Stanford: AI experts fear catastrophe – 4/11/2023 – Tech

A new report from Stanford University shows that more than a third of AI (artificial intelligence) researchers polled believe that decisions made by technology have the potential to cause a catastrophe comparable to nuclear war.

The data was obtained in a study carried out between May and June 2022, before the popularization of generative AI platforms, such as ChatGPT.

The pessimism of 36% of scientists in the field in relation to advances in the sector is, however, just one of the trends observed by the AI ​​Index Report 2023. The report compiled studies, quantitative and qualitative research to point out the trends in artificial intelligence in the last year.

The sixth edition of the AI ​​Index Report is divided into chapters covering eight aspects of the theme, such as “Research and Development”, “Technical Performance”, “Economy”, “Governance” and “Public Opinion”.

Turning to the economy, the study shows that despite the excitement, global private investment in AI declined for the first time in a decade — it stood at $91 billion (R$ 462 billion) in 2022, a drop of 26.7% compared to 2021. Despite the retraction, the amount is still 18 times greater than in 2013, with the US being the largest investor.

In 2022, private investment in AI in the country was US$ 47.4 billion (R$ 240.9 billion), about 3.5 times the amount invested by China, which appears in second position with US$ 13.4 billion (BRL 68 billion).

The same study that sought to know the opinion of researchers in the field, carried out by NYU (New York University) and published in the Stanford report, also shows that 73% agree that artificial intelligence could soon revolutionize society.

However, most researchers believe that private companies have too much influence in the sector. For example, in 2022, companies created 32 relevant machine learning models, compared to just three produced by the academic community.

Ray Perrault, co-director of the AI ​​Index, told IEEE Spectrum magazine that he finds the results interesting because they represent people familiar with the subject. “This data was obtained almost a year ago. It would be interesting to see it now, given what has been happening,” he said.

The report also shows that governments are increasingly interested in AI. Analysis of legislative records from 127 countries showed that the number of enacted bills containing “artificial intelligence” grew from just one in 2016 to 37 in 2022.

Furthermore, the amount of AI-related spending by the US government has increased by about 2.5 times since 2017.

See the main conclusions of the AI ​​Index Report 2023, by area:

Research and Development

• US and China had the most cross-border collaborations on AI publications from 2010 to 2021, although the pace of collaboration has slowed;
• The total number of AI publications has more than doubled since 2010;
• Large Language Models (LLM) are getting bigger and more expensive.

technical performance

• AI continues to deliver state-of-the-art results, but year-on-year improvement in many tests remains only marginal;
• Generative AI is gaining popularity, but these platforms can be prone to hallucinations, producing incoherent or false responses, and are still difficult to trust for critical applications;
• Language models have continued to improve their generative capabilities, but new research suggests they still struggle with complex planning tasks;

ethic

• Generative models arrived, and with them came their ethical problems;
• The number of incidents related to AI misuse is increasing rapidly;
• Interest in AI ethics continues to grow.

Economy

• Demand for AI-related professional skills is increasing in virtually every industry sector in the US;
• In 2022, the most invested AI focus area was healthcare ($6.1 billion), followed by cloud data management and processing ($5.9 billion) and fintechs ($5.5 billion);
• AI is being deployed by companies in a variety of ways;
• AI tools like Copilot from Github are helping workers in a tangential way.

Education

• The proportion of new computer science doctoral graduates at US universities who majored in AI jumped from 10.2% in 2010 to 19.1% in 2021;
• In 2021, 65.4% of AI specialists (PhD) worked in AI companies, more than double the 28.2% who worked in academia.

Politics and governance

• Policymakers’ interest in AI is increasing. An analysis of parliamentary records on AI in 81 countries also shows that mentions of AI in global legislative processes have increased by almost 6.5 times since 2016;
• In 2022, there were 110 AI-related legal cases in U.S. state and federal courts, about seven times more than in 2016.

Diversity

• Computer science students in North America who are pursuing bachelor’s, master’s, and doctoral degrees are becoming more ethnically diverse. In 2011, 71.9% of new bachelor’s degree computer science graduates were white. In 2021, that number dropped to 46.7%.

Public opinion

• Chinese citizens are among those who feel the most positive about AI products and services. According to the 2022 Ipsos survey, 78% of Chinese respondents said they agreed that AI brings more advantages than disadvantages. In the US, only 35% agreed. In Brazil, the proportion was 57%;
• Men tend to feel more positive about AI products and services than women;
• The NYU study also points out that, for 41% of researchers, natural language processing (NLP, its acronym in English), such as OpenAI’s GPT, should be regulated.