AI Training: Ethics or Coverage?

Some authors on the internet are against using their content without permission to train the models behind generative AI. As a blogger, this question often comes to my mind. On one hand, I believe that training without permission poses an ethical issue. I am unsure if we have genuinely addressed this question as a society. On the other hand, I question the danger of many authors wanting to silence their voices by blocking the training process to access their online content. Now more than ever, I feel that all voices are essential in ensuring that model training reflects our diversity as much as possible. I don’t want bots and haters to win the race. We, as authors, need to offer resistance. I’m not sure how this challenge will be solved, if ever.

One Big Regret of My Digital Life

One of the things I regret the most is not having had the idea of creating a blog in the 90s (and keeping it until today). Surprisingly, I learned about HTML, web servers like Apache and Netscape when it became popular1. I didn’t click with the idea of owning a small portion of the Internet to share what I was becoming at that time. What a missed opportunity.

  1. It’s not exactly true. I used to have and maintain a website about a subject that I still find fascinating: meteorology, circa 1994. ↩︎

Yesterday, while writing and editing a report for one of our clients, I used ChatGPT for two different use cases. One use case was to ask for a summarization of what “firmware” is and how critical it is. The second use case is to define the pillars of a data management and governance policy in the enterprise. On that one, I asked for more details about managing unstructured data. The ChatGPT results were mind-blowing. I know a lot about this specific IT field, and I could validate the correctness of the answers. I saved a lot of time because of ChatGPT.

But what about my ethics?

Should I write a disclaimer in this report that says GenAI was used to put together some portions of this report? Is the client ready and mature enough to read this disclaimer? Will he understand that ChatGPT is in fact like an assistant to whom I asked to summarize what a data governance policy is? How do I cite my sources?

Microsoft supports the Apple Vision Pro headset with Office 365 Suite, Teams and Copilot. At least five apps, on day one, for a use case (productivity) still to be demonstrated as viable. For the most probable day-one use case of the Apple Vision Pro, Netflix refuses to create a single app, a “simple” video library browser and player. WTF? Politics. 😑

Avoiding Vendors Lock-in

The notion of “avoiding vendors lock-in” in information technology is interesting. I would argue that it’s impossible to completely be without some sort of vendor lock-in1. At some point, there is always a required commitment level. You commit to open-source software. You commit to a cloud vendor. You commit to a platform. I often give the example of a company building an application internally with a team of developers. In that scenario, the company is committing to something: the application, the data tied to it and its operational model. Applications are hard to replace in many environments. When you decide to invest in software development, you commit to the end product for many years, if not decades, until the organization decides it’s time to transform the application into something else2. There is this concept of “security by design” and of “portable by design,” which should apply to any technical or application architecture. The rest is marketing nonsense.

  1. I do understand data portability concepts and loose coupling principles, though. ↩︎

  2. When it’s time to adopt new architecture paradigms like the cloud is imposing. ↩︎