Academia

Universities, institutes, and government agencies undertake pure research under the traditional structure of laboratory, group, department, school. Large- and long-scale research programs concerning the conceptual and experimental content of traditional fields rise and fall over many interconnected careers. The core unit of this experience is the PhD, in which a promising student is advised by an experienced supervisor, and must come up with an independent project spanning 3-5 years that generates a novel and interesting piece of scientific or theoretical research. Methods and findings are developed and discussed in an open-source manner, and published in academic journals.

Funding for these organizations comes from teaching large numbers of undergraduate students, winning large grants from governmental, transnational, or philanthropic funding bodies, and through the interest on an institution’s endowment. The main benefits are the ability to explore deep and diverse questions freed from direct economic incentives, access to a wide and high-quality range of scientific connections (for supervision, peer-related discussion, and enlisting others to work on exciting projects). Academia still provides the fundamental training for the employees of all stakeholders in AI, and still directly contributes the most to fundamental AI research, and indirectly through advancement of adjacent fields. Universities have maintained good access to large-scale computing clusters.

Tech companies

Modern large technology companies like Google DeepMind, OpenAI, Microsoft Research, Meta FAIR house large AI research departments that contribute towards fundamental improvements and new directions in their core underlying technologies. Rather than being focused on developing specific products these research teams try to solve large and general computational or environmental problems like reinforcement learning, language prediction, or protein folding. The core unit is a team research project aimed at providing one of these solutions or extensions, usually led by a senior research scientist with extensive experience in academia and industry.

Funding comes from the operations of the parent organization, which are typically some of the most wealthy private companies in the world. The main benefits are blue-skies thinking towards large-scale and impactful goals, as well as access to large amounts of engineering support, data and compute. Employee pay is typically much greater than in academic equivalents. Findings are published and code made open-source, typically with some delay or selection to maintain economic competitive interest.

Focused Research Organizations (FROs) and Non-profits

FROs are a recent addition to the 501(c)3 charitable non-profit classification (see Convergent Research, the spearheading “FRO FRO”). They aim to research and solve scientific problems that fall outside of or between traditional academic disciplines. For example, Future House is building AI methods for science—a fundamental challenge that is difficult to address in Academia or for-profit sectors because the difficulty of the problem requires a long-lead time and a group of top-tier scientists engaged in pure research but not a traditional field. FROs are “focused” in that they progress through prespecified, quantifiable, technical milestones.

Funding comes from private donors, grant applications, and equity and revenue from downstream for-profit spinoffs. Their internal structure and incentives are similar to most start-ups, where a small team of founders grows a team of scientists to solve practical, technical problems in a fast-paced manner. Pay is competitive with technology companies.

Start-up

Start-up are nascent and growing private companies and organizations that work intensely to develop an AI prototype solving a well-specified need or problem. They then scale their core technology to better fit a wider range of people or business partners. Start-ups can have a very diverse set of goals and projects, but some key successful examples are Databricks (now Mosaic ML), which brought large ML model training and data storage to the cloud; RoundTable, which uses advanced cognitive science and ML to improve online bot detection; and Presentient Technologies, which uses advanced mathematics to determine when a clinical trial can be stopped early to divert funds elsewhere. Other start-ups like Just Ask A Question (JAAQ) provide rich data sources and user interfaces for AI to tackle global problems, like poor mental health.

Funding comes from private sponsorship and investment, and from operationalizing the business idea. Benefits include initial blue-skies and revolutionary thinking, followed by fast progress on well-circumscribed problems. Founders and first hires have the potential to accrue a large amount of money. But, start-ups usually fail.

Venture Capital Firms (VCs) and Funders

VCs provide funding to start-ups and tech companies, and pro-bono support of other kinds of entity. They aim to maximize the return on their investment by backing projects that are interesting, impactful, and marketable. To make this assay they host large internal AI teams and external consultants that conduct due diligence on potential investments. These can be extremely large funds, managing a wide range of AI-related projects like Andreessen and Horowitz, or smaller, seed-stage specialists like Fifty Years, PillarVC and Adavita.

Funding comes from previous and current successful projects and the fund managers’ other investments. Working as an AI researcher within a VC fund gives access to a wide range of interesting and changing ideas and the potential to become a partner. Or, the ability to do due diligence at good pay rates within a team and a secure professional career.