Hear from the CIO, CTO and other C-level and senior executives on data and AI strategies at the Future of Work Summit on January 12, 2022. Learn more
In the week that 2021 approached, the tech news cycle in general died. Even industries as fast as AI sometimes need relief – especially in the new Covid-19 variant, which supports plans and major conferences.
But that doesn’t mean the end of December was eventful.
One of the most talked about stories came from the South China Morning Post (SCMP), which described an “AI prosecutor” developed by Chinese investigators who could identify crimes and suppress charges “with 97% accuracy.” The system – which trained over 1,000 “symptoms” derived from 17,000 real-life crimes from 2015 to 2020, such as gambling, reckless driving, theft and fraud – recommends sentences with brief text descriptions. According to SCMP, it has already been piloted in Shanghai Pudong People’s Procuratorate, China’s largest district prosecution office.
It is not surprising that a country like China – which, like parts of the US, has embraced speculative crime techniques – is pursuing a black-box stand-in for human judges. However, given how historically unequal algorithms have been shown in the justice system, the implications for those who may be subject to the AI plaintiff’s judgment are nonetheless worrying.
A study by researchers from Harvard and the University of Massachusetts, published last December, found that the Public Safety Assessment (PSA), a risk-measuring tool that judges can choose to use, determines whether a defendant should be released before a trial. Recommend that is too serious. Moreover, according to researchers, the PSA is likely to impose cash bonds on female detainees versus male detainees – a potential sign of gender bias.
The U.S. justice system has a history of adopting AI tools that have since been shown to be biased against defendants belonging to certain demographic groups. Perhaps the most infamous of these is Northpoint’s Correctional Offender Management Profiling for Alternative Sections (COMPAS), designed to predict a person’s potential to become a rehabilitator. A ProPublica report found that COMPAS was more likely to incorrectly assess black defendants as having a higher risk of relapse than white defendants, while at the same time flagging white defendants as a lower risk than black defendants.
New research has shown that training of projected policing tools aimed at reducing bias also has little effect, making it clear – if not before – that it is impossible today to deploy these systems responsibly. That’s probably why some early adopters of speculative policing tools, such as the Pittsburgh and Los Angeles police departments, have announced that they will no longer use them.
But with less law enforcement, courtrooms and municipalities plowing forward, regulation driven by public pressure is probably the best condition for technology to govern and set standards. Cities, including New Orleans, including Santa Cruz and Auckland, have a complete ban on speculative policing equipment. And the non-profit group Fair Trials is calling on the European Union to include in its proposed AI regulatory framework a ban on alleged criminal instruments.
“We do not condone use [of tools like the PSA]Ben Winters, author of a report by the Electronic Privacy Information Center that called pretrial risk assessment tools a strike against individual liberties, stated in a recent statement: ” Should. “
A new approach to AI
It is unclear whether even the most sophisticated AI systems understand the world as humans do. That’s another argument in favor of regulating presumptive policing, but one company, Sycorp – which was profiled by Business Insider this week – wants to codify common human knowledge so that AI can use it.
Sycorp’s prototype software, which has been in development for nearly 30 years, is not programmed in the traditional sense. Psychop can make predictions that the author can expect from the human reader. Or he may pretend to be a confused sixth-grader, assigning users the task of helping him learn sixth-grade math.
Is there a way for AI with human-level intelligence? It’s a million dollar question. Experts such as Yoshua Benjio, Facebook’s vice president and chief AI scientist, Yan Lacun, and well-known professor of computer science and artificial neural networks expert, do not believe it is within reach, but urge others to distance themselves from it. One promising direction is neuro-symbolic reasoning, which merges learning and reasoning to make algorithms “smart”. The idea is that neuro-symbolic logic can help incorporate common sense logic and domain knowledge into algorithms, for example, to identify objects in a picture.
New specimens, such as “artificial brains” made up of living cells, may be on the horizon. Earlier this month, researchers at Cortical Labs created a network of neurons in a dish that learned to play pong faster than an AI system. Neurons weren’t as skilled at pong as the system, but they only took five minutes to master the mechanics compared to 90 minutes of AI.
Pong rarely reflects the complexities of the real world. But in conjunction with forward-looking hardware such as neuromorphic chips and photonics, as well as novel scaling techniques and architectures, the future looks bright for a more capable, potential human-like AI. Regulation will catch up, with any luck. We’ve seen a preview of the results – including false arrests, sex work recruitment and erroneous grades – if that doesn’t happen.
For AI coverage, send news tips to Kyle Wiggers – and be sure to subscribe to the AI weekly newsletter and bookmark our AI channel, The Machine.
Thanks for reading,
AI Staff Writer
VentureBeat’s mission is to become a digital town square for technical decision makers to gain knowledge about transformative technology and practices. Our site delivers essential information on data technologies and strategies so you can lead your organizations. We invite you to access, to become a member of our community:
- Up-to-date information on topics of interest to you
- Our newsletters
- Gated idea-leader content and discounted access to our precious events, such as Transform 2021: Learn more
- Networking features and more
Become a member