The User & AI
“The best AI is the one that is not seen” is the mantra followed by all of us in the AI world. We want to make AI seamlessly work in any user interaction or for that matter user not involved action. For example, we do not realize that when we are talking to devices like Alexa or Google home, there is inherent AI involved for the device to respond. However, an important part of AI is the learning it needs and as it stands today, the learning mainly comes from the human interactions that the machine learns from. For example, learning the nuances of your speech when talking to the home assistants, determining if a photo is of a cat, playing chess, figuring out diseases by looking at the CT Scans etc. The basis for AI is to learn from humans and whenever there is a new scenario, unless AI evolves, it will be a human tagging and interpreting what the new scenario is and teaching the machine how to deal with it.
This becomes even more stark in the enterprise environment where we are using AI to help with either decision making or automating tasks that users have been doing leveraging their experience ( sometimes also called ‘tribal knowledge’) but not leveraging the overall shift towards digital and the resulting abundance of data. For example, a Factory Manager taking decisions on specific downtimes in the factory, an E-commerce Marketing Manager deciding on timing of promotions on their website, a HR person deciding on shortlisting a candidate based on the resume etc. In all these examples, the user decision is the one that the AI engine is learning from and then driving the outputs or the decisions. So, the user always stays in the picture.
Also important is the adoption of AI within the enterprise hinges on the acceptance of AI. The initial doubts on the AI engines ability to surpass human decision making is questioned not just because we don’t believe it can do more with data but because we don’t believe that it has all the variables that should be considered while making the decisions. Hence there is always a human element involved for “validating” whether the AI engine is delivering. The objectivity of this is up for debate (and another blog) but the fact is that the adoption of AI within the enterprise is still a very User Centric decision.
That brings us to the interaction of the User and the AI engine which is a very important ingredient as to whether the AI engine adoption will succeed or not. The key tenets to this succeeding are
- User needs to “feel” that they are still in control
- The users should be able to override the AI decisions
- The AI engine should act as an aid rather than the overall driver of the decision making/automation
- The interactions should allow the AI engine to learn from the user decisions
- The toughest part – the user should be able to understand why the AI engine proposed something that was proposed (Explaining the AI)
The above tenets make it very important for a user centric design that empathizes with the user of the AI engine to make sure the adoption is successful. This requires bringing in the left side of the brain (creative user design) with the right side of the brain (the AI algorithm design) to deliver an application that is easily adopted by the users.
At Tailwyndz, that is our approach to making sure that the application we design does not just deliver on the AI side but is well adopted within the enterprise. Do you also follow a similar approach ?