【美国科学家用AI预测犯罪:犯人施暴前先服刑!?】

【美国科学家用AI预测犯罪:犯人施暴前先服刑!?】宾夕法尼亚大学的科学家们利用机器学习技术让AI系统研究了2万8千份家暴案例,并成功在90%的案例中确认最不可能重犯的人员。研究人员称,如果当地官员采用这项技术,即根据AI预测释放预测结果为两年内不会重犯的服刑人员,那么将只有10%的嫌疑人最终可能会被逮捕,这将给美国沉重的监狱开支带来福音。这不禁让人想起美国电影《少数派报告》,电影讲述了2054年的华盛顿特区,谋杀已经消失了。未来是可以预知的,而罪犯在实施犯罪前就已受到了惩罚。

Minority Report-style machines are 'solving' crimes BEFORE they happen: AI can successfully identify the likelihood of domestic violence in 90% of cases

  • Scientists used machine learning to look at more than 28,000 court cases
  • Offenders were grouped based on likelihood to carry out domestic violence
  • It accurately identified those least likely to be re-arrested within two years
  • The method is reminiscent of 'pre-crime' in the movie Minority Report, and could potentially avoid thousands of cases of abuse

Reminiscent of the movie Minority Report, in which agents are able to convict people before they commit crimes (Tom Cruise in the film pictured), researchers are using machine learning to analyse the outcomes of thousands of court cases and identify those who are most, or least likely to commit further crimes

Accurately predicting if someone is likely to commit a crime or not is still largely in the realm of science fiction.

But researchers in the US may have taken us a step closer, using machine learning to identify those who are likely to re-offend.

They report that in the case of abuse in the home, using computers to pass judgement on an offender's likelihood to re-offend could potentially avoid thousands of cases of domestic violence.

Scientists at the University of Pennsylvania used the technique to look at 28,000 cases of domestic violence in which the offender was charged and release. By using machine learning, researchers were able to successfully identify offenders least likely to be arrested within two years in 90 per cent of cases. Stock image
Reminiscent of the movie Minority Report, in which agents are able to convict people just before they commit the crime, machine learning is emerging as a viable tool to analyse the outcomes of thousands of court cases.

In particular, they can be used to identify those who are most, or least likely to commit further crimes.

Scientists at the University of Pennsylvania used the technique to look at 28,000 cases of domestic violence in which the offender was charged and released.

They found that after the court case, offenders were likely to follow one of three paths within two years.

In the most serious category, offenders would be arrested for domestic violence where they had caused or threatened to cause physical injury.

Others would either be arrested for domestic violence without causing physical injury, or would not be arrested for domestic violence.

‘The most pressing goal was to find a subset of offenders who could be released with no conditions and who were good bets not to be rearrested for domestic violence,’ the authors explained.

They report that in the area they studied, one in five offenders released are re-arrested for domestic violence within two years.

But by using machine learning they were able to successfully identify those least likely to re-offend in 90 per cent of cases, which would mean a reduction of 10 per cent.

 

Figures from charity Living Without Abuse show that in the UK, domestic abuse will affect a quarter of women and one in six men in their lifetime. Perhaps most shocking are the statistics domestic abuse leads to two women being murdered each week and 30 men a year.

Any means to identify those most likely to re-offend could save further damage, and could even save lives.

Writing in the Journal of Empirical Legal Studies, the authors added that thousands of cases of domestic violence could be avoided each year if judges followed the predictions.

‘If magistrates used the methods we have developed and released only offenders forecasted not to be arrested for domestic violence within two years…as few as 10 percent might be arrested.

'The failure rate could be cut nearly in half.’

They said that in the urban area they studied over a two-year period, more than 2,000 post-conviction arrests for domestic violence could have been avoided.

Rather than immediately handing out jail time, those identified as likely to re-offend could be followed up with regular check-ups.

http://www.dailymail.co.uk/sciencetech/article-3483859/Minority-Report-style-machines-solving-crimes-happen-AI-successfully-identify-likelihood-domestic-violence-90-cases.html


Comments are closed.



无觅相关文章插件