Are companies able to use artificial intelligence to increase efficiency and are their employees adequately qualified to leverage these technologies? Prithwiraj Choudhury applies to the US Patent and Trademark Office for a case study.
Considering the rapidly developing era of artificial intelligence, employers are excited about the productivity gains that such tools can bring, while they rather calculate the time remaining before R2 – D2 take their job.
“Jacques Bughin and his co-researchers estimate that in the future, 50% of all tasks currently performed by humans can be done through machine learning and artificial intelligence,” says Prithwiraj (Raj) Choudhury, assistant professor at Harvard Business School. Overall, this could increase global productivity by 1% or more.
It turns out, however, that for a long time before robots are a massive substitute for workers, they use AI-based tools for their work, as we already see in radiologists using these X-ray and X-ray interpretation tools to study machine learning to find cases from the past that set a precedent for legal arguments.
“If somebody in the past has been exclusively in the old technology world and suddenly uses a machine learning tool, they are less productive.”
Choudhury realized that there was little research on the skills needed by the workers to fully utilize the artificial intelligence-based tools. And it’s an important piece of information when companies plan to invest what Accenture estimates in the US by 2035 for $ 35 trillion in cognitive technologies. The ease of adding AI tools automatically improves productivity when users can not. the technology right.
“Artificial Intelligence tools can be good for predictions, but if they’re not used properly, it does not make sense to invest in such tools,” says Choudhury.
Choudhury wants to close this gap with a new discussion paper. Different settings for different people: experimental data on complementarities between human capital and machine learning. The newspaper, written with Evan Starr and Rajshree Agarwal of the University of Maryland, suggests that companies think carefully about the skills they need for hiring or training so that their employees can get the most value for their money. her new AI.
Choudhury has devoted his career to human capital research, including companies such as Microsoft, Infosys and McKinsey, to analyze what makes knowledge workers most productive. Several years ago, he became interested in the US Patent and Trademark Office (USPTO), which uses innovative practices for remote workers.
“I found the US Patent Office intriguing,” says Choudhury. “It’s not just a large organization with more than 10,000 employees, but an organization that shapes the innovation system, what are they doing for the entire US economy?
In the course of writing the fall of the Harvard Business School case in the Patent Office, he noted that the agency implemented a new state-of-the-art machine learning program called Sigma-AI with the aim of reducing the time needed to review patent applications.
Patent examiners can use Sigma-AI to ensure that the applications truly offer new ideas, not designs or techniques previously used in other patents known in the art. “It means going through hundreds of thousands of documents,” says Choudhury.
The Office would like to provide the Applicants with a first answer within 10 months. With patent applications increasing nearly 20% in five years, half a million late applications are currently pending, resulting in delays of six months or more.
In the past, employees used a Boolean search tool similar to Google’s to find the state of the art and search for specific keywords to find earlier cases. The new machine learning tool automates this process, says Choudhury. “The document will be introduced to this tool and will then reveal the documents relevant to an examiner.”
Is a computer training necessary?
Choudhury and his research colleagues wanted to know if education in computer science and engineering (CS & E) could improve the ability of patent attorneys to use the tool of artificial intelligence to make them more productive.
To make sure that previous office experience does not distort the findings, the researchers “recruited” patent examiners who would be a white board: HBS MBA students. For the experiment, they gave each of the 221 students a patent application with five relatively unclear claims for which the prior art existed. Half of the students were randomly assigned to the Boolean search tool and half to the machine learning tool.
In addition, they gave half of each group access to expert advice to help them carry out their research. This advice was surprisingly important to researchers to get the right answer.
“Without the advice, nobody gets the quick fix – it does not matter if you use Boolean or automatic learning,” says Choudhury. “It’s a testament to the human expertise of a real patent examiner with years of experience.” Chalk one for human.
Researchers found that workers’ productivity increased or decreased depending on their background. Those with experience in CS & E achieved better results with the Machine Learning Tool, the ones that do not fare better with the Boolean tool.
For this experiment, the researchers have not tried to find out which tool is the best. However, this is not relevant, says Choudhury. In fact, many companies have already used artificial intelligence technologies to improve their productivity. However, according to Choudhury, “it is used in the vast majority of situations by people without computer experience.”
It’s like asking someone with a background in humanities to use macros in Excel – they may one day understand it, but they will not be as productive as those with statistical exercises. If companies do not balance their employees’ lack of IT experience, they may not be able to leverage the technology they use to improve their operations.
“When a person’s experience is solely in the old technology and they are given a machine learning tool, it’s less productive, even if it’s a great tool,” says Choudhury.
This does not mean that companies necessarily have to hire computer scientists. It may be that employees who do not have such knowledge can learn through extensive training on how to use machine learning tools effectively. Choudhury is currently preparing for a more ambitious 1,000-topic experiment. In this case, people without CS & E receive hands-on training to see if they can improve their skills.
“We will see if these people will catch up in the second phase and reduce the productivity gap,” says Choudhury.