Coding the Way to a New Working Style

In 2006, Fei Fei Li began work on her landmark ImageNet database, which would ultimately make AI-driven image recognition possible for the first time. In 2018, Joy Buolamwini released shocking research that led Microsoft and Google to question the race bias in their own AI systems. These two, while exceptional, aren’t the only women that have made significant contributions to the field of AI—despite only 12% of all AI researchers being women.

Insights spoke with Miku Hirano, CEO of AI company Cinnamon, Inc., to talk about her experience as a woman in the field and how AI is changing working styles, especially for women.

“The responsibility to prevent bias is on the developers.”

Insights: Do you think your experience as the CEO of an AI company has been different from your male counterparts?

Hirano: Many people have asked me what kind of hurdles I’ve faced as a woman, but I don’t actually know because I’ve never been a man. I can’t compare myself to male entrepreneurs in that way. Even when I face a difficulty, I have no way of knowing if it’s a problem of me being a woman, or something else. All I can do is try my best to overcome it.

As women, we need to do our best whether or not our challenges are because we are women. But the rest of society also needs to empower and support women. Otherwise the percentage of women working, the percentage of women helping make society better, won’t change.

Really, there is no difference between men and women. Of course, men tend to be able to carry heavier things, [laughs] but other than that, there’s no difference. That said, in my experience, my CEO counterparts are usually only men. So I do believe that, when making AI, involving women is very important.

If you want a cautionary tale, all you have to do is look at Amazon. Because they receive so many resumes, they used AI to pull out good candidates for recruitment. But that AI started to evaluate men more highly. It was biased.

If a woman had made that AI, she could have probably predicted it would be biased. AIs learn from training data, and the strength of the training data is dependent on humans. So if the training data is already biased, the AI will be biased. Having diverse developers is very important. If you want to make an equal system for gender and race, you need to hire with the opposite bias in mind.

Insights: So you believe the bias of AI is the developer’s responsibility?

Hirano: Oh yeah, I think so. To go back to the Amazon recruitment system, the developers didn’t consider the possibility that their AI would evaluate men higher. Developers have the responsibility to consider those possibilities, and they didn’t.

Actually, accountability is a hot topic in AI right now. Humans can explain the reasoning for their decisions, but AI is kind of a black box—it gives you output, but it can’t usually give you an explanation. This is one of the reasons companies are sometimes reluctant to use AI.

Some researchers are trying to address this. There’s a technology under development called LIME that can explain why an AI gets certain results. For example, maybe we’ve asked an AI to identify an object—my notebook. Using LIME, the AI can explain why it determined the object to be a notebook: “This shape is a rectangle,” or “There are two pages”, or “There are some small lines,” or even “This is handwriting.”

This kind of technology will help us understand and train AI better. But even without that sort of accountability tool, I believe the responsibility to prevent bias is on the developers.

“If we want a new working style for women, a new system should be developed.”

Insights: If you could create an AI solution for one women’s issue with unlimited funds and resources, what issue would you solve, and how?

Hirano: Women around the world tend to have difficulty working and maintaining their private life at the same time. Especially in Japan, women are supposed to raise their kids, so they can’t pour their full potential into their work. We need a tool to harness the little time they have without sacrificing efficiency. If we can solve this using AI, this could be a breakthrough. There are three aspects of this problem that AI can solve.

The first is the standard of working in eight-hour shifts. Historically, working hours were standardized to make work more efficient. But the key to a more flexible working style is working in chunks. AI could be used to break large, multi-step tasks into chunks more appropriate for a one- or two-hour time block without sacrificing efficiency.

The second is connecting those chunked tasks to the right people at the right time. Women at home don’t always know when they’ll be available, but if AI can connect women to tasks tailored to the amount of time they have when they are available, those women could get a lot more work done. This would also satisfy the company.

Third, to make things easier on beginners, AI could be used within each task. Take, for example, a call center—AI can be used to advise employees on the best answers to customer questions. Call center operators usually need to handle hundreds or thousands of these. For beginners, it is very difficult. It usually takes new hires between three months to two years to be professional about it, but most people quit these jobs within three months. If AI can help the operators decide how to handle customer inquiries, the work would be far easier for beginners.

In general, if you think about using AI to simplify a task, you need to think from the perspective of a new user. Not only chunking down what each task should be, but ensuring that each task can be done by a beginner, and then finally connecting individuals to tasks.

It would take a lot of time and money to totally create this system, but if we want to accelerate a new working style for women, something like this should be developed.

“People could have a safer life in the future.”

Insights: What is your kokorozashi?

Hirano: My kokorozashi is to extend human potential.

The human world has evolved so much—we went from using fire and inventing the wheel to building the internet and smartphones. Our lives are more convenient now. We can do so much more. Before trains or cars, we could walk maybe thirty kilometers a day at maximum. Now, in a plane, we can fly to another continent in hours. This is an incredible extension of human potential.

I want to do something to continue that extension. I want to contribute something big—as big as possible—to humanity. And I believe the next big thing is AI.

The democratization of specialist knowledge is key.

For example, lawyers are very expensive. Sometimes their services start at a minimum of $10,000 USD. The only people that can use them are companies or rich people. You’d never use a lawyer, for example, if you loaned $100 to your friend and they didn’t pay you back. Or if Facebook updated their policy, and you were worried about how they handled your personal information. But if AI dramatically increases the efficiency and productivity of lawyers, using a lawyer could cost as little as a dollar. We could use lawyers for super small things. It would completely change the entire market and benefit consumers.

Take the car: the car was invented 150 years ago, and at the time only crazy rich people could afford one. Decades later Henry Ford democratized the car, and now everyone can use one. Using AI, the same thing could happen to specialists.

Insights: Is this something you’re working on now?

Hirano: Yes, and the technical advantage we have at Cinnamon is that our programs can understand unstructured data.

Unstructured data is, for example, a PowerPoint or a Word document, email or photos, images, audio files. Structured data is the stuff you’ll find in a database: a client’s name or address, or how much revenue you’re going to get this month—this kind of clean data is structured data. AI and robotic process automations can handle structured data.  

But actually, 80% of all data is unstructured. So if you really want to make change or transform to a totally digital business model, unstructured data is the key—but of course it’s more difficult for AI to understand. So our technology focuses on understanding unstructured data.

If we can use technology to sort through unstructured data, it will help us democratize access to specialists. Then poor people, single mothers, and neurodiverse individuals (such as those with ADHD) could access the quality of life that only rich people have now. The sort of person that might have a really difficult life today could have a safer life in the future. That would make a big difference. That’s the future I imagine.