"The Near Future of AI: Action-Driven Platforms and Better Rules for the Internet"

"The Near Future of AI: Action-Driven Platforms and Better Rules for the Internet"

In recent years, the field of Artificial Intelligence (AI) has made significant advancements, and its impact on our daily lives is becoming more apparent. From language models to question-answering systems, AI has the potential to revolutionize how we interact with technology. However, to unlock its full potential, we need to shift our focus towards action-driven platforms and better rules for the internet.

One of the key findings in AI research is the effectiveness of Language and Learning Models (LLMs) when prompted to "think step by step." Researchers have discovered that LLMs perform better at question-answering tasks when they break down the problem into smaller, actionable steps (Kojima et al. 2022, arxiv). This approach allows the model to understand the problem more thoroughly and provide more accurate responses.

But what if LLMs were given external cognitive assets? These external resources could include search engines, code interpreters, or even chats with humans. By incorporating these cognitive assets into the thinking process, LLMs can enhance their performance and provide even better results. The model needs to understand the power of its own tools and grasp the desired outcomes of the user.

ReAct, a framework that takes three iterative steps - Thought, Act, and Observation - leverages external cognitive assets to supercharge the model's capabilities. By combining the thinking process, choice of action, and observation of outcomes, LLMs can optimize their performance. This approach opens up new possibilities for reinforcement learning, where the system can be trained to produce better results by measuring a specific metric of interest.

On the left side, we have the External Cognitive Assets that can enhance the model's power. These assets can be any function that takes text as input and provides text as output. For example, incorporating searches, code interpreters, or engaging in conversations with humans can provide valuable insights and expand the model's capabilities.

However, the real challenge lies in task-oriented training. While techniques like instruction tuning seem straightforward to implement, developing effective training methods remains a complex task. It requires a careful balance of providing the model with the necessary tools and ensuring it understands the desired outcomes. The development of robust and comprehensive training methods is crucial to make action-driven AI platforms work seamlessly.

Shifting our focus to the internet, we need to address the underlying issues that contribute to the current state of online discourse. The key to creating a healthier online platform lies in flipping the power dynamic and giving people the control over what they pay attention to. Instead of algorithms dictating what content users see, we should empower individuals to curate their own feeds based on their preferences and interests.

The root of the problem lies in the business model that sells people's attention to advertisers. This model incentivizes companies to prioritize content that manipulates people's emotions and captures their attention. To foster a more thoughtful, civil, and intellectually diverse discourse, we need to explore alternative business models that allow people to choose to pay with money instead of attention. By doing so, we can create an environment where platforms serve the interests of the users rather than the advertisers.

Merely demanding stricter regulations or giving social media giants more curating power is not the solution. We need a fundamental shift in the business models that drive these platforms. Instead of relying on attention-driven models, we should seek to create profit-generating models that prioritize serving the users. This change will not only benefit the consumers but also foster a healthier online ecosystem.

In conclusion, the near future of AI lies in action-driven platforms and better rules for the internet. By incorporating external cognitive assets and leveraging reinforcement learning, we can unlock the full potential of AI systems. Simultaneously, by shifting the power dynamic and fostering business models that prioritize user interests, we can create a more inclusive and thoughtful online environment.

To facilitate this future, here are three actionable pieces of advice:

1. Invest in the development of comprehensive training methods for action-driven AI platforms. This will enable the models to understand the power of their own tools and align their actions with the desired outcomes.

2. Explore alternative business models that allow users to pay with money instead of attention. By incentivizing platforms to serve users rather than advertisers, we can promote a healthier online discourse.

3. Support research and development efforts that focus on the ethical and responsible use of AI. As AI becomes more integrated into our daily lives, it is crucial to ensure that it is used in a way that benefits society as a whole.

By implementing these actions, we can shape a future where AI and the internet work in harmony, empowering individuals and fostering positive change.