Stanford Publishes 2023 AI Index on Notable Trends and Advancements in the Industry

[ad_1]

Printed: 5 April 2023, 10:31 am Up to date: 05 Apr 2023, 10:37 am

The Stanford Institute for Human-Centered Synthetic Intelligence has revealed a 386-page-long AI report on notable tendencies and developments within the AI {industry} in 2023, enabling decision-makers to take a human-first method to advancing AI responsibly and ethically.

To trace progress in synthetic intelligence, Stanford’s AI Index Steering Committee – an interdisciplinary group of specialists from throughout academia and {industry} – collaborated with numerous organizations. These embody the Middle for Safety and Rising Know-how at Georgetown College, LinkedIn, NetBase Quid, Lightcast, and McKinsey.

Key takeaways

With eight chapters masking analysis and growth, technical efficiency, technical AI ethics, the economic system, training, coverage and governance, variety, and public opinion, these are the important thing takeaways from the report:

Business has taken over academiaMost important machine studying fashions have been launched by academia up till 2014. Nevertheless, since then, the {industry} has raced forward, producing 32 important industry-produced machine studying fashions, in comparison with simply three by academia. That is as a result of giant quantities of information, computing, and cash assets that {industry} actors possess and might put money into machine studying fashions.

Number of Significant Machine Learning Systems by Sector — by way of HAI

AI fashions are quickly rising and getting pricier to trainReleased in 2019, GPT-2 was thought-about to be the primary giant language mannequin (LLM) with 1.5 billion parameters and value an estimated $50,000 to coach. PaLM, one of many flagship giant language fashions launched in 2022, had 540 billion parameters and value an estimated $8 million—round 360 occasions bigger than GPT-2 and value 160 occasions extra. LLM and multimodal fashions are rising sooner and changing into pricier throughout the board.

Estimated Training Cost of select Large Language and Multimodal Models — by way of HAI

12 months-over-year non-public funding in AI has decreased for the primary time in over a decadeGlobal AI non-public funding was $91.9 billion in 2022, a 26.7% lower since 2021. Of that, $47.4 billion got here from the US and $13.4 billion from China. Nevertheless, AI funding as an entire has considerably elevated over the last decade. The quantity of personal funding in AI in 2022 was 18 occasions higher than in 2013.

Private Investment in AI — by way of HAI

Extra companies are adopting AI, whereas AI-related job postings are growing throughout each American sectorThe variety of firms adopting AI in 2022 has greater than doubled since 2017. Organizations which have built-in AI reported decrease prices and better income. Despite the fact that many fear about having their jobs changed by AI, the variety of AI-related job postings throughout each American sector for which there’s information has elevated on common from 1.7% in 2021 to 1.9% in 2022.

Cost Decrease and Revenue Increase From AI Adoption by Function — by way of HAI

Policymaker curiosity in AI regulation is quickly rising alongside the rising misuse of AIAn evaluation of the parliamentary data on AI in 81 international locations reveals that mentions of AI in world legislative proceedings have elevated practically 6.5 occasions since 2016. The AIAAIC database, which screens cases of unethical utilization of AI, has reported a surge in AI-related incidents and controversies by 26 occasions since 2012. In 2022, some noteworthy incidents have been recorded, corresponding to the discharge of a deepfake video that includes Ukrainian President Volodymyr Zelensky surrendering and the utilization of call-monitoring know-how by U.S. prisons to watch inmates.

Number of AI Incidents and Controversies — by way of HAI

AI techniques might have severe environmental impactsLuccioni et al. (2022) discovered that BLOOM’s coaching course of resulted within the emission of 25 occasions extra carbon than a person taking a one-way flight from New York to San Francisco. Nevertheless, new reinforcement studying fashions like BCOOLER present that AI techniques can be utilized to optimize vitality utilization.

CO2 Equivalent Emissions by Selected Machine Learning Models and Real Life Examples — by way of HAI

Conclusion

Other than these key takeaways, the report additionally incorporates many different insights relating basis fashions, together with their geopolitics and coaching prices, the environmental impression of AI techniques, Okay-12 AI training, public opinion tendencies in AI, and way more.

The important thing takeaways solely scratch the floor of a few of the most urgent considerations and noteworthy tendencies within the AI {industry}. These occupied with diving deeper into any specific subject can obtain the total report.

Learn extra:

[ad_2]

Source link