 years computer processors double in power. We witness the achievements by IBM’s Deep Blue, Watson and recently, its Project Debater’s success in debating against humans.[105] We hear that quantum computing and driverless cars will soon be available for everyone. Colin Rule is already predicting that future negotiations will be conducted between Bot to Bot, and Bot to human. But human negotiators are still much needed. We need the human wisdom, morals and hearts to design trustworthy ethical systems. We need humans to continue to train and refine robot negotiators. So who is a better negotiator – a human or robot? I think the best negotiator will combine the best of humans and robots.


[1] Evan King, ‘Artificial Negotiators’ (Roger Fisher and Frank Sander Writing Prize Paper, Harvard Law School, 2016-2017) <http://hnmcp.law.harvard.edu/wp-content/uploads/2012/02/King-Evan-Winner.pdf> . The paper is winner of the May 2017 Roger Fisher and Frank Sander Writing Prize.

[2] ‘What is Negotiation? - PON - Program on Negotiation at Harvard Law School’, Harvard Law School (Website, 2 November 2020) https://www.pon.harvard.edu/daily/negotiation-skills-daily/what-is-negotiation/. The Seven Elements Framework was developed by the Harvard Negotiation Project to help people negotiate effectively. The seven elements include ‘interests’, ‘options’, ‘alternatives and the Best Alternative To a Negotiated Agreement (BATNA)’, ‘legitimacy’, ‘commitments’, ‘communication’ and ‘relationships’.

[3] See ibid.

[4] Project Debater (Website) <https://www.research.ibm.com/artificial-intelligence/project-debater/live/>. In February 2019, after participated in a debate on the topic of ‘preschool should be subsidized’, the IBM Project Debater was voted by the audience to have enriched everyone’s knowledge on the subject matter, more than the opponent (Harish Natarajan), who is a world debating champion and world record holder for most debate competition victories. In November 2019, Project Debater won the debate on the topic ‘AI will do more harm than good’ at the historic Cambridge Unio.

[5] ‘Program on Negotiation Teaching Materials & Publications: Working Conference on AI, Technology, and Negotiation,’ Harvard Law School (Website, May2020) < https://www.pon.harvard.edu/teaching-materials-publications/working-conference-on-ai-technology-and-negotiation/ence on AI, Technology, and Negotiation - PON - Program on Negotiation at Harvard Law School>.

[6] In this paper, the term ‘technology’ broadly refers to any application of scientific knowledge for practical purposes. So ‘technology’ includes (but not limited to) artificial intelligence, algorithm, social media, machine learning, Big Data, deep learning, robots, Application Programming Interface (API), cloud, avator, Zoom and many other inventions.

[7] Roger Fisher, ‘Seven Elements Explanatory Memo’, (Harvard Negotiation Project, Harvard College of Law, 1989) (‘Seven Elements Explanatory Memo’). See also UNSW Moodle LAWS8980 Principled Negotiation General Course Material https://moodle.telt.unsw.edu.au/pluginfile.php/6844903/mod_resource/content/2/7%20Elements%20Preparation%20Memo%20Info%20%28V2%29..pdf.

[8] Roger Fisher and William Ury, Getting to Yes – Negotiating An Agreement Without Giving In (Penguin Books, 3rd rev ed, 2012) 21. (‘Getting to Yes’)

[9] Ibid 43.

[10] James Sebenius et al, ‘Dealmaking Disrupted: The Unexplored Power of Social Media in Negotiation’ (2021) 37(1) Negotiation Journal 97, 100. The term ‘social media’ refers to not only the major platforms such as Facebook, Twitter, Instagram and LinkedIn, but also less well known or obvious sources such as Reddit, Glassdoor, 4Chan, Pinterest, Yelp, WeChat, Snapchat, TikTok, Line, Sina Weibor, Vimeo, Wikipedia, WhatsApp, Viber, QQ, Meetup, Google Search, Google Maps and many other forums, channels, discussion groups, and platforms that often prove invaluable sources of data and information.

[11] Ibid 97.

[12] Sebenius (n10) 105-114. Refer to Case Study#1: Oldhaven, Massachusetts. See also PON HLS, ‘PON AI, Tech, & Negotiation Conference Session C: Potentially Critical Roles of Social Media’ (YouTube, 26 May 2020, 1:05:36) <https://www.youtube.com/watch?v=qKhLtrud_sA>. (‘PON Session C’) For privacy reasons, I have avoided mentioning any names, even though it was a well-publicised real-life event.

