Guest Post: Spencer Williams, Generative Contracts

Spencer Williams (right) is an Associate Professor at the Golden Gate University School of Law, where he researches and teaches at the intersection of business law and innovation. He writes on topics such as AI contracting, smart contracts, quantitative modeling, complex systems theory, algorithmic price gouging, and the venture capital industry. Prior to teaching at GGU, he was a fellow at Stanford Law School in the Program on Corporate Governance and Practice. Earlier in his career, he practiced corporate and securities law at Wilson Sonsini Goodrich & Rosati in Silicon Valley and served as the General Counsel of a Y Combinator-backed technology startup. He received a JD from Stanford Law School and a BS from MIT.

Generative Contracts
How consumers and small businesses can use generative AI to write their own contracts.

Since the public release of ChatGPT last year by OpenAI, generative artificial intelligence (a form of AI that can generate text, images, and other content) has quickly made its way into knowledge-based industries such as law. In the context of contracting and transactional law, most of the attention thus far has been focused on using generative AI to improve the process of reviewing, negotiating, and managing contracts for medium and large-sized businesses. Examples include BlackBoiler, ContractKen, Evisort, Genie, Ironclad, Lexion, Ontra, and Robin.

Less attention has been given to using generative AI to write entire contracts. This is mostly due to the fact that contracts for medium and large-sized businesses are often too long, complicated, and complex to be generated by currently available AI models. An underexplored area of opportunity, however, is to use generative AI to write shorter, simpler, and less complex contracts for consumers and small businesses. This is particularly appealing from an access to justice perspective because most consumers and many small businesses cannot afford to pay commercial lawyers to write basic contracts.

To explore this possibility, I engaged in a proof-of-concept hypothetical in which two consumers, Alice Brown and Carl Delmonte, use ChatGPT to write a contract for Alice to buy Carl’s used car. A few notes about the hypothetical. First, in an attempt to represent how consumers might engage with such a product, I used prompts that required no prior knowledge of contract or commercial law. Second, I did not use prompt engineering. Third, I did not fine-tune the underlying model (OpenAI’s GPT 4.0) for contract-specific applications. A real-world consumer contracting AI product would likely use prompt engineering and fine-tuning, particularly given that fine-tuning has become substantially cheaper and faster due to the use of low-rank adaptation (LoRA). I plan to examine prompt engineering and fine-tuning with LoRA in future posts.

For the hypothetical, I began with the following prompt:

“Write a basic car sale contract in California for Alice Brown to purchase Carl Delmonte’s 2014 Honda Civic with VIN# 12345ABCDE for $5,000.”

In a matter of seconds, the model produced the following contract:

BILL OF SALE AND PURCHASE AGREEMENT

THIS AGREEMENT, made this 16th day of May, 2023 (the “Effective Date”) by and between Carl Delmonte (the “Seller”), and Alice Brown (the “Buyer”).

1. DESCRIPTION OF THE VEHICLE

   Make: Honda

   Model: Civic

   Year: 2014

   VIN: 12345ABCDE

2. TERMS OF SALE

   The Seller agrees to sell and the Buyer agrees to buy the above-described vehicle (the “Vehicle”) for the purchase price of Five Thousand Dollars ($5,000.00), payable as follows: full payment upon the execution of this Agreement.

3. AS IS SALE

   The Seller expressly disclaims any implied warranty as to fitness for a particular purpose and any implied warranty as to merchantability. The Vehicle is sold in “as is” condition. The Buyer acknowledges that the Buyer has had an opportunity to inspect the Vehicle, and is not relying on any representations or warranties of any kind whatsoever, express or implied, from the Seller as to any aspects or characteristics of the Vehicle, other than as expressly stated in this Agreement.

4. TRANSFER OF TITLE AND DELIVERY

   The Seller shall endorse the proper documentation for the transfer of the Vehicle and deliver it to the Buyer at the time of sale. The Buyer is responsible for registering the Vehicle in their name.

5. GOVERNING LAW

   This Agreement shall be governed by and interpreted in accordance with the laws of the State of California.

6. ENTIRE AGREEMENT

   This Agreement constitutes the entire agreement between the parties with respect to the sale of the Vehicle and supersedes all prior discussions, negotiations, and agreement, whether oral or written.

SELLER:

Carl Delmonte

Signature: _______________________

Date: ___________________________

BUYER:

Alice Brown

Signature: _______________________

Date: ___________________________

NOTARY ACKNOWLEDGEMENT

State of California

County of ___________________

On ____________, 2023, before me, a Notary Public, personally appeared Carl Delmonte and Alice Brown, known to me (or satisfactorily proven) to be the persons whose names are subscribed to the within instrument and acknowledged that they executed the same for the purposes therein contained.

