Weekly Research and Commentary on the Future of Business and Technology.
AI Writes Code – But Is It Legal?.
Jul 4, 2022
A few weeks ago, at an event from the wonderful Entrepreneurs’ Organization, the event organizer played Ryan Estis’ talk “The Simple Secret To Happiness”. It’s a tad cheesy, but so, so good (and true). It’s how we approach every interaction here at be radical, it is how, I, personally, approach life, and it is, essentially Newton’s Third Law of Physics at play: To every action there is always opposed an equal reaction. The “action” is within our control.
And now, this…
Practical Futurism // Decode. Disrupt. Transform.
You might have heard about GitHub’s Copilot — using OpenAI’s Codex AI, which has been trained on billions of lines of code, Copilot suggests code and entire functions based on descriptions such as “Determine whether the sentiment of text is positive by using a web service”. Copilot takes this input and spits out a complete function — when you first see it in action, it feels magical.
The product was in an invite-only beta for a while and recently made its debut as an open-for-all SaaS service. For $10/month, every programmer can now have their Copilot. And this is where the trouble starts: First, Copilot is not perfect — far from it. GitHub itself admits that “in a recent evaluation, we found that users accepted on average 26% of all completions shown by GitHub Copilot.” The issue here isn’t the 26% (which isn’t great, but likely will get better over time as the AI learns from all those user interactions), but rather the fact that, if you aren’t a great program, that those 26% of accepted code suggestions might be buggy (and likely will be, as it is near impossible to write error-free code).
The challenges don’t stop here. The much bigger issue is the overall legality of tools like Copilot. To make Copilot work, the AI system was trained on code which is publicly available in the GitHub code repository — much of it in the form of Open Source code. And herein lies the problem: A lot of this code comes with specific licensing terms. Open Source licenses such as GPL are legally binding agreements which limit what you can do with the code and what not. And most likely, ingesting code by an AI system to train the system breaks multiple of these license agreements. Here is just one of many articles discussing the issues at hand.
All of which points towards a future for AI systems which is fraught with legal challenges: Who owns the training data AIs are being trained on? How can it be used to train the AI, and what happens with ownership of the derivative work an AI creates? Who owns the output of an AI system? Good time to be an IP lawyer these days… (via Pascal)
What We Are Reading
🍀 4 Types of Business Transformation Business transformations can take many forms and often occur at the same time. Understanding the root of the cause and the timing to pivot are fundamental when defining your approach and critical to success. Jane ⇢ Read
🛒 Should You Buy Now, Pay Later? Tread Carefully. Too good to be true. How consumers are losing control and going on shopping splurges using “Buy Now Pay Later” that turn out to be messy and a direct hit to one’s credit score. Mafe ⇢ Read
🏙️ The smart city is a perpetually unrealized utopia The technologies meant to enable smart cities have always failed to recognize humans as anything beyond “convenient repositories of data” and the city itself as a complex, rich, and emergent thing. Jeffrey ⇢ Read
😨 The great Silicon Valley shake-out Having lived through this cycle twice now (2000/2001 and 2008/2009), I have to say: Nothing new under the sun, but rather a predictable pattern which repeats itself every ten years or so. Pascal ⇢ Read
— Pascal, Mafe and the three Js (Jane, Jeffrey, and Julian)
Subscribe to the radical Briefing, your ‘everything, everywhere, all at once’ source for our latest insights and
commentary. Explore ‘radical’ perspectives on the future of business, with cutting-edge news, research, lessons, and
ideas on new technologies, business models, methodologies, cultural shifts, and leadership trends.