By Jane Mcentegart
When you need advice, who’s your first call? A trusted mentor? Maybe a specific set of people whose opinions you respect? Of course, more advice doesn’t always make things clearer (actually, it rarely does). What if a machine could collect advice from hundreds (or thousands) of people and give you the Cliffs Notes – short, actionable advice based on the collective expertise of those who have walked the path before you?
Intel recently collaborated with the Society of Women Engineers (SWE) and the Athena Alliance on the #AIforWomen initiative. As part of an International Women’s Month celebration, AIforWomen polled career professionals (inside and outside Intel) asking what advice they’d offer to those trying to grow their own careers. People from all around the world responded and shared advice on social media.
For a person to trawl through those posts for the recurring themes and the most prolific pieces of advice would be no small feat. For an AI, it’s no sweat. A collection of algorithms can take all of that text (15,700 words in total), sift through it in short order, and turn it into five key pieces of advice in minutes.
Opportunities, not comfort
Look for opportunities outside your comfort zone and choose what makes you feel most uncomfortable. That’s where you can learn and grow most, where you get the courage to experiment and discover new worlds.
Learn at work
Throughout your career, you can learn at work. Believe in yourself and do what you enjoy. If you like what you're doing, you perform better, and success follows.
All together for gender equality
The opportunities for women today are endless thanks to the trailblazing efforts by those who came before us. For our daughters and future generations, we all need to support and empower women, and give each other strength every day.
Find a mentor
Reach out to your connections for help and find a mentor who inspires you. Cultivate a network of colleagues in every position. Most importantly, believe in yourself and your abilities.
Be true to yourself
Set your personal mission and be unapologetically authentic! Bring your whole self to whatever you do – work, school, home – and make yourself a priority. Make your unique voice heard.
How to get advice from a machine
Last year, Intel engineer Shira Guskin led a project that leveraged AI to find the most common career tips for women in Israel (where gender equality lags behind most developed countries). As part of this year’s AIforWomen initiative, Rita Brugarolas, an AI software solutions engineer with Intel’s AI group, took the project to a global level and expanded on Shira’s work to try get more out of the data.
Rather than just analyzing the data for common themes, Rita wanted to see if it was possible to extract fresh insights from the aggregated text – a truly crowdsourced piece of advice. She did this by leveraging multiple AI models that tackled the outpouring of responses in three different ways.
When I talk to Rita on the phone, I’m pretty direct: “How much AI is happening here? Like, is it just that this can categorize a spreadsheet of entries faster than I can, or is there more to it?”
There’s definitely more to it.
“There's quite a bit of nuance that the model is capable of understanding,” Rita says. “In the past it was a lot of word counting and stuff like that, but these results are quite insightful.”
At first, Rita tried giving the machine predetermined categories for sorting (including an “other” bucket for advice that didn’t really fit anywhere), but she thought it would be more insightful to eliminate the bias of pre-selected categories and let the model discover common underlying themes on its own.
“This is the most impressive thing about this work,” she explains. “That without any labels or human input, it’s able to find some very sensible themes.”
A transformer model, typically used in natural language processing, is a powerful model that learns context in sequential data (like the individual words in sentences that make up a piece of advice) and can find trends or relationships in the data. In this case, it was used to generate a vector to represent each tip in the campaign, while another algorithm was used to cluster the tips into groups. Then, another algorithm extracted the four or five most descriptive terms for each topic cluster. These are the highlighted themes above.
The resulting themes are pretty easy to interpret, but Rita took the analysis one step further to try and draw out real advice. She fed the groups of tips belonging to each theme to a summarization model as one monster chunk of text, and the computer boiled the text down into once succinct piece of advice.
My once-over is selling it short — ask enough questions and Rita will break down how the models deal with punctuation, combine similar topics that are close together but not a total match, based on plotted inter-topic distance, and more talk of clustering and how linguistic principles (like contractions, grammar and lemmatization) can affect the entire process. The results themselves are (thankfully) a lot easier to understand.
You can explore advice submitted as part of the AIforWomen campaign by searching the #AIforWomen hashtag on Twitter, LinkedIn and Instagram.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.