A 386-page bid from Stanford has described the intricate and ever-evolving trends of Artificial Intelligence. The adept researchers of the Institute of Human-Centered Artificial Intelligence joined hands with scholars from private industries and academia for this report. Leveraging this joint venture, the team has gathered a bunch of value-driven information and predictions on the AI matter.
While this initiative may not be the latest one to analyze the AI domain, this survey seems to be an important one as it helps individuals figure out what’s going on in the industry. It’s being said that the team has already started working hard to lay out the next updates on the AI happenings and modules.
Details Included in the Report
The said report closely monitors new foundation models and analyzes their geopolitics and training costs. In addition, it also covers advanced topics like K12 AI education, the impact of AI systems on the environment, public opinions on AI, AI policies in different countries, etc.
According to the report, AI development has evolved dramatically over the last decade.
From an academia-led domain to an industry-led trend followed by a substantial margin, the advancement shows no sign of taking a pause. The report also claims that it’s gradually becoming challenging to scrutinize models on conventional benchmarks, and this shortcoming calls for a new paradigm.
The report adds that the energy footprint of AI training and utilization has become considerable. However, it is yet to be discovered if it relates to additional efficiencies. The study has also monitored the AI controversies and incidents which took place in the last decade. According to the researchers, the said controversies have somehow been raised by a factor of 26 since 2012.
The report has found that job postings and AI-related skills are indeed increasing. However, the pace is not as fast as people think.
Leveraging AI’s growth, policymakers and businesses are trying to make some quick money. Investment in AI ventures has been slowed down temporarily after an unimaginable rise over the last decade.
The report has also revealed China, Saudi Arabia, India, and America’s take on the advancement of AI. While approximately 70% of participants from China, Saudi, and India find AI extremely useful, only 35% of Americans are ready to praise human competitive intelligence.
More Findings and Suggestions
Besides the ones mentioned above, the report also covers a lot of other topics and sub-topics in a readable and user-friendly language. For instance, chapter three of the report titled ‘Technical AI ethics’ highlights the well-talked bias and toxicity matter. According to the report, these two metrics are challenging to reduce.
The report also illustrates different AI updates, including controversies and incidents, through bullets and diagrams.
However, the tests suggest unfiltered models are more prone to lead individuals to the problematic territory. To address this issue, it’s crucial to employ instruction tuning.
It will help add an additional layer of preparation, passing the outputs using a second mediator model. This may not remove the said issues holistically. However, it may improve things to some extent.
The report reveals that better-performing language models are usually more gender biased. While the reason behind this tendency is undiscovered as of now, the fact states that the optimization of AI is probably not going to be as straightforward as society’s expectations.