The CEO's Corner: Artificial Intelligence (AI) Trainer
By Mal Shaw
I was privileged to recently represent DDLS at Microsoft Inspire (the new name for Microsoft's annual worldwide partner conference) that was held in Washington, DC, in July this year. This year's conference attendance was about 17,000 including more than 300 from Australia.
I attended the keynotes and sessions on Blockchain, Artificial Intelligence, Learning & Readiness and even learned how to air tap in a Hololens application!
One Artificial Intelligence application that I thought was wonderful is shown in this video that describes the journey of the incredibly inspiring Microsoft Software Engineer, Saqib Shaikh, and the work of Saqib and his team in harnessing AI to improve lives.
At the same session, the presenter showed an application where his company developed an automated agent to respond to enquiries in Microsoft Teams using the Microsoft Bot Framework. When asked how the Bot knew the answers, the presenter responded by saying that he had "trained" it. Of course, that grabbed my attention and I started contemplating the new career of AI Trainer. So, how does one train a machine?
In this case, the bot, supported by Microsoft Cognitive Services, already had significant natural language ability. If a question were posed, the bot would respond with some options of phrases that it already knew and knew how to respond. This, then, was the training exercise: enter all the questions that the application's users may pose and train the bot to respond to them. This certainly does look like a Learning & Development Consulting project, i.e. analyse the learning requirement, design the information gathering piece and the delivery model (to the machine!), develop the learning solution, implement to one or more machines and then evaluate the success. I guess it's likely that different machines will learn in different ways, depending on the algorithms that they use. For example, some will respond to reinforcement learning, try different answers until a positive response or negative response is received, or supervised learning where a known set of inputs and outputs is provided by a machine "teacher". Part of the role of the AI Trainer would be to identify the learning style of the machine and adapt the delivery model to achieve the best outcome. For a real world example, a quick internet search reveals the role of Fashion AI Trainer advertised, being a position where a person's domain expertise is required to improve an AI system's sense of style. These training models can achieve huge scale using machine learning models applied to massive, accurate datasets. This then leads us to also wonder if the AI Trainer can also be automated so that we have machine teaching machine. Some of our greatest minds have already contemplated these scenarios, and others, and have described the opportunities and risks in the open letter: Research Priorities for Robust and Beneficial Artificial Intelligence.
Our role, training humans, is also evolving. More high quality training resources are coming online. I'm fascinated to see some students speed up online video training content to 1.5x speed so that they can learn at the pace that suits them. Of course, we will always also teach in the classroom. In our recent client satisfaction survey, 92% of the clients that responded preferred classroom training. I'm sure that they agree with us that this is the optimal way, based on how humans learn, to learn more, faster. So, if any machines turn up in our classroom, based on the way they learn, we will know who (er, what) you are!
Thank you for choosing or considering DDLS for your learning journey.