The Future of AI // not boring
AI shouldn’t be considered as monolithic thing, there are already useful tools, but expectations are too high
Fundamental limits of any intelligence (natural or artificial) shouldn’t be forgotten, despite any hype and marketing.
The future of AI isn’t boring. There are many useful AI tools and services, many more will be built, but expectations are too high, above the fundamental limits.
There is no AI vs NI, the future of AI is to be helpful and useful for NI.
Here is a summary of the key points from the talk:
The speaker, Dr. Melanie Mitchell, discussed the past, present, and future of artificial intelligence (AI). She provided a historical overview, tracing AI’s tumultuous early years marked by overly optimistic predictions and periods of disillusionment known as “AI winters.”
The rise of machine learning, especially deep learning and neural networks, led to major breakthroughs and renewed optimism around 2010. However, these systems still exhibited failures of true understanding and brittleness when faced with novel situations.
The advent of large language models like GPT-3 and image generators has been astounding but has also raised questions about the nature of their “intelligence” compared to humans. There is ongoing debate over whether scaling up these models can lead to general intelligence akin to humans.
While holding hopes that AI may revolutionize fields like science and medicine, Dr. Mitchell expressed concerns over AI magnifying biases, fueling misinformation, disrupting jobs, and being entrusted with tasks beyond its competency. Key challenges include developing AI that truly understands the world and human values.
Ultimately, she emphasized that the future of AI is not inevitable but something society must thoughtfully shape together for the benefit of humanity.
Not boring things // a new paradigm for generative AI, generative interactive environments (Genie)
We introduce Genie, the first generative interactive environment trained in an unsupervised manner from unlabelled Internet videos. The model can be prompted to generate an endless variety of actioncontrollable virtual worlds described through text, synthetic images, photographs, and even sketches. At 11B parameters, Genie can be considered a foundation world model. It is comprised of a spatiotemporal video tokenizer, an autoregressive dynamics model, and a simple and scalable latent action model. Genie enables users to act in the generated environments on a frame-by-frame basis despite training without any ground-truth action labels or other domain-specific requirements typically found in the world model literature. Further the resulting learned latent action space facilitates training agents to imitate behaviors from unseen videos, opening the path for training generalist agents of the future
A Foundation Model for Playable Worlds
The last few years have seen an emergence of generative AI, with models capable of generating novel and creative content via language, images, and even videos. Today, we introduce a new paradigm for generative AI, generative interactive environments (Genie), whereby interactive, playable environments can be generated from a single image prompt.
Genie can be prompted with images it has never seen before, such as real world photographs or sketches, enabling people to interact with their imagined virtual worlds-–essentially acting as a foundation world model. This is possible despite training without any action labels. Instead, Genie is trained from a large dataset of publicly available Internet videos. We focus on videos of 2D platformer games and robotics but our method is general and should work for any type of domain, and is scalable to ever larger Internet datasets.
Generalist Embodied Agent Research
