“Finding a refrigerator will do when you are hungry, if you know to look inside it, and not to eat it. None of this is obvious to a machine.” – Roger C. Schank and Peter G. Childers, The Cognitive Computer: On Language, Learning, and Artificial Intelligence (1984)
Last time, I wrote about some potential problems surrounding artificial intelligence (AI) in a financial institution setting. A lot of the problems AI researchers encounter have to do with training the computer and setting up the tasks, as Schank suggests in the quote above. If you set the parameters incorrectly, or provide a flawed set of data, you can get unintended results. Recent trends in machine learning, such as reinforcement learning using deep neural networks, intensify the upside of augmenting human efforts with a computer—you can solve problems better and faster than a human—but also potentially intensify downsides if you mess up the training.
AI and its cousin, automation, are like any other business approaches: when you get things slightly wrong you waste a lot of time and resources. Therefore, any change in direction needs to be evaluated in terms of strategic goals and return on investment (ROI). Any time we set a computer to do a task, we want to make sure it is set up to achieve the goal at least as well as a human.
AI Is My Co-Pilot
Consider an example from my field of Ph.D. research, computer vision, which has been crucial to developing autonomous vehicles. People are quite bad at driving cars and have fatal crashes all the time. So it should not be too hard to meet the benchmark of being at least as good as a human (well, when it comes to laws and ethics, that is a different story). In practice, though, operating a car is really complicated and requires enormous investment. While autonomous vehicles’ vision systems are now quite functional, they can still be easily hacked. For instance, the cars can be made to mistake a stop sign for a speed limit sign, or (theoretically) can be trapped in a circle.
Mesh computer abilities with human ones, though, and you have a winning formula, for the time being. At the low end, lane departure warning systems alert the driver when the car’s sensors detect that he or she has entered a dangerous situation. The driver will then hopefully return to their proper lane. At a higher level, the car can take over control of steering with lane-keep assist or road departure mitigation. The human and computer skills work together to augment or amplify the available intelligence around driving.
Moving the Needle Toward AI
I’ve been talking about cars because they are a little bit easier for most people to conceptualize than marketing campaigns, operational workflows, and other situations where Katabat is applying powerful computing to eliminate human error and enhance efficiency. The overall principle, that there is a gradual progression from humans completing tasks, through using computers, to automating tasks, to using AI to assist in or take control of tasks, is the same regardless of the field of technology. I am constantly monitoring that gradient in order to determine when the business case is there for our clients to adopt a cutting-edge or disruptive new level of the technology.
We’re developing exciting ways to enhance the customer experience and inform marketing strategy using AI. For now, we automate outgoing omnichannel messaging on schedules and when triggered by events, then manage incoming responses from customers’ preferred channels using sophisticated decisioning to connect them with appropriate resources or an available agent. Our strategy for upcoming releases of the Katabat platform will see implementation of sophisticated AI optimization and predictive analytics throughout marketing, communications, and customer experience.
I hope you will continue to follow this blog space as I discuss the impact of automation and AI on financial services in general and Katabat clients in particular.
Please contact me at any time at email@example.com to discuss ideas! And remember, don’t eat the refrigerator.