Journal of Business welcomes Yanek Kondryszyn, intellectual property attorney and partner at Lee & Hayes PLLC, for its latest Elevating The Conversation podcast.
Elevating The Conversation podcast is available on Apple Podcasts, Amazon Music, Spotify, and more. Search on any of these platforms or Journal website To hear the entire conversation, here are five takeaways from the just under 40-minute episode.
1. The term artificial intelligence is often used too narrowly. AI is a very large umbrella term that describes a more philosophical way of thinking. I get a little repulsed when I hear people talk about AI tools, and especially when they apply the term to describe large-scale language models (a category of natural language processing that's currently booming). .
I think the reason it's such a problem is because machine learning is just a method. The big fuss we're seeing is centered around generative models at the moment. A set of specific machine learning models that perform exceptionally well in text generation, text-to-speech, and speech-to-speech.
But the problem is that at some point, there will probably be forms of AI other than machine learning models. So whether it's using biological circuits in some way, whether it's using biological logic, chemical logic, quantum computing, we're all starting to associate AI specifically with large-scale language models and diffusion models. Once you put it away, it starts to become useless.
A further reason not to use this term for generative tasks is that AI could certainly be used for non-generative tasks in the future as well. For example, robotics. Generally speaking, it's just another use.
2. Be careful about the client data you input into your AI or machine learning tools. If you are a company that receives personally identifiable or sensitive information of any kind, you need to be aware of what is being provided as input to the machine learning models you are using.
For example, if you want to generate an image or process a summary of information to incorporate into your model, you can do this: you may be in real trouble If you haven't yet looked into the terms of use for that model.
Free versions of these models often allow you to use your data to train the technology in exchange for being free.
And what exactly does that mean? I don't know the official name of his one of those toys, but imagine a game where you put a coin in and it goes through a bunch of nails and you get some sort of prize. Essentially, figuratively speaking, the training data moves the position of those nails. So someone else could take the input and end up with an output close to what was produced for you. If you suddenly have a customer whose personal information is exposed because their fingernails are moved in a certain way, that could be a problem.
This is especially risky right now, especially as many of these providers are competing to improve their benchmarks in ways that can set them apart from their larger competitors. It's not “Hello, how are you doing?” The input they are struggling with. It's a strange thing, an outlier. And guess what? That outlier data may contain a small amount of information that you entered.
3. Be equally careful about how you use the finished machine learning model. We've covered quite a bit about the inputs to the model. I think another thing to be careful about is the output from the model and how you use it.
The U.S. Copyright Office has made the fact that copyright applies only to creative works created by human beings very clear. As the example of Photoshop shows, digital tools can be used to aid human creation. However, unless a person is deeply involved in the creation, there is no copyright.
If you want to create a jingle for your business or advertising material, you don't use a machine learning model to generate it in a pure sense. I'm not going to just give the model a prompt and use everything it has to offer. Instead, use it for review. For example, take a photo and use tools to modify it. This is close to the way the Copyright Office seems to lean toward the idea that things are subject to copyright.
Consider a situation where an ad includes a jingle. If you generate it using a purely machine learning model, you run the risk of your competitors plagiarizing your work, and there's nothing you can do about it. Granted, that doesn't go into trademark issues, but I still think you need to be aware of how the output is used and how copyright applies.
4. In this day and age, a company's customer data can become a trade secret. Managers need to recognize that they also potentially have a treasure trove of high-quality human data.
Any system that tracks interactions within human-written documents becomes valuable training data for machine learning models. In my view, the real new oil in this economy is human-generated data. This is because there will be a huge influx of AI-generated data.
It can be difficult to sift through the noise and find good training data to train these models. And a lot of companies, even here in Spokane, are probably storing some of these mountains of data, and with the advent of these generative models, all of a sudden a lot of documents could become trade secrets overnight. You have to realize that there is.
That may mean consulting an attorney and finding a trade secret strategy to protect your documents. What may not have previously been a trade secret may suddenly become one because it is valuable as training data for industry models.
5. Take a lifelong learning approach when it comes to AI. It would be foolish not to become a learner, right? These tools must be used in a manner that does not expose personal data or sensitive information. We highly recommend trying out the free versions of these tools to familiarize yourself with their features.
But don't leave it there. I think there are a lot of posts on LinkedIn about prompts and how to get the right output, but it's really just about how to do a Boolean search. I see myself in something like a Boolean search phase in the innovation cycle. In some ways, you have to do a lot of quick engineering to get the results you want.
You might say, “What do you think of a business that has bicycles out front?'' But that's not exactly what you wanted. So you actually need a slightly more sophisticated set of prompts, like aperture type and setting type. In the future, we will see innovations that reduce the friction of interfacing with these tools.
Thankfully, we can go back to the basics of business: what you need to consider for your business. We need to get back to the basics of investigating how this technology can benefit us, what the real business advancement is, and whether we need it.
As business owners, what are the biggest friction points for us? And are there models in place today that actually help those friction points? To be able to answer these questions, Of course, you both need to mind your own business, but you also need to have a good understanding of what the tool actually does.
