Last night was the first of several planned Artificial Intelligence Panels put on by @hired_hq moderated by Tom Taulli (@ttaulli). @ttaulli is a writer for Forbes and author of the book “Artificial Intelligence Basics: A Non-Technical Introduction”. The panelists were from a number of Seattle companies working on AI tech: Reza Rassool from Real Networks, Jamie Chung from KenSci, Charna Parkey Ph.D. from Textio, Jeremy De Souza from LivePerson, and Andy Martin from the host organization Zillow. The panel hit on a number of interesting AI topics including: careers, ethics, bias, and the future of the industry. Summarized below are some of the key takeaways:
Careers: For newcomers to the AI field, find roles that are tangential to your current work. Subject matter expertise can help you get a foot in the AI door. For example, if you work in healthcare, identify AI roles in healthcare as a starting point.
When vetting future employers, pay close attention to the job descriptions. Companies with mature AI ecosystems will have well thought out roles that do not require one person to be an entire team. Also, talk to potential employers or recruiters in advance to get a feel for how they are using machine learning and the specific requirements for the role. Find out if they are consumers of ‘off the shelf’ AI solutions or if they are building their own models. Determine which would be the best fit for you.
Ethics and Bias: In summary, both topics need serious and careful consideration when developing an AI project.
Bias: The main takeaway is to consider (and try to eliminate) bias before building your model and keep thinking about it on the way to production. Basically, engineer it in from the beginning because it’s hard to fix it once it’s ready to ship. Also, start with the right data and use the least amount of information necessary to create an accurate model. For instance, in facial recognition, do not label the data unnecessarily (i.e. drop the gender, race, ethnicity labels) and be sure to have enough images of all different types of people (not just proportional to your population, that might not be enough).
Ethics: Summed up by ‘First, do no harm’. If it is unclear if a machine learning algorithm will cause harm, ask yourself how it will be used. The panelists walked through examples of AI that may may encroach on civil liberties (i.e. an AI to identifying ‘political dissent’ in private communications for governments). They also talked through ethics of ML that is intended to change behavior. In these cases, it is important to consider a person’s intention and educate not manipulate.
Future of AI: Last but not least! The panelists jumped in with a lively discussion about the direction of AI. As AI moves into new markets like medicine ‘explainable AI’ will be important. For instance, it’s not enough to know with 98% accuracy how long a patient will stay in the Emergency Department. Medical staff need to know which factors are impacting length of stay and how to intervene.
If you’d like some more interesting word soup checkout AI Words you need to know by @ttaulli. Hired will likely post the panel discussion on their YouTube channel as well if you’d like to take a look.