By: Matthew Jurgensmeier

Machine learning —a subset of artificial intelligence that teaches a machine to complete a task through iteration— provides significant possibilities for the future of production across many mediums, including video, music, and text. As these machine learning technologies evolve, they present opportunities for users to increase their efficiency and focus on less structured tasks.  However, machine learning may also present problems on both a legal and a societal level.

Simple automatic text generators which are programs that predict what a user will type before they finish typing, are a somewhat new technology that provide benefits for technology users. They have been integrated into cell phone keyboards to allow users to communicate more effectively. More recently, popular email client Gmail rolled out this feature for personal emails. Indeed, these word-level and sentence-level predictors can be quite useful, saving users time with each keystroke. These programs also make intuitive sense, as a person is provided with options for how to continue to the next word, or how to complete a commonly used sentence or phrase. However, these technologies have clear limitations, especially when it comes to creative endeavors.

That is, these technologies had clear limitations.

What is OpenAI?

Enter OpenAI, a company focused on researching and building safe artificial general intelligence software. OpenAI’s focus is developing tools that promote automation for the benefit of society. Whereas this non-profit research company typically shares its tools widely to accelerate artificial intelligence research, its actions relative to its most recently developed tool have been more conservative. The company’s newest machine learning algorithm, trained on Reddit using millions of web pages, takes the concepts of text prediction and generation to a completely different level, producing pages of text in a matter of seconds based on prompts ranging from a few words to a paragraph or more.

What are the benefits of OpenAI?

In a paper published by OpenAI, the company details its progress, as well as its hesitation for publishing a full-fledged version of the model. While the advancement of this technology will doubtless provide benefits for people in a variety of fields, it provides significant potential for abuse. Some of the potential benefits are artificial intelligence writing assistants which would increase user efficiency, more capable dialogue agents which would improve user experiences with automated systems, unsupervised translation between languages, and better speech recognition systems.

What are the concerns with OpenAI?

A few of the malicious purposes outlined in the paper include using the model to generate misleading news articles, impersonate others online, automate the production of abusive or faked content to post on social media, and automate the production of spam/phishing content. One relatable example is that with fine-tuning – training on a more specific and nuanced dataset – the program could write reviews on Amazon given conditions of star rating and category. These automated reviews could pose a real problem for manufacturers whose main sales platform is within these online marketplaces.

How will copyright law handle this machine-generated content?

In addition to concerns regarding public policy and misuse, the current state of intellectual property law will likely be stressed by innovations like this program. Rather than a traditional tool for production, like a typewriter or even an assistive program like Microsoft Word, OpenAI’s program and other machine learning text generators will function as the producers themselves, meaning that traditional copyright laws may need to be updated. As it is currently constructed, domestic copyright law will not attribute copyright ownership to a machine. Indeed, the threshold question of whether the creator of the program or the user of the program should be provided with copyright protection jumps to the fore. As it stands now, at least in the Ninth Circuit, it appears that works created by a non-human – a machine – even a very well-developed machine, may reside in the public domain.

Due to the potential abuses of its technology, OpenAI is taking a patient, careful look at the current state of the world and preparing itself and the deployment of its technology for the worst. As a non-profit enterprise, OpenAI is empowered to make decisions like this because they do not have to focus on being first-to-market. In leading by example and practicing patience, OpenAI is setting a standard in hopes of curtailing bad actors before they ever have a chance.

For an example of how machine learning is being used to generate fake images and videos, visit https://wjlta.com/2018/02/15/deepfakes-a-disastrous-merger-of-ai-and-porn/