[13] Sebenius (n10) 113.

[14] PON Session C (n12) 1:13:22.

[15] Ibid 1:13:24.

[16] PON Session C (n12) 1:14:14.

[17] Ibid 1:16:42.

[18] The discussion of tools and techniques to analyse social media information is outside the scope of this paper. For further information, see Sebenius (n10) 97.

[19] Figure 5 is a graphic illustration of a Maltego relationship map and is an example from another case study, not the current Massachusetts marijuana production case study. See Sebenius (n10) 107 for further source article discussions.

[20] PON Session C (n12) 38:22-43:32.

[21] Tech, ‘'Fake' Amazon ambassadors baited on Twitter’, BBC news (online,16 August 2019) < https://www.bbc.com/news/technology-49372809>

[22] Sebenius (n10) 103-104.

[23] Seven Elements Explanatory Memo (n23) 1.

[24] Alan Limbury, ‘Seven Element Explanatory Memo-Expanded’, (UNSW Moodle LAWS8980 Principled Negotiation General Course Material, 2021) <https://moodle.telt.unsw.edu.au/pluginfile.php/6844904/mod_resource/content/1/Seven%20Elements%20Expanded%20Memo.pdf> 1. (‘Seven Element Explanatory Memo-Expanded’)

[25] Getting to Yes (n8) 58.

[26] Seven Element Explanatory Memo-Expanded (n24) 1.

[27] Getting to Yes (n8) 70.

[28] Ibid 70.

[29] Ibid 67.

[30] Seven Element Explanatory Memo-Expanded (n24) 1.

[31] PON HLS, ‘PON AI, Tech, & Negotiation Conference Session B: Lessons from Online Dispute Resolution’ (YouTube, 26 May 2020, 1:00:36) < https://youtu.be/OQm-4Qfap7w>. (‘PON Session B’) Leah Wing explained that this finding was made during an early study she did (a number of years ago) with the National Centre for Technology and Dispute Resolution along with the National Mediation Board.

[32] Don Philbin Picture It Settled (Website) <http://www.donphilbin.com/picture-it-settled/> . The “Picture It Settled®’33 software was created in 2013 by the award-winning US attorney-mediator Don Philbin and others.

[34] Don Philbin, ‘Pictured It Settled Lite Overview’ (YouTube 19 Nov 2011) < https://www.youtube.com/watch?v=7ifbExaAwD8>. (‘Picture It Settled Lite’)

[35] See Ibid.

[36] Colin Rule, ‘Smartsettle 2020 Election Demo for International Council for Online Dispute Resolution (ICODR)’ (YouTube, 11 April 2020, 0:00-58:45) < https://www.youtube.com/watch?v=NrxRlSsId_E>. (‘Smartsettle’)

[37] Smartsettle (n35) 10:31.

[38] Ibid 12:17.

[39] Smartsettle (n35) 20:21-35:40.

[40] Ibid 37:04-43:34.

[41] Seven Elements Explanatory Memo (n23) 1.

[42] William Ury, Getting Past No – Negotiating with Difficult People (Random House Business Books, Version 1.0 (ebook), 1992) 20-21. (‘Getting Past No’)

[43] Seven Element Explanatory Memo-Expanded (n24) 1.

[44] Getting Past No (n41) 20-21.

[45] Sebenius (n10) 114-118.

[46] PON Session C (n12) 48:34-1:02:16.

[47] Sebenius (n10) 115-116.

[48] PON Session C (n12) 53:25-54:02. Also see ibid for source paper discussions.

[49] Ibid 117-118.

[50] Cathryn Ashbrook and Alvaro Zalba, ‘Social Media Influence on Diplomatic Negotiation: Shifting the Shape of the Table’ (2021) 37(1) Negotiation Journal 83, 86-89.

[51] Ibid 87.

[52] Ibid 88.

[53] Seven Elements Explanatory Memo (n23) 1.

[54] Seven Element Explanatory Memo-Expanded (n24) 1.

[55] Colin Rule, Keynote Workshop on Private International Online Dispute Resolution (Stanford University, 2017).

[56] Ethan Katsh and Janet Rifkin, Online dispute resolution: Resolving conflicts in cyberspace (Jossey-Bass, 2001). Janet Rifkin and Ethan Katsch created the concept of ‘fourth party’ as the metaphor of ODR in 2001. The traditional parties at a negotiation table include the two disputants (party 1 and 2), and a human neutral (ie the mediator). The fourth party is referred to as the technology components such as the Zoom meeting app or other technologies which mange the communication and processing of information that is at the heart of every ADR process.

