DATAx New York keynote examines how to curb the "dark side of ML"

Sarah Hoffman, Vice President of AI and Machine Learning Research at Fidelity Investments and keynote speaker on Day One of DATAx's Machine Learning Innovation Summit, explores the current state of ML in industry – and where we need to be heading


In her keynote speech at DATAx New York Sarah Hoffman, vice president of AI and machine learning (ML) at Fidelity Investments, starts off by sharing some of the more newsworthy examples of ML in the media, including the AI Chinese anchor now delivering news from the uncanny valley and AI capable of creating art which this year went for $432,500.

"AI is even taking my kid out of a job," jokes Hoffman. "We now have AI which can find Waldo!"

However, Hoffman explains that there is also a "dark side of ML" leading the general public to become more and more afraid of it. "Americans on average now fear computers and robots taking their jobs more than they do death," says Hoffman, agreeing that it is a valid fear to an extent as many jobs are going to be deemed redundant due to AI. And equally as pertinent, bias within AI is becoming more of an issue.

"Bias is an extreme issue in ML, not because the technology is flawed but because the data behind the algorithms was created by humans who are biased," explains Hoffman. She cites examples such as the Amazon recruitment tool which had to be scrapped when it kept downgrading female applicants because the data it was trained on led it to the conclusion that female applicants were not as good as their male counterparts and the Twitter AI which was taught how to be racist in less than a day.

This is why not only outspoken fearmongers like Elon Musk – who recently claimed AI was more dangerous than nuclear bombs – but also supporters of the technology are sounding the alarm. Hoffman even quotes Stephen Hawkings' fear that unless we learn how to prepare for and avoid the potential risks AI, it could be the worst event in the history of our civilization and he had an incredibly positive outlook on the future of AI.

However, Hoffman claims she is not afraid of this because, "humans have always been scared of new technology taking their jobs; it's part of human history".

She also believes that we can prepare for these worries and prevent these risks. For instance, one key way to get around an issue such as bias is explainable AI. "It is no longer acceptable for bankers or recruiters to just say: "I can't help you because the AI said so". We need to know why AI is doing what it is doing."

While this may not eliminate bias, it will go a long way to helping minimize it and this is why Google has an "explainability" function attached to its tensor flow tool, as does IBM.

The other large fear is that AI will take lots of jobs from the economy. While this is largely inevitable, the reality is that many more jobs will be created in their stead.

"It will shock you if you go on to Glassdoor and type "ML" how many jobs there are which aren't at all connected to engineering but are instead roles around the field, such as ethical ML roles which are equally essential to the future of ML."

However, Hoffman does say that an issue we are still struggling with is reskilling individuals. It is great that we are creating all these new roles, but if there is no one to take them who does it help?

"Humans plus machine equals superpower, and that's the future of ML," states Hoffman, quoting Paul Daugherty's book, Human + Machine. "It comes down to asking yourself what parts of your job – you know, the really mundane parts – could a machine probably do better?

"There is still so much machines can't do, like innovation and creativity, and once we allow machines to take the load of all the boring stuff we do it will give us superpowers," she added.

Hoffman concludes by sharing some of the applications of AI she is most excited about. In the healthcare space, diagnoses and treatment opportunities through ML are really taking off. "The idea of living longer and healthier lives due to ML really excites me".

Likewise, as an employee of Fidelity Investments, she also elaborates on tools like robo-advisors and chatbots, which are still very basic right now but show so much potential. "They can help us make better financial choices because they can help everyone, 24 hours a day. Future chatbots may be able to give you very specific advice regarding your financial wellbeing."

Finally, she outlines the greater potential for the democratization of AI as time goes on. While there is so much to be gained from ML, it is still being done by "a very small pool of people and that is not enough", Hoffman argues.

"Yeah, we can make more degrees, but we also need automated ML platforms, so others can take advantage of the technology in new and interesting ways."

While it is still early, AI is already being incorporated into our day-to-day lives. "In China, instead of only playing with dolls, kids are playing with, and learning from, AI robots. Even our kids, with stuff like Alexa, are already miles ahead of where we were at their age."

Sarah Hoffman was the keynote speaker at Day One of the Machine Learning Innovation Summit track at this week's DATAx New York festival. She was followed by Zachary Cohn of Meetup, who spoke about using ML to foster better connections, and Hangjun Xu's presentation on how Airbnb has been using customized regression models for the firm's dynamic pricing. To learn more about how ML is transforming all works of life and industry, click here.

How big data can protect infrastructure small

Read next:

How big data can protect infrastructure