In 1973, the artists Richard Serra and Carlota Fay Schoolman bought television airtime to broadcast a short video titled “Television Delivers People.”1
The video consists of an elevator music soundtrack playing while sentences in white text on a blue background slowly scroll upward. The video’s message was the product of commercial television, which was the mass audience delivered to advertisers. A recurring phrase ran through the video:
It is the consumer who is consumed.
You are the product of t.v.
You are delivered to the advertiser who is the customer.
He consumes you.2
When I wrote a business case for Grammarly, widely used at UTA, I started thinking about us being the product in this new age of AI. Developing business cases for potential UTA use is one of the services OIT’s Innovation Architecture team provides, which I’m on.
Currently, UTA has around 529 Premium (paid) and 11,000 Free Grammarly individual licenses. I learned from discussions with Grammarly that the company uses client data differently depending on the license type. Grammarly categorizes the Free and Premium accounts as self-serv since you can sign up directly from their website.
Suppose you use a Free or Premium license. In that case, some of your data is being used to train their large language model (LLM)3, where customer data training controls (CDCT) are automatically on, which means a small amount of anonymized, disassociated information trains the large language model (LLM). Individuals can contact Grammarly to ask to turn CDCT off, and then no client data will be used to train the LLM.
Grammarly for Education (G4e) is an enterprise license where CDCT is automatically turned off, and client data is not being used to train the LLM.4 Additional features in G4e licenses support governance and heightened data security. These features give organizations insight into how their users use Grammarly, such as an analytics dashboard that shows the types of suggestions typically taken and the # of prompts used.
Using self-serve licenses of generative AI tools such as Grammarly raises concerns about non-published UTA research data that may leak out to LLMs. Further research is needed to determine if other AI vendors use the same business model as Grammarly.
When working with AI vendors, know the free and paid license models we describe. Vendors such as Box have made their AI Principles clear.
Grammarly’s security rating on BitSight is 790, at the top of the technology industry’s range. Grammarly has been in business for 15 years and scaling responsibly and securely. It works with 500k apps, including the full Microsoft suite.
As we work with AI vendors, we will ask about their security rating and make that information available to the UTA Community.
We find ourselves in a back-and-forth arms race regarding student and vendor use of AI. Awareness of how data is used on LLM for free and paid licenses is a first step toward finding a sustainable way to live with AI.
There is consensus that AI detection tools are unreliable, and we must figure out how to live with generative AI. One path is to work with vendors and establish low-risk pilots. UT Austin is starting to launch a Grammarly for Education Generative AI Pilot.
By the way, Richard Serra produces monumental Corten steel sculptures worth experiencing…by walking through them.
A trial account of Grammarly for Education generated all article versions below.
Prompt: Improve the article:
By Dana Miller, Business Architect at OIT Innovation Architecture
AI Delivers People?” by Dana Miller, a Business Architect at OIT Innovation Architecture, presents a thought-provoking exploration of the evolving relationship between technology users and the platforms they engage with, framed through the lens of a 1973 short video by artists Richard Serra and Carlota Fay Schoolman titled “Television Delivers People.” With its minimalist presentation of text scrolling over a soothing elevator music backdrop, this video delivers a potent critique of commercial television’s role in commodifying its audience for advertisers, encapsulated in the chilling assertion, “It is the consumer who is consumed.”Drawing parallels to the current era of artificial intelligence, Miller reflects on the implications of being “the product” in the context of the widespread use of Grammarly at the University of Texas at Arlington (UTA). With UTA’s adoption of approximately 529 Premium and 11,000 Free Grammarly licenses, the conversation shifts to how these tools, emblematic of generative AI, utilize user data. Notably, the distinction between the data handling of Free and Premium individual licenses versus Grammarly for Education (G4e) licenses highlights a critical discourse on privacy, data security, and the ethical use of AI.
While Free and Premium users contribute, albeit minimally and with safeguards, to the training of Grammarly’s Large Language Models (LLMs) through Customer Data Training Controls (CDCT), G4e licenses stand apart by excluding client data from LLM training entirely. This bifurcation raises questions about the potential for unintended data leakage from non-published UTA research through self-serve licenses, prompting a call for further investigation into AI vendors’ data practices.
Miller’s narrative extends beyond data privacy concerns to encompass the broader security and ethical considerations of engaging with AI technologies, underscored by Grammarly’s impressive security rating and responsible scaling. The dialogue culminates in reflecting on the “arms race” between AI utilization and detection, advocating for informed, cautious integration of AI tools in educational settings through pilot programs like the one initiated by UT Austin.
