After listening to the conversation between John Thornhill and Marc Filippino on GPT-3, I felt an uneasiness. I recalled Brian Arthur’s articulation of technology.
I am rephrasing here: all technology is a phenomenon packaged in a useful frame. Hence, basic and applied research in the traditional R&D model. Basic research focuses on the phenomena and applied (and later development) will focus on the frame. Think of it as steam packaged into an engine.
Having spent years researching AI and augmentation, I took a break from the field, understanding its limitation. When a new development shows up and generates a stir, as GPT-3 is, I find it my responsibility to revisit my hypotheses. As an innovation consultant, helping both the technology and the people (companies/brands) that use it, I care about language a great deal. After years in the field, working with academics in linguistics and robotics, I ascertain that no machine can do anything unstructured. That would be general intelligence.
When OpenAI, its funders, founders, or ambassadors make claims of seeing a path to generality, we must look closely and consider that claim’s soundness. Arthur’s definition could be useful.
On the phenomena level: there is nothing categorically new. The computers are faster, index more data, and run more permutations, but their design has no new modalities. The machine does not understand what it is statistically generating. You could argue that GPT 3 is a new packaging of statistics, more than machine intelligence.
On a frame level, OpenAI is introducing some newness, but not a categorical one. It goes for the commercial incrementalism that Silicon Valley is known for. Indicative by the shorthand that GPT-3 is a “super-powerful auto-complete.”
OpenAI is responsible and slow in their work, allowing a limited number of collaborators to access the code and moderate its use. FT did an excellent job reporting this story. But the chasm I am describing comes crashing when OpenAI is placed next to Turing. Turing went into the unknown, navigated darkness, and in his brilliance, designed a fundamental phenomenon we still use to this day. OpenAI is iterating on a frame, a commercial application.
The way we describe technology matters a great deal. We should ask ourselves what does intelligence mean and who is calling something intelligent? As I am writing these words, am I a writer the same way Shakespeare was a writer? In our race to relinquish human endeavors, language might be the real victim.