Day 1 Of The New GPT-3 Pricing Model: Are pre-AGI Technologies Common Goods?
The outcry was almost palpable, certainly audible, across the globe’s digital communities. It seemed that the spectre of paid pre-AGI API access rallied AI and automation critics, SJWs, economists, artists, writers and ML Engineers around the world against OpenAI to form a united front against what was perceived as a restricted pricing model barring everyone from access to the promised land.
What is the fuss about?
GPT-3, the world’s most massive transformer-based language model developed by openAI and released earlier this year, has seen a slew of innovative applications developed that leveraged the state-of-the-art zero to few-shot learning capabilities that ranged from automated web app development and deployment, text-based adventure games, advanced conversational bots, e-mail writing assistants (I would seriously launch this as a product) many, many more.
A lot of these applications have proven to be oil on the campfires around which naysayers that criticize AI and advanced ML tools as an existential risk congregate to spin up dark visions of our future where AI displaces humans, relegating us to a trifle and hollow existence.
I, for one, welcome our new neural overlords.
The pricing model for GPT-3 is not in any way revolting, revolutionary or in any other way questionable. Yes, I just said that. And no, I didn’t get beta access to GPT-3 in exchange for these words — I never had beta access in fact.
The API economy has been fast-growing sector of the wider digital economy for at least the last decade, if not longer. With SaaS products covering mainstream end-user needs via a subscription model for the last 6 to 7 years, nobody can claim the freemium pricing to be outrageous.
With the free tier, research and development on the most advanced natural language processing model is available to anyone with a stable internet connection and a enough time (barring economic sanctions against certain IP address ranges). There is no disputing that. The discussion about pricing and even about the perceived treachery of OpenAI “selling out” to Microsoft is a proxy for a deeper issue: what aspects of technology should be public goods and available for all and what aspects should follow a market logic?
Business Ethics, not AI Ethics, are the problem
The yearning for clarity about what is coming for us is as deep-seated as the need for belonging to a tribe. OpenAI was launched on the premise that developing AGI technology is an important step on the evolution of humankind and that pre-AGI technologies are going to be the stepping stones toward that future. The company brands itself as a fair and ethical actor in the space of AI research, claiming
“[OpenAI] will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome.”
The reasoning that the outraged give the public is that the pricing tier of the GPT-3 API contradicts the claims in the OpenAI charter. This debatable position misses one important aspect: OpenAI never had a choice. Irrespective of management, ownership and corporate benefactors: in the context of a capitalist market economy any organization that incurs significant R&D costs needs recoup these costs within a finite time horizon. The idea that cool, paradigm-shifting technology should not rest in the hands of the few and powerful is a romantic ideal that never really occurred in real life: Nikola Tesla was very tolerant of the illegitimate use of his patents and inventions on a global scale. What Earth’s atmosphere is to carbon emitters Tesla was to businesspeople: a means to externalize costs that would otherwise be cost-prohibitive under existing market realities.
The only way such massive R&D projects can be given to the public for free is by externalizing the cost of development: through philanthropy, which might be a hard sell in the OpenAI case, through tax collection (which would ruffle the feathers of Unions and civil rights activists) or by letting everyone chip in and become a stakeholder — which is cost-prohibitive mostly from a legal standpoint. The reality of how business is done in our current economic paradigm forces the hand of the well-intentioned to either think small or monetize early.
Roasting Microsoft or OpenAI for not taking a decade-long time-horizon into account when making funding and ROI decisions is silly. Until the moment that advances in AI become state-subsidized and instantly public-domain classical business ethics will define what we get to see and how we get to use it, thus making the process of AI development more predictable but also less accessible. Making the entire process of value creation open to radical, global participation in financial and technical terms is the only way we can get to the other side of te chasm. It may not be obvious to you but the foundation for such a reality are being laid as we speak — but not where you would expect them.
Tesla’s works (and the person) were gifts to the public — a public that did not know how to honour and govern knowledge that should be part of a global Commons of Knowledge. The freemium model for GPT-3 is a much, much smaller gift than promised — but it is still an incredible gift that can lead to incredible practical discoveries in automation. Use it to validate your use-case — then submit to the invisible hand of the market to see if it agrees with you.