A Spotify-based Model for Knowledge Royalties as an Alternative to Mass Unemployment
“It is difficult to get a man to understand something, when his salary depends on his not understanding it” -Upton Sinclair
Goldman Sachs estimates that 300 million jobs will be negatively affected by advances in modern AI. Goldman estimates that 44% of legal jobs could be lost, with a 42% loss in administrative jobs. Elon Musk has posed that intelligent machines might be “the workforce of the future.”
Background
Back in 2015 in the Gangnam district of Seoul, a group of friends and I hosted a weekly roundtable called Super-AI and the Future of Work at Maru 180. Usually the roundtable was capped at 10 people. We strived for a diverse set of profession and nationality representations, usually with a 50% local participant base. In the meetings, the core organizers would give a quick update on some advances in AI that week, followed by the general topic of discussion. A given weekly topic was usually centered around which policy decisions would be most appropriate in an age where AI was smarter and more capable than humans. What would we do? How would future humans earn a living? Would AI exterminate us? What would be the purpose of higher education? Would we merge with AI and become hybrids? Would society split into AI-friendly nations? What did that mean for x-person’s current situation?
During these meetings, everyone would get up to 2-minutes to ask a question, provide some insight, propose a solution, and/or express concerns. You could pass and save your time for a longer discussion later in the session. I usually kept the books on that with a notepad. If you interrupted someone, you lost your turn. We had doctors, lawyers, programmers, startup entrepreneurs, marketing professionals, government employees… the occupational diversity was quite substantial. Everyone brought something new to the table with each turn building off the last. Usually, there was a core group of three coders, including myself, and the others were semi-regulars, usually popping in once a month or so.
Elon Musk
One day in 2015, Elon Musk said that he’d been frightened by something he’d seen with a new development around AI. This is when he invested in OpenAI, which was working on a technology that would become the GPT-4 of today. In his worry, he exclaimed that governments needed to get involved immediately. Most waved him off as crazy. Many still do today. A lot of people in our sessions also thought he was crazy. A lot of the coders had seen AI/worked on it, and they thought AI becoming as advanced as Musk claimed was at least 2–3 decades away.
Yet, here we are, 8 years later, and generative AI is passing bar exams and writing whole books.
Like transformers (behind OpenAI’s tooling), I can’t attribute the following concept to a specific instance. There are many experiences during these roundtables, building AI-based IoT devices, developing NLP technology for call centers in my early-20s… even working now on AI-baased products to solve problems around models and valuations in M&A and capital raise transactions. However, there was a specific instance, that directly led to the new idea of a Spotify-like revenue-sharing platform for professionals. For example, if I had to compensate each person individually who I’d worked with in the past for any commercial success I have today in any particular engagement: that might prove quite difficult. In fact, it is like this for GPT-4, as well. However, what if you could measure and attribute that knowledge that you gained before and weigh its impact on generated revenue?
A few days ago, we unveiled GiveFlag by Giving Desk. A tool that started as a loose set of internal tools for helping us analyze the risks and valuations of companies. As we sat through meetings with a number of businesses involved in acquisition and investment negotiations, we realized an inefficiency. GiveFlag was deployed as a way for us to explain complex financial models, do scenario analysis, and help our portfolio companies (and partners) fulfill their goals around deals… but at scale.
The Other Problem
Small to medium-size businesses often find it inhibitive to benefit from the management accounting and risk analysis capabilities that larger firms have often built-in. We figured we might put the new tool to the test and get feedback from financial analysts and other professionals to improve the tool. One thing that came up was that a lot of people didn’t want AI to do a lot of the things GiveFlag and other AI tools are capable of doing today (and getting better at). There is a fear of job loss in the market, certainly. It’s palpable.
However, expert feedback was leveraged with regards to this particular launch instance and really enhanced the value of the model. In this particular case, the model was evaluating a bank (SVB) like a normal business, among other things. Essentially, the model lacked some banking experience and nuance as it analyzed SVB financial statements (SEC filings) “by the book”. However, as it was trained on the feedback from experts, it improved quite rapidly.
But this was a commercial product. Shouldn’t those contributors/experts be compensated? Surely. Therefore, we launched an Expert Partner program quickly after. This program is in its early stages, but it allows experts to enhance our models by looking at the financial statements and then reviewing the GiveFlag AI models’ analyses. As the product gains traction and delivers value, the expert partners benefit accordingly. This is how it should be. This is done by applying another AI model that analyzes model output improvements and measures how previous contributions led to better results. It’s quite complex and very early, but practical.
Conclusion
In summary, while it’s hard to attribute how any experience improved our lives, it is currently about as challenging for GPT-based tools to do this. However, by creating expert network ecosystems, people with specific knowledge-bases around various topics can advance GPT-based models (and other machine learning solutions), receiving royalties perhaps into perpetuity. Maybe even passing those assets on to their children and grandchildren. This could reduce the impacts of potential mass unemployment caused by highly advanced AI.
