Business Lawyers in Columbus, Ohio
By Drew Stevens - September 18, 2018 - Technology & IP
Hailed by some as one of the more promising tools in artificial intelligence and machine learning, neural networks present both a range of exciting applications and legal contract issues. As an AI lawyer, I always find working on AI, cloud platform, and neural network-related services contracts engaging, based on the unique and interesting issues that crop up. This post will break down what neural networks are and some of the legal provisions that neural network and AI services providers should consider when negotiating AI-related contracts.
Despite the recent rise in the use of neural networks in advanced AI applications, not to mention increased tech-press coverage of neural networks, the technology, and underlying ideas are far from new. The theoretical base for neural networks was first advanced in the late 1800s, but basic advances in the field did not begin in earnest until the 1940s. Development picked up in the 1970s and 1980s but building early neural networks was time-consuming, labor intensive, and severely limited by the technology available at the time. Thanks to significant breakthroughs in the last decade, neural networks have begun to offer the potential and applications that technologists and AI experts have dreamed of.
At their core, neural networks are models designed to somewhat mimic the activity of the neurons in the neocortex, the wrinkly part of the human brain. Your brain consists of billions of neurons that accept input from external stimuli and other neurons. When you have a thought or move your arm, prior to such action taking place, signals travel through neurons, which then, in turn, connect with even more neurons, progressively making even more connections until the action takes place.
Somewhat similarly, neural networks are composed of artificial neurons, usually referred to as nodes or units, that are arranged in layers. The first layer, the input layer, is where the neural network receive information from external sources. The final layer, the output layer, is where the processed information is generated and presented. In between the input layer and the output layer lie hidden layers. The hidden layers are where processing and feature extraction happen. The more hidden layers that are involved, the deeper the neural network.
There are a variety of neural networks that exist. One of the easiest to understand and visualize is the feedforward neural network. Here, information travels in a one-way direction from the input layer, then to the hidden layer, and finally to the output layer. The connections do not form a circle or loop in the network. Contrasted to the feedforward neural network, the recurrent neural network features information that flows in multiple directions. The recurrent neural network can be ideal for processing more complex functions like speech and language recognition. Other types of neural networks include Hopfield networks, convolutions neural networks, and Boltzmann machine networks.
There has been no shortage of coverage and conjecture regarding what neural networks can be used for. Broadly speaking, neural networks can be ideal for processing big data and inferring relationships and patterns – tasks that would take human beings countless hours to perform. Popular examples include image processing and recognition – think facial recognition for social media and photo apps and fraud detection for banks in which issues with customer’s handwriting are detected.
Beyond image recognition, even greater potential lies in forecasting applications. Neural networks have the capacity for analyzing and processing complex data sets and drawing nonlinear relationships. For example, this could enable financial businesses in the lending industries to better predict whether a customer will fulfill or default on a loan.
With advanced technology and great potential comes a variety of legal issues and contract concerns. AI and neural networks services and services contracts can take a number of forms. This includes (a) services providers licensing access to neural network platforms to customers, or (b) offering to perform research and analysis for customers based on data submitted by the customer. Regardless of the services arrangement, the following areas are issues that AI service providers and the customer should contemplate.
Without question, neural networks can involve extraordinarily expensive technology platforms and services and, at times, even more valuable data. If you’re the service provider, you should strongly consider incorporating the usual disclaimers into your services agreement, including disclaiming the warranties of fitness for a particular purpose and merchantability. As neural network platform services may process vast amounts of big data, service providers should also contemplate disclaiming any liability related to losing data or deletion of data. Taking this a step further, a service provider may want to require that the customer stipulate that the customer will be responsible for backing up specific projects and data.
As no piece of technology or software is perfect, it can be common to also address errors and service interruptions. Some provisions will stipulate that the service provider is not representing that the technology platform will operate 100% error free. In terms of interruptions, a service level agreement may want to be included or an acceptable service level may want to be defined. If you’re the service provider, depending on the value and nature of the services, you may want to flatly state that you cannot guarantee that the platform will operate uninterrupted.
Finally, both parties should contemplate actual dollar limits on potential liability. Usual limits for advance technology and AI-related agreements include one or both parties disclaiming liability for punitive damages, consequential damages, incidental damages, and damages for lost revenues. A cap on damages may also save thousands or millions of dollars, including limiting damages to the amount of money exchanged by the parties under the contract.
Another key area that the parties should address is the rights and limitations of both the service provider and the customer when it comes to the actual use of the neural network’s services. For example, will the service provider be able to access and actually review the data submitted by a customer? Such data could involve extremely sensitive confidential information and trade secrets. There are a number of scenarios where a customer may wish to limit the service provider’s access to the data. However, situations may also arise where the service provider might actually need to review such data – think troubleshooting system errors.
AI service providers can spend millions of dollars developing neural network tools and platforms, and accordingly, should take steps to protect such development. From a contractual perspective, this includes explicitly stating that a customer cannot reverse engineer, decompile, disable, copy, or otherwise attempt to pick apart or modify the underlying software.
Depending on the industry associated with the data, the service provider may also want to exclude certain uses of the neural network platform, or at least disclaim that the tools are not meant to be compliant with industry-specific regulations. For example, healthcare technology companies that process data can invoke HIPAA. The neural networks provider may want to stipulate in such a case that the responsibility for handling and processing of such healthcare data in compliance with HIPPA falls on the customer.
Finally, having a good acceptable use policy that is attached as an exhibit or schedule to the services contract can save both parties a lot of heartache. By making a clear disclaimer that a neural network platform cannot be used for a certain subject matter, a service provider can create another layer of liability protection and even stipulate that breach of acceptable use terms is grounds for immediate suspension of access. Common components of acceptable use policies include prohibiting using the services for unlawful or infringing purposes or using the services to generate spam and mass unsolicited advertising.
Well defined payment provisions can provide clarity for both parties. Generally, with neural networks platforms and tools, the parties should consider project and milestone-based payment or monthly pricing. Procedures for invoicing and deadlines for payment should be clearly stated. In terms of taxes, especially with international businesses, the service provider should be sure that the customer is responsible for all taxes.
The parties should also contemplate procedures for late payments or failure to make payment. Specifying an interest rate for late payments is common. In the event the customer falls excessively behind on payments or stops paying completely, service providers should have a termination or suspension of services provision. Services providers should take care in crafting such a provision and ensure that they don’t find themselves held liable for abrupt termination-related consequences in the event the customer is in the middle of a neural networks project.
Finally, both parties should be clear regarding their intellectual property rights. In the case of the service provider, it should generally be stipulated that all right, title, and interest in the neural network platform remains with the provider. From the customer’s perspective, consideration should be given to the rights concerning customer data. All intellectual property rights in the customer data should generally remain with the customer, but the parties may want to stipulate that the customer will license the use of the data to the service provider or license the right to use anonymized data.