Just for the sake of an intellectual exercise, imagine that through some unusual stroke of luck (or misfortune) someone finds themselves at the head of a venture which spins out of an AI-related academic project. As if one of those (usually hopeless) EU education projects actually produced something that somebody else not only wanted but was willing to pay a lot of money for. A number of things follow on from this.
Firstly, the university who (probably) made life very difficult for the people who came up with and developed the idea, probably sneered at any claim that "this is important work", or at appeals to protect key people, late in the day turns round and says "this will make us millions! it's our intellectual property". While market conditions change quickly, the university drags its feet in negotiating a handover of IP and the writing of patents. Over a year goes by, everyone tears their hair out, but eventually things are signed. Universities have become very weird organisations that ape commercial practices without really understanding why they do it, or thinking about whether it is sensible.
Secondly, a spin-out company with freedom to operate is one thing, but this needs funding. The mode of thinking for academic spin-outs is similar to the mode of thinking of academic projects - how to get funding? It should be said that VC funding cannot be gained unless you have experienced people who know how to deal with VC firms. But say, for the sake of argument (through another stroke of luck) that this is in place. The danger of this mode of thinking is that getting funding becomes the prime objective. There may be a point, however, where it is so obvious that a spin-out product is so desirable to potential customers, that the getting of funding is not a question. That raises the third question:
What kind of a business are we?
So you might have funding which might keep your operation going for a year or so before you need to be raising revenue through sales. What are the conditions for your viability? This is where an AI business is weird and interesting, and this sheds light on Elon Musk, Twitter and OpenAI.
Successful and viable businesses typically have a set of operations which produce things - products, services, etc - for a customer base which pays for those products and services. Among the different regulating mechanisms within any such business will be some kind of operational management which ensures effective coordination of the production operations, marketing and so on. Since all businesses operate within changing market conditions, all viable businesses will develop an R&D arm which is scanning the horizon for new opportunities and advising on strategy. Some business will hire software developers to develop new solutions to internal operational challenges. R&D looks to the future and potential scenarios, operations are focused on the present - there is often tension between them, and good businesses balance one against the other. Interesting to note that Elon Musk's current restructuring of Twitter is basically trying to rebalance the relationship between R&D and operations within that company (which is losing money).
An AI is a specific kind of technology. In the above scenario, it fits within a company's R&D structure. In itself, it is not about operations. Musk's OpenAI is a good example. It makes itself available as an API which can be plugged-in to the R&D operations of other businesses who will use it to automate writing tasks that would once have been a function within the operations of a company. Through adopting OpenAI services, those operations are restructured, people moved (or removed), and the operations restructured.
Now look at OpenAI itself as a business. As a business, it appears to have few customer-facing operations apart from sales and marketing. It develops and provides access to machine learning models which sit on the internet (although from a technological perspective, these models are just files which could sit anywhere - even on individual devices). Its customer-base is a community of users who integrate its services into high-end heavy usage corporate operations for which they pay subscriptions. OpenAI must maintain the scarcity of what it does (in the face of continual innovation in AI), and ensure that customers keep buying its services. That means that OpenAI's own R&D must outpace the R&D of its customers - or rather, OpenAI's customers see that a good chunk of their own R&D is best outsourced to OpenAI.
I think this is a problematic business model because effective R&D relies on having a good model of the organisation of which it is part. R&D without a concrete set of business operations attached is potentially root-less - it's not part of a viable operation, and could therefore lack coherent direction. This may be the most important reason why Musk was so keen to buy Twitter: it gives him an operational infrastructure, to which he (no doubt) believes his R&D company (OpenAI) can restructure and make profitable.
With a set of operations to manage, an AI business can grow its services and see the effect of its developments on the viability of the whole organisation. Some things will work, other things won't. Sometimes operational requirements will override whatever new innovation is suggested by R&D. Other times, the R&D is critical to maintain organisational effectiveness. Moreover, an AI business in this situation could extend its reach beyond a "host" organisation, offering services to other organisations. The only problem is that in doing so, other organisations might become competitors to the original host organisation. This requires new thinking about corporate cooperation and market competition.
This is the most fascinating question about all AI businesses. They are surrogate R&D operations without operational attachments. If an AI was a human system it would be like the pathology of when a university's management believes it is the university (see this many times!), and that the current operations (academics, administrators) could be replaced by another set of operations. Equally mad is the belief that management is generic and transplantable, as in the idea of "institutional isomorphism". Management without operations isn't viable.
But it's technological form is different - AI exists as a concrete coherent thing that provides services to R&D which can be genuinely useful. These services require R&D themselves - which is the regulatory domain of the AI company itself, but the whole thing demands some kind of operational "host". An AI company is a kind of "virus", and its best chances of preserving its viability is reproduction in other hosts. Reproduction of the AI is in the interests of the original host because it grows the AI business, but it must do so in such a way that other hosts do not become competitors to each other.
The dynamics of this are different to the traditional ways we think about organisational viability and competition. Traditional businesses compete for resources (sales, income) by acquiring market share in the products they produce. They may seek to establish monopolies by acquisition of competitors to remove threats and increase profits through creating scarcity in the market (which then requires regulation by government). But AI is presenting a dynamic of what might be called "organisational environmental endogenisation". That is to say, something in the environment which threatens the viability of organisations - AI - is endogenised (assimilated) within an organisational structure in order to transform that organisational structure so it is better able to maintain its viability and profitability. As part of maintaining its viability, growing the endogenised element and then getting it to "infect" other entities becomes a critical part of the viable operation. This is not to neutralise competition, but rather to increase the strength of the ecology within which organisations sit and within which they can continue to grow and develop better R&D operations.
There is something a bit odious about Musk. But equally, there is something important happening around technology at the moment which presents organisational questions which are unavoidable for anyone looking at the future of business, organisational viability and society. It's urgent that we think this through. I'm incredibly fortunate to be in a position where I'm grappling with this at first hand.
