Union Square Ventures: A picture of the four possible directions of AI technology in the future

Authors: Grace Carney, Matt Mandel, Rebecca Kaden & Nick Grossman

Compiled by: DeepTech TechFlow

In this era full of opportunities and uncertainties, the rapid development of artificial intelligence has opened countless possibilities for us, but it has also brought many profound and unanswered questions. Will we continue to move forward along the exponential growth curve of technology, and even achieve a leap in self-improvement? Or will we slow down due to the emergence of technological bottlenecks? Is the limit of technological development close at hand, or still far out of reach? Will the future world be dominated by a closed ecosystem led by a few giants, or an open ecosystem of multi-party competition, cooperation and symbiosis?

The future direction and the greatest opportunity may depend on two key factors: the speed of technological progress and the openness of cutting-edge models.

The speed of technological progress: reflects the speed of development of artificial intelligence models. If the development is rapid, we may reach the critical point of "intelligent explosion" in the next few years; if the development slows down, it may be delayed by ten years or even longer due to encountering technical bottlenecks.

Openness: Measures the scope of power controlled by artificial intelligence models and the economic value extracted from it. In an open scenario, the performance of open models is close to that of closed models, or there are many closed models of similar quality in fierce competition, which means that the models cannot monopolize economic benefits. In a closed scenario, a few giants are far ahead with closed models and control the majority of value creation.

While there will be new development opportunities in every scenario, we believe that the quadrant ultimately landed in will determine which projects, strategies, and business models have the most advantage.

Large-scale basic model: lower left quadrant

In the technology stack, there has always been tension in the value distribution between the infrastructure layer and the application layer. In 2016, USV analyst Joel Monegro pointed out in his blog post 'Fat Protocols' that 'the value distribution of the traditional Internet technology stack shows the characteristics of 'thin protocols' and 'fat applications.' Market development indicates that investing in applications has a higher return, while direct investment in protocol technology usually has a lower return.' However, in the blockchain technology stack, this relationship has been reversed, with most of the value concentrated in the protocol layer. With artificial intelligence becoming the new dominant platform, we need to re-examine this relationship - whether the model itself occupies the main value, or the applications built based on the model are more valuable?

In the lower left quadrant, technological progress continues to accelerate, ultimately forming a closed ecosystem, with the model almost mastering all value. The stronger the model, the less we need applications or other models built on them.

In this scenario, the dominant players will be a few giant basic models and their required infrastructure. To maintain the expansion of these models, we need to consume a huge amount of energy (including more efficient new energy). The construction of infrastructure around hardware, software, and energy production will become an important area of value creation. For example, companies such as Radiant, Blixt, Lydian, and Glow in the USV portfolio are dedicated to innovation in energy generation and management. In addition, there are opportunities to support model architectures that adopt different technological paths, such as decentralized training, which may become the winners in the future market.

The rise of niche models: upper left quadrant

(Editor's note: Niche Model refers to an artificial intelligence model that focuses on meeting specific industries or specific needs. Unlike general-purpose models like large language models, niche models are typically optimized for specific industries, tasks, or datasets, and have higher targeting and professionalism.)

In the upper left quadrant, although the ecosystem is closed, technological progress has slowed, creating opportunities for niche models to fill the gap in basic model capabilities. Currently, we have seen this trend in multiple areas, such as biotechnology, healthcare, robotics, emotion analysis, personalized datasets, and physical world data. Some of these areas have already received investment from USV. In this case, the data is more specific and targeted, and the utility of the model is also clearer. By narrowing the scope of application and increasing targeting, the accuracy and practicality of the model can be significantly improved. In this quadrant, we will see the continuous growth of dominant niche models in many areas, becoming important building blocks.

Applications with network effects: upper-right quadrant

In the upper right quadrant, the technological advancement of the basic model has slowed down, but the ecosystem remains open. The model has certain limitations in terms of capabilities and a more specific scope of application, which provides opportunities for startups to transform AI into productized applications and build on top of the model.

This result is not unfamiliar to the venture capital ecosystem. It is not only a technological breakthrough, but also a test of imagination. The opportunity lies in redesigning our interaction with AI, thereby changing our work and life. We can foresee the emergence of new applications, support layers, and interaction interfaces in the future, completely revolutionizing the way we learn, research, provide medical services, create, entertain, socialize, execute tasks, and manage businesses. By directly reaching users, learners, patients, or customers, costs will be greatly reduced and usability will be significantly improved.

As in previous technology cycles, many developers will compete for the same opportunities. Therefore, network effects (whether through user growth or data accumulation) will be a key factor for enterprises to accumulate and protect core value.

Truly Open AI: Bottom Right Quadrant

If the model achieves rapid development in an open ecosystem, then we will usher in the most imaginative scenario: a large number of super-intelligent and continuously evolving models interacting at high frequency and high skill levels.

In this scenario, many open models will dominate their respective market areas, but these models need tools to efficiently interact with each other. In a world dominated by intelligent agents (AI agents), these agents need to trade frequently. Cryptocurrencies may become the simplest and most flexible form of digital currency between intelligent agents, which will benefit tokens themselves, decentralized infrastructure, and interaction protocols like Farcaster. It is conceivable that in the future, there will be a large number of intelligent agents with wallets and budgets, engaging in frequent transactions in a digital economy.

Summary

This framework, though simplified, undoubtedly provides us with a starting point for our thinking. Approaches that dominate in each quadrant may stand out in specific scenarios. Especially those products that can inspire new user behaviors, create entirely new value, and drive proactive user adoption will have the potential to create tremendous value.

The division of these quadrants helps our team to deeply consider, in this special period full of uncertainty, "What conditions are needed for a certain method to truly take advantage?"