Things to consider
Firstly, if a project is to have any value, then it should reimagine, not replicate, an existing process. You have to be creative and experimental in the approach you take. If you are merely doing things the same way but with different tech, then it’s unlikely you have a good AI project.
Secondly, the implementation of the project will include several training phases for the AI. The most obvious will occur during the initial loading of data and/or knowledge base, which most people recognize. However, what’s different is that the system needs to go through a phase of refinement or learning once the software has been released. During this phase, things like gain and variance for Machine Learning, or intent training for NLP and maybe model refinement for cognitive reasoning need to be improved. So, during this phase, it is essential to carefully release the software and measure how well it’s performing over a 6-12 week period, at the least.
The most significant thing to remember about AI implementations is that there are over 20 different types of AI technologies and techniques, and not one fits all situations.
It is critical to understand when Machine Learning is strong and when it is weak. The same is true for Deep Learning or Cognitive Reasoning.
With this sentiment in mind, it is important to not get caught up on “chatbots” or “Alexa” as the AI: these are just UIs. Think about what Forester called the “Cognitive Moment”. That’s the moment when the AI does something that makes the user’s life easier. As an example, last year’s iPhone upgrade. Out of nowhere, it seemed, the iPhone started to look at our meetings in iCalendar and told us when we needed to leave for the meeting based on traffic volume and potential delays. That’s AI. It’s also a cognitive moment and not a chatbot.
10 AI commandments
To frame the best AI project, it’s worth considering these 10 AI commandments, as accumulated by Rainbird’s experts:
1. Don’t pretend AI is human: AI is mathematics, not magic, so stop referring to it as a person and realize that it is, in fact, a computer.
2. Follow the full customer journey: it’s better to answer a handful of questions well than answer a ton [and badly].
3. Don’t get hung up on the idea of a chatbot: focus on the cognitive experience and explore other UIs which may be better for your situation, such as email, voice or CRM.
4. Don’t be blinkered: there are other types of AI out there beyond Machine Learning, Deep Learning, and Neural Nets… cognitive reasoning, for example. Be sure to check out all the possibilities before committing to one.
5. Don’t use AI for the sake of it: if you can do something better by using traditional technology then do it that way.
6. Understand the business case: be sure to define ‘success’ or your proof of concept will be nothing more than an interesting toy.
7. Automate: focus on one task at a time and don’t try to automate a whole person.
8. Enhance humans: look at how you can amplify their skills – don’t try to replace them.
9. Don’t suspend reality: if you are in a regulated industry, for example, and the rationale of the system needs to be audited, then Machine Learning won’t work, no matter how clever the algorithm.
10. Get started, experiment and learn from your mistakes: AI is more accessible than ever and remember that no one has all the answers.