This blog post was originally published at Intel’s website. It is reprinted here with the permission of Intel.
Are robots coming for your job? Well, maybe. In a recent Intel on AI episode, Intel’s Sandra Rivera and Abigail Hing Wen sat down to talk about how artificial intelligence (AI) will impact the future of work. It’s a deeply interesting conversation as Sandra is the Chief People Officer at Intel, working to hire and retain some of the most skilled workers on the planet. The conversation is all the more pertinent as the world is still dealing with the global COVID-19 pandemic.
As a semiconductor manufacturing company, Intel must keep workers onsite in our fabrication facilities even during a health crisis. But there’s no real touching of the wafers by human hands. These facilities are some of the most high-tech robotics environments in the world, and it’s one of the reasons why Intel has been able to maintain a 90% on-time delivery rate throughout the pandemic.
During this time of social distancing and proper safety protocols, Sandra notes that Intel is able to run fabrication facilities with less than 50% of the onsite workers onsite under normal operations. Intel worked quickly to set up remote working capabilities for thousands of workers, including remote operations centers for these fabrication facilities. Even with necessary IT, security, and operational staff such as cafeteria and cleaning services, Intel was able to change working strategies so that less than 10% of those teams needed to be onsite each day.
“You still today cannot replace human judgment—the reasoning, the creativity, and the communication skills that experienced professionals have.”
Using AI for Repetitive Tasks
In a previously blog about Pieter Abbeel, we looked at how smart robots might be implemented in warehouses as part of a larger shift towards more automation and why robots that can learn will be crucial to move past the repetitive task phase. Similarly, Sandra sees any level of automation for repetitive tasks as an overall positive as it can free up capacity for uniquely human work that requires nuanced judgement, creative problem-solving, and higher cognitive functions.
Does this mean that robots and AI are going to take over certain jobs? Sandra thinks so. Perhaps not entire jobs, but certainly a lot of day-to-day, repetitive tasks for data collection and processing that can be done more efficiently and more error-free by machines. Even at a company as technologically advanced as Intel, Sandra knows employees spend too much time collecting, sorting, and organizing data. What she’d like those employees to be doing instead is gaining information and insights from that data to make better quality decisions. For example, a team within Intel developed an AI system that mines millions of public business web pages and extracts an actionable customer segmentation for the sales team.
Using AI for Talent Acquisition
Another area Intel is using AI is in talent acquisition, development, and retention. Between full-time jobs, contract workers, and other roles, Intel hires over 20,000 people every year. Appropriately matching the company’s needs with the available skills in the market, or even in Intel’s own database, could mean analyzing several million contacts. Trying to do this manually, of course, is not efficient. But with Intel’s own internal model, the hiring team has gone from several weeks to only a matter of hours to find a pool of qualified candidates.
Another important area where AI might help employers is training and retention. In the podcast Sandra says that when looking at core engineering skills, whether electrical engineering or software or a general computer science background, many of those skills are transferable to other roles. Assuming that engineers have almost 80% of the skills they would need to do something else withing a company, it makes business productivity sense to invest in current employees to help them close that 20% gap in knowledge rather than go out and hire a brand new slate of employees, who will have to be onboarded into the organization.
One of the things Intel is looking to do is use more AI tools to better understand a skills database of existing employees, and then track employees’ desires for mobility within the company. This will give Intel greater retention, allowing career progression or even lateral moves to different departments that employees have always been interested in pursuing. And it’s not just Intel. Two years ago, AT&T set a goal to retrain 100,000 employees in skills like data analytics, cybersecurity, and cloud computing, giving them access to online trainings with Coursera and a portal that tracked what types of current and projected jobs the company needed. Giving more training to existing employees is also a route that many organizations are using to create teams that can apply AI to core business problems. This incremental, organic “grow your own” approach helps circumvent one of the main pitfalls for organizations needing to add new technical disciplines: how do you hire well for skills you don’t already have?
Using AI to Overcome Bias
Of course, anytime you replace human decision making with machines there’s an issue of exacerbating underlying errors and other issues, especially around bias. In the same podcast episode as Sandra, Abigail spoke with Ben Taylor, Chief AI Evangelist at DataRobot, about how to fix AI models to avoid biases in the field of human resources.
In the podcast, Ben notes that AI technology can act as an amplifier for human mistakes. The team at DataRobot seeks to mitigate these this issue by tracking and measuring the outcomes from AI models to better understand if a system is amplifying the bias, and then dropping certain features and components to remove the problem. The fact that people are beginning to address issues of bias is extremely heartening to Ben. He believes that protecting against bias, considering the ethical implications of technology, and being open to data regulation will make AI more explainable, which makes it much more useful, and therefore more powerful. Tools that improve “model explainability” can help us to understand decisions made by very complex models—important for detecting bias, for building confidence in decisions by AI, and for debugging purposes.
While very early AI models were effectively self-documenting—simple decision trees that were really just mechanized flowcharts—this is not true of most models trained today. As models have become more complex, “explainability” methods have had to evolve. Interestingly, some “black box” techniques are able to give insight into decisions without access to the internals of the AI model, which means we can also apply them to decisions made by humans. While human decisions seem very accessible—can’t we can just ask the decider what they were thinking? —in practice there is a large literature within psychology showing that post-facto explanations are often just rationalizations, even when our interlocutor is really trying to be transparent. Much of our own thinking is inaccessible to us. (For a pithy and entertaining summary, read The Mind is Flat by Nick Chater.) There is every chance that better tools and standards for AI models will lead us to more transparent and fairer human decision making.
Using AI to Give Humans More Time
Ben also talks about the impact of AI on employment. One example he gives is that DataRobot created a proof of concept for a media company, showing how AI could out-perform humans in sourcing. Instead of firing the people who used to do that work, the media company instead expanded their team as it opened up more resources for humans to do the creative work customers really needed.
Ben also notes that advancements in storage, especially Intel® Optane™ memory, are allowing researchers to handle data sets previously too large to analyze—something that was holding back enterprises from seeing the true potential of what AI can do. This is consistent with the experience that we had when Intel partnered with the National Center for Missing and Exploited Children, to help analyze and route the growing volume of tips they receive. We were able to offload lots of routine work to AI, focusing the human expertise where it is most needed.
To learn more about Intel’s work in AI, visit: https://intel.com/ai
To hear more Intel on AI episodes with host Abigail Hing Wen, find your favorite streaming platform to listen to them at: intel.com/aipodcast
Data Scientist, Intel