On artificial intelligence, museums, and spaghetti
When I was a kid, spaghetti for dinner was a big treat. Like Kool-Aid-level-big treat…we just didn’t have it that often.
There was a special bonus of a spaghetti dinner: the meal prep. My mother would pull a noodle or two out of the boiling water and throw it at the kitchen wall. If it stuck, or slowly lumbered downwards, dinner was ready. If it bounced off, then the spaghetti simmered a few more minutes. In either case, “helping” my mother by throwing noodles was hot fun for 6-year-old me. (Throwing spaghetti at walls is still a part of my mealtime QA testing.)
Generating use cases for machine learning (a branch of AI) and museums is kind of like spaghetti throwing. We’re in the stage of “seeing what sticks” with the promises of AI, the realities of AI, and relevance to shared museum goals. It’ll take experimentation and time to understand the full impact of machine learning on museum operations.
Thankfully, there’s a market that can speed up the process of experimentation and learning. Technology that was once the domain of AI engineers is increasingly accessible in the form of machine learning as a service (MLaaS). Google, Microsoft, IBM, and Amazon all offer these (frankly, amazing) services to developers. There are tons of options across products and features, from building your own chatbot to image recognition to generating complex, multivariate predictions.
So, in the spirit of what might “stick” in museum-oriented use cases — and what AI and machine learning as a service offer — there are some early alignments emerging.
Some notes on the above:
- The “Effort/Cost” axis can be loosely interpreted as a “Build vs. Buy” continuum. The far left represents moderate in-house or partner effort, and the far right contains much of what you’d buy from a third party source. That said, you could take anything positioned on the left and build up its complexity, and drive up cost or effort. (Definitely the case with chatbots.)
- Content optimization refers to the use of machine learning to analyze content for entities, categories, standardized concepts (for example, aligned to DBpedia), structures, etc. for the purpose of improving online content for voice queries, SEO, and future machine learning use cases.
- Voice skill refers to voice assistants like Amazon’s Echo and the “skills” that can be used.
- CRM/IA optimization refers to third party platform “AI” features, such as Salesforce’s Einstein or Blackbaud’s SKY, and similar solutions.
- Facilities management refers to facility monitoring that uses AI-assisted tech to optimize energy usage, occupancy, security, and related functions.
- “Robots” refers to public-facing robots like Pepper the humanoid robot at the Smithsonian. (Not MLaaS.)
Necessary caveat: This is purely subjective and very early; YMMV. I’m sure my thinking will evolve as these services mature, I gain additional experience with these services, and the barriers to entry shift around. (And you may disagree with my assignment of relevancy. Tell me in the comments!) Of course, what’s offered through MLaaS doesn’t represent the full set of possibilities in AI and machine learning…but it does provide an accessible set of tools for museums to consider.
(Sidebar: After experimenting with different data sets using IBM’s Watson services, I’m seeing some promising use cases for web site redesigns and content discovery…depending on how the experiments go, I’ll report back.)
Acknowledgment: Neil Hawkins and his case study with the Baseball Hall of Fame Museum for some recent inspiration on what’s possible. (I’m also a fan of IBM’s Watson offerings in MLaaS — thumbs up to their free trials in natural language processing, content discovery, sentiment analysis, image recognition/classification, and more.)
*The term “AI” is a little overplayed. …OK, a lot overplayed. “Machine learning” strikes me as a more accurate term to describe much of what I talk about here. Please forgive the shorthand.