In closing, Miller’s essay not only revisits the historical critique of media consumption but also positions the discourse within the ongoing challenge of navigating the complex landscape of AI ethics, security, and governance. Through this lens, “AI Delivers People?” resonates as a timely meditation on our role and agency in the age of AI, inviting readers to contemplate the delicate balance between technological advancement and the preservation of human values.
Prompt: Make the article sound more on-brand:
**AI and the New Paradigm of User Data**
By Dana Miller, Business Architect at OIT Innovation Architecture
In 1973, artists Richard Serra and Carlota Fay Schoolman utilized television as a medium to broadcast a thought-provoking video titled “Television Delivers People,” highlighting the consumer’s role as the actual product in commercial television delivered to advertisers. Fast forward to today, and we find ourselves in a similar discussion with AI at the forefront.
At UTA, our engagement with tools like Grammarly, widely adopted across the university, is a prime example. Our Innovation Architecture team, which I’m part of, delves into the intricacies of these technologies, evaluating their fit within our ecosystem. With Grammarly, we’ve come to understand the nuances between their Free and Premium licenses, particularly in how user data contributes to training their sophisticated algorithms. Interestingly, Grammarly differentiates its licenses by the level of data utilization, with an opt-out feature for those concerned about privacy.
Moreover, the introduction of Grammarly for Education (G4e) licenses marks a significant shift towards enhanced data protection, automatically excluding user data from algorithm training. This model, prioritizing data security and governance, offers an analytical lens into user engagement without compromising privacy.
This exploration raises important questions about the transparency and ethics of AI tools, particularly those that leverage user data to refine their models. As we navigate this terrain, engaging with AI vendors that uphold clear data usage and security principles is crucial. Grammarly’s commendable security practices and robust integration across various platforms set a benchmark in the industry.
As we forge ahead, the dynamic between student use and vendor offerings of AI tools demands our vigilant attention. Understanding the intricacies of data usage in AI and distinguishing between free and paid licenses is a critical step toward a harmonious coexistence with these technologies. Our initiative at UT Austin to pilot the Grammarly for Education Generative AI Pilot exemplifies our commitment to navigating this landscape responsibly.
In closing, the conversation around AI and user data is reminiscent of the message conveyed by Serra and Schoolman decades ago, underscoring the enduring relevance of critically examining our role within the digital ecosystem. As we continue to explore the potentials and pitfalls of AI, let us remain mindful of our position, not just as users but as stewards of our data.
Prompt: Shorten the article:
**AI: A New Era of User Data Utilization**
By Dana Miller, Business Architect at OIT Innovation Architecture
In 1973, Richard Serra and Carlota Fay Schoolman’s broadcast, “Television Delivers People,” provocatively highlighted how commercial TV turned viewers into products for advertisers, encapsulating the message “It is the consumer who is consumed.”
This notion has resurfaced with the advent of AI, prompting reflections on our role as ‘products’ in this digital age. At UTA, our exploration into Grammarly’s use unveils a nuanced approach to user data. Grammarly employs data from its Free and Premium users to refine its algorithms, albeit with measures for anonymization and an opt-out option. However, its Grammarly for Education (G4e) licenses take a stricter stance, not utilizing user data for algorithm training, offering enhanced data security and governance.
The distinction between self-serve and enterprise licenses raises essential considerations regarding data privacy and the potential exposure of sensitive information. Understanding the specifics of data usage becomes crucial as we navigate the complex landscape of AI tools. Grammarly’s high-security standards and clear differentiation in data handling practices exemplify vendor transparency.
The evolving interaction between AI tools and users underscores the need for vigilant research and responsible partnerships. UT Austin’s initiative to pilot a Grammarly for Education Generative AI Pilot represents a step toward balancing innovation with privacy. Our awareness and actions will shape a sustainable coexistence with AI technologies as we continue this journey.
Prompt: Write the article as Haiku:
In screens we trust, sold—
Consumers consumed, now AI
Shapes us, unseen hands.
References
- Quote Investigator. (2017, July 16) You’re Not the Customer; You’re the Product. https://quoteinvestigator.com/2017/07/16/product/
↩︎ - Wikipedia (2023, September 18) Television Delivers People. https://en.wikipedia.org/wiki/Television_Delivers_People
↩︎ - Grammarly note: With G4e licenses, LLM training is always off. When off, Grammarly will not retain anonymized, de-identified, or pseudonymized user or enterprise content for product improvement. Training is on as a default for all self-service customers; this means Grammarly may retain random samples of text to help us improve the algorithms underlying our software and services. All random samples of text are de-identified, anonymized, and sanitized. Access is extremely limited and monitored to ensure data never leaves secure confines.
↩︎ - Grammarly note: We have consolidated our Enterprise Data Control behaviors into a new control for educational institutions (and enterprise business customers). CDTC ensures we will not improve our product using customer data nor train on data from education-licensed customers. ↩︎