Currently, many different organizations do cutting-edge AI research, and some innovations are shared freely (at a time lag) while others are kept private. I'm referring to this state of affairs as 'multipolar,' where instead of there being one world leader that's far ahead of everyone else, there are many competitors who can be mentioned in the same breath. (There's not only one academic center of AI research worth mentioning, there might be particularly hot companies but there's not only one worth mentioning, and so on.)

But we could imagine instead there being one institution that mattered when it comes to AI (be it a company, a university, a research group, or a non-profit). This is what I'm referring to as "monolithic." Maybe they have access to tools and resources no one else has access to, maybe they attract the best and brightest in a way that gives them an unsurmountable competitive edge, maybe returns to research compound in a way that means early edges can't be overcome, maybe they have some sort of government coercion preventing competitors from popping up. (For other industries, network or first-mover effects might be other good examples of why you would expect that industry to be monolithic instead of multipolar.)

It seems like we should be able to use insights from social sciences like economics or organizational design or history of science in order to figure out, if not which path seems more likely, how we would know which path seems more likely.

(For example, we may be able to measure how much returns to research compound, in the sense of one organization coming up with an insight meaning that organization is likely to come up with the next relevant insight, and knowing this number makes it easier to figure out where the boundary between the two trajectories is located.)

  • $\begingroup$ As research is progressing in various organisations(Specifically);it's at this stage(Young stage) that we cannot decide whether country/single institution/organisation is multipolar or monolithic.Accept the one with more goals or visions to shape civilisation.Therefore,one might think according to the globe perspective, that someday AI will be monolithic basing on who owns brilliant visions or goals in terms of Applicability of broader AI safety insights to multipolar outcomes. $\endgroup$ – quintumnia Feb 10 '17 at 9:56

Oligopoly vs Monopoly

The terms in the question, multipolar and monolithic, appear to be referring to the micro-economic concepts of oligopoly and monopoly respectively. Although these concepts are not AI specific, they certainly apply to such development in the way the question suggests. Leading AI R&D is occurring in a relatively small number of corporate, governmental, and university research labs.

A simple search for Oligopoly vs Monopoly will cover the differences between these two scenarios sufficiently for all sectors of products and services, and those concepts and observations apply to a large degree in the AI case.

Those who have not studied economics or political science might incorrectly conclude that oligopoly is necessarily related to capitalistic competition and monopoly is related to communism. However, R&D monopolies frequently exist within capitalistic economies and oligopolies frequently exist in a planned economies, such as communist states.

Strengths of Each

Research oligopoly strengths:

  • Lead to diversity of approaches and target products and services
  • Lower the probability of ethical, environmental, or social technology abuses because multiple independent teams, even with corporate confidentiality are in place, exert the force of at least a modicum of accountability

Research monopoly strengths:

  • Efficient in terms of return on research investment for two reasons (1) the division of a fixed amount of available research funding between several labs can increase the R&D required to produce usable results, and (2) unfruitful approaches are more quickly abandoned when their workers can be shifted to other teams within the same organization
  • Attract all those who have the most talent and experience in the fields needed by the specific R&D into one place to potentially collaborate

Predicting Transitions Between Monopolistic and 'Oligopolistic' Forms

The leading edge of any R&D tends to begin as a monopoly or a 'biopoly'. Three or more units beginning the same research at the same time or being funded by government at the same time (as in planned economies) is unusual.

Over time, employees tend to leave and start their own organizations, more so in capitalistic societies, but also in planned economies, where government may hear cases to branch off and grant permission or the scientist may, in one way or another, seek relocation to a free economy.

We can see these trends throughout technological history, back to early forms of writing, crop irrigation and plowing, chariot development, and ship building. Modern examples include the application of the Westinghouse-Tesla model of electrical energy distribution and sales, genetic engineering, speech recognition, and HTTP clients (browsers starting with Mosaic, then Netscape, then IE, then Opera, Chrome, and others).


How far ahead one global AI leader is than another cannot be so easily predicted. Some companies choose to monetize R&D results sooner in the research cycle than others, depending on their strategy and the diversity of their revenue stream. For instance, as of this writing, Apple and Facebook can wait a decade to release to the public manifestations of R&D without jeopardizing their financial status.

Furthermore, governments may declassify material (another form of what the question calls time lag) several decades after discovery. In the case of the Von Neumann model for achieving critical temperature and pressure of fissile or fusion materials, the government may never declassify it. What the NSA or equivalent organizations outside the United States may decide must fall under an equivalent strategic wall of secrecy cannot be known outside of treasonous disclosure.

The insight in an organization is not sustainable in practice. Most people do not even know where first insights occurred and two whom they occurred first because the current icons in business related to the products and services stemming from the insight are otherwise completely unrelated to origins.

Few would buy a new car and thank Isaac Newton or even Henry Ford. Absolutely no one buys a cell phone and thanks Alonso Church, Claude Shannon, or Gene Roddenberry when the first call or SMS message goes through. We don't thank the water wheel researchers of Fourth Century Alexandria when we turn on a light, or even Nicola Tesla.

We do in some ways worship computer technology entrepreneurs when we buy an iPhone or require a document in docx format and vaguely understand that it is a MS Office document type. Few know that not a single innovation we attribute to those entrepreneurs came from within their respective organizations. Everything from personal computing, desktop publishing, touch screens, and windowing interfaces were developed in the R&D facilities of other corporations.

In some ways, the company to come up with the next insight is less likely to be the last one that did. Innovators usually find better pay after they have a choice item on their resume like, "Invented first adaptive cell tower switching protocol for mobile device communications." That person is not only no longer working for the company that filed the corresponding patents, but is retired, and the company is now struggling and without recent productized innovation.

Insurmountable Competitive Edges

Thus far competitive edges have always become surmountable. The proliferation of nuclear weaponry, now spanning at least seven countries, is a current and certainly important example of this. German and Japanese aeronautics R&D edges are another.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.