Future of AI in 2024

Towards the end of 2023 experts from Stanford's HAI had released a set of predictions on what they expect in AI in 2024. It has just been 2 months into the year and as I was reading these once again, I could not help but think that we may need a new set of predictions for the second half of the year, especially in areas related to the capabilities of the new Gen AI models. As we are releasing new models and raising the bar every week, I suspect that we soon will need a new set of paradigms to even make sensible predictions into the future of AI. Scary??

?Here are a few of the HAI predictions with an overlay of where I think we will be headed into the second half of 2024.?

White Collar Work Shifts:?

Mass adoption of AI by companies will start delivering productivity benefits that have been long anticipated. Knowledge workers, including creative professionals, lawyers, and finance experts, will see changes in their jobs. AI won't completely automate jobs but will augment and extend what we can do, making our work more efficient and enabling new possibilities. ?

Gemini 1.5 models with 1 million token context windows (tested with 10 mil) can accelerate the adoption in almost all areas of knowledge work.?

?

Deepfake Proliferation:?

Expect the emergence of big multimodal models, particularly in video generation. However, this also means we need to be vigilant about deepfakes. Both startups and big boys like OpenAI and Google will continue releasing larger models with new capabilities. ?

It is Feb and we already have Sora which is pushing the boundaries of our imagination.?

Functional AI Agents and Multimedia AI:?

Expect more functional AI agents capable of executing end to end tasks like making reservations and planning trips. Progress in multimedia AI, especially video processing, could revolutionize data interpretation and usage. ?

2023 was the year of chatting with AI, 2024 could be the year of doing with AI!?

?

GPU Shortage:?

The global shortage of GPU processors, crucial for running large AI applications like LLMs, will pose serious challenges as companies and countries seek in-house AI capabilities leading to high demand. Innovators must create cost-effective alternatives. There is a lot of research already happening in this space to create alternative processing paradigms including low-power solutions. This will be a priority to democratizing AI access.?

Sam Altman is seeking up to $7 trillion to overhaul the semiconductor manufacturing industry including the large-scale production of AI chips.?

?

Privacy Challenges in the AI Era:?

Privacy and data protection legislation will significantly impact AI development. Data remains the foundation of all AI systems, and developers' hunger for training data will continue to grow. Existing privacy laws are insufficient to address data acquisition races and resulting privacy harms at both individual and societal levels. Legislation around AI is also on the horizon; the EU is finalizing widespread AI rules, while the U.S. may not see major regulation due to an election year. ?

Recommendations to mitigate privacy risks:?

???? - Denormalize data collection by default (shift from opt-out to opt-in).?

???? - Focus on the AI data supply chain for transparency and accountability.?

???? - Develop new governance mechanisms and technical infrastructure for individual data rights.?

??

In summary, AI continues to evolve rapidly, with larger models, new capabilities, privacy challenges, and practical applications across various domains. The second half of 2024 promises even more exciting developments in this fast-evolving space.?

??

Please note that these predictions are based on expert insights and ongoing research at Stanford's HAI (Human-Centered Artificial Intelligence). As always, the future remains dynamic, but these trends provide valuable context for understanding where AI is headed.?