Notary Public: ___________________

My Commission Expires: ____________

DISCLAIMER: This is a simplified car sale contract for illustrative purposes and may not be legally sound or complete. It’s always recommended to consult with a legal expert or attorney when drafting or entering into a legal agreement.

—

A few observations about this contract. First, while not perfect, it is a completely functional and enforceable contract. Second, GPT 4.0 now uses standard contract structure principles such as sections, numbering, and definitions which were not always used by previous versions of the model. Third, ChatGPT now includes a disclaimer at the bottom of any legal documents it generates, though this disclaimer can easily be removed with the prompt, “Modify the above contract to remove the disclaimer at the end.”

One of the main advantages of generative contracts relative to other consumer-facing contract products such as templates (e.g. LegalZoom), which are fixed and static, is that generative contracts are incredibly flexible and customizable using simple prompts. For example, Alice and Carl can easily modify the contract using prompts such as, (1) “Modify the above contract to clarify that Alice can pay via Venmo,” (2) “Modify the above contract to say that Carl has to bring the car to Alice’s house located at 1234 Main Street, Faketown, California,” and (3) “Modify the contract to include a section for Carl to write in the odometer reading on the day of the contract.” The model seamlessly incorporated all of these requests into the contract.

Beyond basic modification, a generative product such as ChatGPT can also answer questions about the contract and then make modifications based on the answers it provides. For example, if Alice and Carl don’t want to bother with the hassle of getting the contract notarized, but are unsure if this is a requirement in California, they can ask, “Does a bill of sale for a car in California have to be notarized?” The model provided the following answer:

As of my knowledge cutoff in September 2021, a bill of sale for a car in California does not need to be notarized. California does not require a notary for a car title or bill of sale. The most important aspect is that the document includes necessary information about the seller, buyer, and the vehicle, and that both parties sign it.

However, laws and regulations can change, and it’s always a good idea to check with the local Department of Motor Vehicles (DMV) or a legal professional for the most current information.

—

If they are satisfied with this answer, Alice and Carl can remove the notary section with the prompt, “Modify the above contract to remove the notary section.” This example demonstrates a weakness of generative contracts (and all generative AI products): their knowledge is time-limited based on the model’s data cutoff date (in this case September 2021).

As another example of dynamic modification, assume Alice and Carl are uncertain about how the contract handles repair costs incurred after the sale. They could ask a clarifying question such as, “If the car breaks down the day after Alice buys the car, does Carl have to pay for the repairs?”, which returns the following answer:

According to the terms of the contract provided, the car is being sold “as is,” which means the Seller, Carl, disclaims any implied warranties. This usually includes the condition of the car after the point of sale. Therefore, if the car breaks down the day after Alice buys it, Carl would generally not be responsible for the repair costs under the terms of this contract.

However, this can also depend on the specific laws of the state in which the sale occurred. Some states have “lemon laws” that could potentially provide some protection to the buyer in such a situation. It’s always a good idea for both the buyer and the seller to understand their rights and responsibilities under the law before entering into such a contract.

It should be noted that this response is based on the given contract and general principles and may not account for all possible legal considerations. It’s recommended to seek advice from a legal professional for specific situations.

—

If Alice is unhappy with the “as is” nature of the contract, they could modify the contract with a prompt such as, “Modify the contract to say that if the car experiences any problems beyond regular wear and tear within the first month after Alice buys the car, then Bob has to pay for any repairs up to $1,000.” In response, the model replaces the “As Is” clause with a “Limited Warranty” clause that reads:

The Seller provides a limited warranty covering any problems beyond regular wear and tear that the Vehicle may experience within the first month following the Effective Date. If such problems occur, the Seller agrees to pay for any necessary repairs up to a maximum of One Thousand Dollars ($1,000.00). Beyond this amount, the Buyer is responsible for the cost of any additional repairs.

—

This hypothetical only begins to scratch the surface of what generative AI can do for consumers and small businesses. In addition to this car sale contract, I have used simple, consumer-accessible prompts to create contracts such as an employment agreement, a consulting agreement, a services agreement, a nondisclosure agreement, a residential lease, a commercial lease, a procurement contract, a partnership agreement, a limited partnership agreement, an LLC operating agreement, a software licensing agreement, and a mobile app terms of service agreement.

I’ll be continuing to explore the application of generative AI to contracting in the coming months. If you are interested, please reach out to me at swilliams@ggu.edu.

In the meantime, if you are looking to sell a Honda, you might try something like this: