🧠 10 AI Predictions for 2024

Subscribe | Contact | Meet The Team

Welcome back AI prodigies!

In today’s Sunday Special:

• 📈Macrotrends

• 👨‍⚖️AI v. Copyright Law

• 📚Bookkeeping

Read Time: 6 minutes

🎓Key Terms

• Foundation Model: machine-learning models trained on a broad spectrum of unlabeled data capable of performing a wide range of text and image generation tasks.

• Vision-Language-Action (VLA) Model: machine-learning models trained on internet-scale amounts of visual data and natural language to control physical robots.

• Copyright Act of 1976: provides the current framework for U.S. copyright law, codifying the doctrine of “fair use” and extending copyright laws beyond a fixed period of the creator’s lifetime.

📈MACROTRENDS

1. VLAs Eclipse LLMs: In today’s discourse, Large Language Models (LLMs) are synonymous with AI models since the most popular consumer-facing applications are text-only (e.g., OpenAI’s ChatGPT). However, as AI becomes increasingly multimodal, LLMs will become less exciting; the diminishing returns of adding parameters and the need for specialization will supersede chatbot hype. Consider AlphaFold, an AI model trained on the amino acid sequences and molecular structures of known proteins to predict the formation of unstudied proteins with “atomic accuracy.” AlphaFold eliminates the costly and time-consuming step of mapping protein structure to develop new vaccines, drugs, and chemistry-based solutions, like how to break down plastic waste. Or consider emerging foundation models in robotics. For example, Google Deepmind’s Robocat is an agent that learns new tasks with as few as 100 demonstrations. Trained on videos and simulations of robots performing the new task, in addition to text-based instructions, it represents a new class of AI called Vision-Language-Action (VLA) models. In 2024, as models become increasingly multidimensional, so will the speed of training, the breadth of use cases, and their descriptors.

2. Products > Demos: In 2023, generalist generative AI models like OpenAI’s ChatGPT (text-to-text) and DALL-E 3 (text-to-image) were the first chatbots to reach 1 million users. They were entertaining, educational, and mildly productive but failed to provide a unique value proposition. Generalist models are testing platforms; they cannot revolutionize healthcare, cybersecurity, driving, realty, decision-making, or replace millions of jobs. Specialized ones can. Launched in June 2022, GitHub Copilot—one of the fastest-growing coding tools—autocompletes code, generating over 3 billion lines for over 20,000 organizations. While demos capture the public’s imagination, products provide tangible economic value. GitHub Copilot is a preview of the next stage in AI’s lifecycle. AI startups must prove product-market fit, differentiation, and long-term profitability, especially after underwhelming venture capital returns.

3. CAIOs Proliferate: As discussed in October, Biden’s executive order requires all 400 federal agencies to appoint a Chief AI Officer (CAIO). To some, this is a microcosm of the excessive bureaucracy in Washington; simply creating an AI position and a corresponding department before identifying AI’s precise value to each agency puts the cart before the horse. Nevertheless, AI czars are coming for the private sector. Just like Chief Cloud Officers (CCO) were the hottest new executives in the 2010s, CAIOs will be the darlings of Fortune 500 companies and their mid-sized counterparts. Since AI is still an infant in corporate America, CAIOs will ensure appropriate ethics, safety, security, and governance guide implementation, all while reassuring skeptical shareholders.

👨‍⚖️AI V. COPYRIGHT LAW

4. AI-related litigation will characterize 2024. Does training generative AI models on copyrighted internet content violate copyright law? It depends on how courts apply the Copyright Act of 1976 and the fair use doctrine’s four-factor test:

   1. The purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.

   2. The nature of the copyrighted work.

   3. The amount and substantiality of the portion used in relation to the copyrighted work as a whole.

   4. The effect of the use upon the potential market for or value of the copyrighted work.

For generative AI models, fair use analysis has two distinct components: whether these systems infringe by incorporating copyright-protected material into the training process and whether generative AI outputs infringe the rights of those same copyright holders. First, copyright owners will undoubtedly argue that using copyrighted materials to train generative AI systems without authorization violates the right to control reproduction, a fundamental right of the Copyright Act of 1976. Given the Supreme Court’s recent guidance in Andy Warhol Foundation v. Goldsmith, the application of factor 1 will hinge upon the factual question of whether there is a real-world market for the right to include works in a generative AI dataset and whether the outputs of that AI system effectively compete with copyright owners. This question is the crux of Stability AI’s lawsuit against photo licensor Getty Images, alleging its generative AI model “competes directly” with Getty’s licensing marketplace for “those seeking creative imagery,” including in generative AI datasets.

Second, whether output from generative AI systems infringes on copyrighted work is likely to hinge on whether it’s “substantially similar” to the original work, as stated in factor 2. The variety of generative AI outputs by modality, length, realism, and over time will complicate matters further as model outputs deviate from copyrighted works on multiple dimensions but rely on them nonetheless. To answer factor 4, scholars will likely determine whether outputs can substitute original works and how they imitate style. Do they regurgitate relevant portions of the underlying works, disconnected from the original creator’s name, image, and likeness, or generate new outputs without using “meaningful” elements from the originals? Only two things are certain. The answers to these questions will be driven heavily by their effects on the market for copyright-protected works. And courts in different jurisdictions will reach different conclusions, so the issue will go to the Supreme Court.

📚BOOKKEEPING

5. Extremely Likely: GPU Shortage Continues Amid Unprecedented Demand!

6. Very Likely: Deepfakes Proliferate Amidst National Elections (e.g., U.S., Mexico, India, and Taiwan).

7. Likely: AI-Generated Video is the AI-Generated Text of 2024, Prompting Disclaimers on YouTube.

8. Unlikely: Federal AI Regulation During an Election Year, Despite Dozens of Bills Introduced.

9. Very Unlikely: Chatbots Eclipse Search Engines in Information Retrieval, Given Hallucinations.

10. Extremely Unlikely: Wearable AI Begins to Replace Smartphones, Following Humane AI Pin’s March 2024 Release.

🩺 PULSE CHECK

📒FINAL NOTE

If you found this useful, follow us on Twitter or provide honest feedback below. It helps us improve our content.

How was today’s newsletter?

• ⭐️⭐️⭐️⭐️⭐️Nailed it!

• ⭐️⭐️⭐️Average

• ⭐️Epic fail.

❤️AI Pulse Review of The Week

🎁NOTION TEMPLATES

🚨Subscribe to our newsletter for free and receive these powerful Notion templates:

• ⚙️150 ChatGPT prompts for Copywriting

• ⚙️325 ChatGPT prompts for Email Marketing

• 📆Simple Project Management Board

• ⏱Time Tracker