AI For Negotiation (And Maybe Even World Peace)

Over many years as an investor, I’ve found that one of the most simple and successful approaches to disruption is this: take a human-driven process and automate it with technology.

When we reflect on most negotiations—whether in the finance industry (funding rounds, acquisitions, etc.), government (legislation, policy, diplomacy, etc.), or otherwise—they usually go through the same motions as they have for years if not decades. ?They’re unautomated, human dependent, emotional, unpredictable, and suboptimal.?And yet they often consume tens if not hundreds of hours and thousands if not millions of dollars.?Why do we insist on reinventing the wheel every time??Could AI be the gamechanger?

Here’s an idea, in several versions of evolution, that I think begs to be pursued (and which I'd love to fund):

Version 1.0: AI-driven negotiation software could automate the process.

The base-level idea here is that our AI negotiation agent would take a party’s initial positions as inputs, liaise with the other side (whether human or AI), use natural language via GPT to iterate between the sides, arrive at mutually acceptable terms, auto-populate a standard legal template, and then deliver completed drafts for lawyers to take across the finish line.?

Might this reduce legal transaction costs by 50-80% while reducing the time from months to days??Might it improve negotiated outcomes by removing some of our self-defeating human tendencies??Might it help each party, and even the lawyers, become more efficient so that they can focus on other, higher value-add activities?

Now someone is going to say, “It already exists.”?Yes, I’m aware of several startups applying this concept to high-volume vendor contract lifecycle management (CLM): Pactum, Intellext, Icertis, ContractAI, SirionLabs, Ironclad, and others.?

Those are a good start, but I’ve got something even bigger in mind here.?Has anyone developed AI to negotiate and optimize big transactions like funding rounds, mergers and acquisitions (M&A), congressional legislation, and international treaties??If so, I’d be thrilled to hear it because taking this concept to the next level is what I’m really interested in.

Version 2.0: Next-level AI-automated negotiation could optimize outcomes.

When I started out as an investor, I viewed negotiation as a tug-of-war.?You say your startup is worth $20M, I say it’s worth $10M, and then we push and pull until one of us gives in somewhere in the middle.?Amateur.?As I matured, I began to take the “give up nickels to get dimes” approach.?Slightly better, but still narrow-minded and adversarial.?Finally, I had a revelation: what if both sides worked together collaboratively toward an optimized outcome??Suddenly, I became partners with founders as we shared what we cared most and least about so that we could solve each other’s problems and create a win-win.

That’s what Version 2.0 looks like: AI software—probably as agent for each side and possibly as neutral honest broker—that helps both sides get the most of what they want while giving up what they don’t mind giving up.?These optimized negotiations leave each side feeling like they got the best deal they could have.?Plus, having partnered together with the AI rather than negotiated as adversaries, both parties conclude the deal amicably without the sense of exhaustion and partial defeat that usually accompanies a closing.?

In other words, AI-optimized negotiations move everyone from amateur to professional all at once.?But how about improving beyond the capabilities of even the best human negotiators?

?

Version 3.0: Path-optimized, behavioral algorithms could create the best possible outcomes over the long-term. ?

Once the AI software has learned how to both automate and optimize existing processes, it can add multiple time horizons, behavioral algorithms, and path dependency.

Put another way, this is thinking multiple chess moves ahead.?Most negotiating parties, however, are only thinking of the present moment without considering how it will affect future negotiations down the road.?The way to teach a longer-term mindset is to walk someone through multi-year scenarios, run simulated outcomes, ask them questions, adapt to their perspectives, and help them reach that understanding on their own.?Not all humans have the patience for this, but AI does.

What’s more, many negotiations—in which both parties starting at Point A would happily shake hands at Point B—fall apart along the way.?One side overreaches, another sides gets emotional, and suddenly someone has walked.?In other words, the path and the sequencing can make all the difference.

AI negotiation software with pathfinding could foresee and manage this.?If Path X and Path Y are not going to lead to an agreement but Path Z will, the AI could steer the parties through that path to get everyone to yes.?In a world where there are often hundreds, thousands, or even millions of paths, AI could well go beyond what any human could do.?Imagine how many more funding rounds, M&A transactions, and even congressional legislation could reach fruition with such software.

So Version 3.0 is now solidly in the realm of AI achieving things that humans could not. ?What else is possible?

Version 4.0: AI could create new opportunities from scratch where none would have otherwise existed.

All of the above presumes that we begin with two parties who have already found each other.?But what if AI could take care of that too with a bit of matchmaking?

Imagine our AI platform surveys an M&A landscape, takes into account all of the key considerations of target acquisitions on the one hand and prospective acquirers on the other, and then proposes a match??For example, it could suggest to a VP of corporate development at a big tech company several start-ups that are likely to boost their growth if acquired or threaten their existing products if not.?Then, the AI platform could advise on when would be the optimal time to acquire them, which valuation could likely get it done at each stage, how to preempt an auction, and how to steer the negotiation to a successful close.

Writ large, this could automate and optimize the entire M&A landscape. ?Lucrative??Highly.?Good for the world??Questionable.?But that’s the ultimate potential application of this AI platform.

Version 5.0: AI could unlock hidden solutions to the world’s most intractable problems among multiple stakeholders.

Nowhere are people more familiar with impasse than the Middle East.?Years ago, while discussing ways that Israelis and Palestinians could finally achieve peace, a Middle Eastern friend told me pointedly, “You Americans are so amusing.?You think that if you put everyone in a room with enough whiteboards and enough time, you’ll find a solution.?In this part of the world, we’ve learned that sometimes there simply is no solution.”?He was right.?I had been naive. ?And, yet, I find my optimism and idealism are an incurable condition.?

Maybe AI can help.?Rather than plot a path toward the best possible, pre-defined end state, perhaps AI could be utilized to find an undefined but least bad outcome that no human could envision?

Imagine a state of geopolitical affairs wherein AI aids diplomats in exploring imperfect but tolerable solutions to long-term conflicts.?Imagine that it could take into account a wide range of stakeholders, balance all of their interests, determine how to navigate red lines, find a palatable end state, and determine the sequencing to guide each party toward that outcome.?The Israelis and Palestinians may always be at odds, but perhaps there are more stable equilibria that all parties could live with such that the conflict becomes sustainably non-violent.

This may be where quantum computing comes in.?AI has already demonstrated—as early as 2016 when AlphaGo wowed the world with its now famous “Move 37”—that it can divine approaches that no human would ever try.?And Meta’s Cicero AI beat humans using game theory in the multi-player game of Diplomacy.?But in real-life geopolitics, a system as complex as peace negotiations may require exponentially more computing power.?Quantum computing may be the new hardware needed to reduce the compute time from hours to seconds for these extremely varied and variable systems.?

Don’t get me wrong, I’m not suggesting that AI can, will, or should replace humans at every level.?Rather, I’m proposing that AI can enable humans to achieve much better outcomes than they could on their own.?AI may never have the imagination, creativity, emotional intelligence, and general wherewithal of people.?Properly channeled though—and with all the necessary ethical safeguards—it could help us do our best and perhaps even unlock possibilities that we could have never achieved on our own.?