AI and ethical licensing
AI is a hot topic right now. Every tech company wants to brag about its AI. They make claims that it will change your business, save lives, keep people safe, and even change the world as we know it. Change the world it will; the important question is whether that will be a net gain or loss. AI doesn’t have a well-established formal definition right now and has largely become a marketing term. This allows companies to claim that they use AI without a way for people really know what’s going on behind the curtain. For the purposes of this article, I’m going to define AI as any system that uses Machine Learning technology, since that’s what most academics mean when they say AI.
Unfortunately, all of the positive media around the real-world use of AI seems to be winning every day people over quite effectively. I say unfortunately because they are only being presented with incomplete information, half-truths, and even outright lies about the ability of these systems and their potential impact.
New technology is always dangerous when used before we understand it. A great example of this is Radithor. This was a drink that was developed and sold during the 1920s and had radioactive thorium added to it, with the purported benefit of keeping you healthy and energized. Today we recoil in horror at this, thinking about the instances of cancer and other negative effects this would have on people. But they genuinely didn’t know that at the time. Had they waited just a few years, lives could have been saved.
We are in a similar situation today. No matter what you hear from corporations on AI the truth is that we don’t yet understand it. We know how they work, but not why they work. If you were to ask a leading expert to explain in detail exactly why an AI model gave a certain output, they wouldn’t be able to do it. Traditional code is mostly deterministic. This means that you can show step by step what happens when it runs for a given input. Each step is explicitly defined by the human that wrote the code. AI\Machine Learning however is not deterministic. It’s a “black box” that we can’t see inside of.
While this is an incredibly exciting field of study, with future potential to do good in the world, our understanding of it is far too shallow to use it in ANY way that could impact someone’s life. Yet this is already happening every day. AI is being used to decide which applicants get an interview for a job. It’s being used to diagnose medical conditions. It decides whether or not you get that home loan you’ve always dreamed of.
They even use it to try and “predict” whether a criminal will commit more crimes if released on parole. That’s right, the fate of someone’s freedom rests on an unproven system that we don’t understand. There are some acceptable uses for it of course. One person has used it to create a game that provides a totally unique experience each time you play it. Others use them to help write blog posts, or to help provide inspiration during writer’s block. Using it for things that don’t impact people’s lives, like recognizing the text on a receipt for research marketing purposes is fine.
And sadly there has already been proof that there are issues in using it this way. Loans have been denied because of the neighborhood a person lives in. Models have rejected applications largely based on the person’s status as a minority. The systems that are used to predict whether someone will commit future crimes have been found to base its decision on things like gender and ethnicity.
As a developer, I refuse to work on any AI projects for companies that will use or sell them for life-impacting purposes. However, The only way we can grow our understanding is to continue the research, and research is most effective when shared. What then is to keep someone from using these AI/Machine Learning models for dangerous or even malicious purposes?
Ultimately there is no way to prevent this from happening. But, there is a way to discourage it, and maybe stop it in some cases. Researchers have recently released BLOOM, a groundbreaking AI/Machine Learning model that is likely the most advanced model ever created. The code is open, and it is free to use, but with a catch. Inside of their software license, they have actually placed ethical restrictions on its use. Usually, licenses have restrictions for legal reasons, such as to protect the author or to limit how it can and can’t be used in a legal context. Open source software has licenses that encourage the sharing of the code, which is an ethical requirement in and of itself; placing ethical restrictions on the actual use of the code itself is a relatively new development. The researchers that created BLOOM are well aware of the ethical dangers around the advanced model they have created, and have made the decision to place parameters around its use.
The restrictions mentioned below aren’t complete, they are simply ones of interest to this article.
The first restriction of note states that you may not use it to “generate or disseminate verifiably false information with the purpose of harming others”. Disinformation is currently a hot topic in our society, and their choice to include this restriction shows one of the ways this tool can be used for malicious purposes. The phrasing presents some questions on how this could be proven if they were to take someone to court over it. Hopefully, it is well enough defined to empower them to stop malicious actors.
The next interesting restriction is that any content generated must come with a disclaimer explicitly stating that it was machine generated and not written by a human. Now, this may sound silly, since you would be inclined to think that it would be obvious to a reader whether it was written by a human or not. However, this model and others like it can generate content that truly is indistinguishable from human-generated content. Even those with backgrounds in the field can often no longer tell the difference. Some may question how this could present an ethical issue, however, I believe that the knowledge that something is machine-generated can have an influence on the impact that it has on a person. If something goes wrong and hurtful or damaging content is generated, the viewer can know it was machine generated and therefore not “take it to heart” as such. Emotional injury is still an injury, and people have a right to know whether a human or machine is behind the content they consume.
It also specifically forbids using it to automate legal decisions that impact a person’s legal rights or using it to provide medical advice or results interpenetration.
And finally using it cannot be used by law enforcement or legal processes to predict or prove that a person committed or will commit a crime.
My understanding is that this license is the first of its kind for AI\Machine Learning. I believe it does a fantastic job at addressing the most common and dangerous uses of AI that can negatively impact people’s lives. I sincerely hope other researchers and open source maintainers, in general, consider this kind of license for their own projects. It’s obviously not needed everywhere but it is important that we start thinking about how our software, both AI and traditional, can be used in ways that cause harm to the world.
