12 AI Trends to watch for in 2024

December 12, 2023

2023 Generative AI Insights

If these past twelve months are any guide, 2024 will be another year of rapid advancements in AI. 2023 saw a continued acceleration of AI capabilities and an emerging consensus around the need to develop shared guardrails for future development of this technology.

Those themes are likely to continue in 2024 according to members of the Vector Institute’s leadership team, as AI reshapes health, business, and our day-to-day lives. Harnessing their collective expertise, they’ve gazed into their crystal balls to offer some predictions for AI in the year to come. 

Empowerment of the individual

AI has never been more accessible, which means people can be empowered to leverage AI in every part of their life. Whether it is writing and editing email, developing a presentation, or creating the first draft of a process checklist, the extent to which we can automate tasks in our lives will only be limited by our creativity. However, many still worry about how AI will impact them and their future. Just like with businesses, it is important to include the public in the changes taking place to ensure people are part of the process.

Cameron Schuler, Chief Commercialization Officer, VP Industry Innovation

Upskilling & reskilling

Expect new and emerging roles to meet the changing landscape. From prompt engineering to get the most out of LLMs to auditing AI systems to ensure they are performing as intended, there are multiple opportunities to learn new skills that will enhance one’s AI toolkit.

Melissa Judd, VP Research Operations and Academic Partnerships

AI is no longer the domain of the IT department

Historically, technology has been relegated to the IT department. But generative AI offers opportunities across an entire business, making AI a top priority for the whole company. The risk of disruption and where it comes from will be far more unpredictable; leaders will have to be more vigilant in balancing the risks from competition while ensuring their own company remains competitive. This will include a move away from curiosity-driven experiments with AI to more focused strategic problem-solving. – CS


Change Management

Companies that invest deeply in generative AI and in change management for their organizations and workforces will reap the greatest rewards in 2024 and beyond. – MJ

LLMs trained on specialized, trustworthy, and debiased datasets will replace the current internet-scraping versions

The LLMs trained on internet-scraped data are subject to all of the biases and mis/disinformation available on the internet. As training datasets become better curated, LLMs trained on those datasets will be higher quality, more reliable, and less likely to produce unreliable or problematic outputs.

Roxana Sultan, Chief Data Officer and VP Health

Copyright laws will adapt to make the development and use of such LLMs feasible

The LLMs trained on internet-scraped data often include datasets that have not been consented for these types of use, which has currently been leading to lawsuits and copyright claims. Policymakers are working to quickly adapt copyright legislation to keep pace, balancing the needs for LLM training with the rights of content creators. – RS

Multimodal models will create a new AI paradigm

While LLMs are an incredibly powerful form of AI, the development of foundation models that can combine data from multiple sources (text, image, speech) will enable a whole new level of AI. Over the past year in health, a number of multimodal foundation models (models trained on diverse data sources, such as text, waveform, imaging, genotype, etc.) and generalist medical AI products have been published. Evidence to date indicates that multimodal foundation models have the potential to enable innovative new health technologies through the seamless integration of diverse data sources and modes of communication. – CS, RS

Federated learning will unlock more value from health data across systems

In health, work is underway to test federated learning — a machine learning technique that helps to preserve data privacy — approaches to training health AI models across hospitals and health networks using distinct electronic medical records and/or data systems. If successful, these models will establish proof of concept for an approach to health AI that lowers the barriers to data centralization and enables development of more robust models than those trained on data from a single centre. – RS

Small, open-source models will enhance efficiency and competitive performance for specific tasks or domains

Small, open-source models will enhance efficiency and competitive performance for specific tasks or domains. As the demand for specialized language understanding grows, these models may outshine their larger counterparts where precision and contextual relevance are paramount. Improvements in smaller models that cater specifically to niche use cases will be driven by continuous innovation in model architectures. The development of more agile and effective models will be led by open-source contributions and collaborative efforts.

Deval Pandya, VP, AI Engineering

Rise of AI Agents

AI agents — programs that make decisions based on their environment — will evolve to demonstrate enhanced context awareness, multimodal capabilities, and a commitment to continual learning, offering users more personalized and adaptable experiences. Developers will integrate ethical practices, edge computing, and industry-specific customization, ensuring the responsible and domain-specialized deployment of AI agents across diverse sectors. This evolution will redefine work dynamics as human-AI collaboration emphasizes the cooperative synergy between AI technologies and human capabilities. – DP

Go fast! No wait, slow down!

The tension between rapidly advancing frontier models and concerns over AI safety & existential risk will continue to unfold in new and interesting ways. – MJ

More Guardrails

Countries will continue to build guardrails for AI be it through voluntary codes, the development of standards and regulations. – MJ


Vector Faculty Member Gautam Kamath
Research 2024

Vector researcher Gautam Kamath breaks down the latest developments in robustness and privacy

Large Language Models

Standardized protocols are key to the responsible deployment of language models

Man typing on laptop
Generative AI

How businesses can balance AI innovation and cybersecurity