[57] Rule (n54).

[58] Ohio Board of Tax Appeals Resolution Centre (Website) <https://ohio-bta.modria.com/>.

[59] PON Session B (n31) 24:09-30:11.

[60] Ohio Board of Tax Appeals Resolution Centre (Website) <https://ohio-bta.modria.com/>.

[61] PON Session B (n31) 30:12.

[62] Ibid 30:11-30:31.

[63] Ibid 31:08-32:40.

[64] PON Session B (n31) 31:08-32:40.

[65] Ibid 39.36.

[66] PON Session B (n31) 39:25.

[67] Seven Elements Explanatory Memo (n23) 1.

[68] Seven Element Explanatory Memo-Expanded (n24) 1.

[69] Getting to Yes (n8) 175.

[70] PON Session B (n31) 38:47-39:20.

[71] Ibid 38:47-39:20.

[72] PON Session B (n31) 13:54.

[73] Seven Elements Explanatory Memo (n23) 2.

[74] Seven Element Explanatory Memo-Expanded (n24) 1.

[75] Yeomans et al, ‘Detecting Politeness in Natural Language’ (2018) 10(2) The R Journal 489 <https://www.mikeyeomans.info/papers/Rpoliteness.pdf>.

[76] PON Session D (n79) 25:27. Also see Ibid 3.

[77] Yeomans (n74) 3.

[78] Ibid 136.

[79] See ibid.

[80] PON HLS, ‘PON AI, Tech, and Negotiation Conference Session D: New Insights into Psychological Processes’ (YouTube, 26 May 2020, 45:12-47:19) <https://www.youtube.com/watch?v=HeFiDSzOers>. (‘PON Session D’)

[81] PON Session D (n79) 38:08

[82] Ibid 55:21-55:35.

[83] Ibid 55:17-55:44.

[84] Cogito (Website) <https://cogitocorp.com/>.

[85] Dinnar et al, ‘Artificial Intelligence and Technology in Teaching Negotiation’ (2021) 37(1) Negotiation Journal 65, 117-118.

[86] PON HLS, ‘PON AI, Tech, & Negotiation Conference Session A: State of the Art’ (YouTube, 27 May 2020,33:00) <https://youtu.be/15yk9GojMJ8>. (‘PON Session A’)

[87] See Ibid.

[88] Raiyan Baten and Ehsan Hoque, ‘Technology-Driven Alteration of Nonverbal Cues and its Effects on Negotiation’ (2021) 37(1) Negotiation Journal 35, 41.

[89] BuzzFeedVideo, ‘You Won’t Believe What Obama Says In This Video! ��’ (YouTube, 18 April 2018, 0:45) <https://www.youtube.com/watch?v=cQ54GDm1eL0>. (‘Deepfakes’)

[90] Deepfakes (n88).

[91] Baten and Hoque (n87) 42.

[92] Michael Wheeler, ‘Introduction to Special Issue’ (2021) 37(1) Negotiation Journal 5, 9.

[93] Baten and Hoque (n87) 43.

[94] Ibid 44.

[95] Seven Element Explanatory Memo-Expanded (n24) 1.

[96] Welcome to IAGO.Explore Your Emotions. (Website) <https://jtmell.com/iago/>

[97] Jonathan Gratch, ‘The Promise and Peril of Automated Negotiators’ (2021) 37(1) Negotiation Journal 13, 18.

[98] Ibid 17.

[99] Ibid 17-18.

[100] Gratch (n96) 17.

[101] PON Session A (n85) 46:06-47:56. See also ibid, for further discussion in the source article.

[102] Ibid 48:40-49:38.

[103] Gratch (n96) 18-19.

[104] Mell et al, ‘Results of the first annual human-agent league of the automated negotiating agents competition’ (Conference paper, International Conference on Intelligent Virtual Agents, 5-8 November) 2018) https://dl.acm.org/doi/abs/10.1145/3267851.3267907.

[105] Project Debater (Website) <https://www.research.ibm.com/artificial-intelligence/project-debater/live/>


AustLII: Copyright Policy | Disclaimers | Privacy Policy | Feedback
URL: http://www.austlii.edu.au/au/journals/UNSWLawJlStuS/2022/6.html