Utilization of Chatbots as Personal Assistants

Personal assistant chatbots are used to handle administrative tasks such as scheduling meetings, setting reminders, retrieving information to answer questions, or pulling information from the internet into a summary. These kinds of chatbots have a variety of different use cases and can be used internally within a company to manage employees, client-facing to schedule meetings with clients in a business-to-business setting, or consumer-facing as an individual personal assistant. The areas where personal assistant chatbots can add value to a business are similar to customer service and eCommerce chatbots. While low involvement AI can still be implemented in these firms, the full potential of AI can only be harnessed by firms with the resources to train more robust systems. Granted this could be counted off as a benefit for larger companies for having grown their businesses effectively, but this uneven distribution of technology may create stark competitive barriers for smaller companies.

Beyond its impact on preventing a fair competitive landscape, AI's over-reliance on data represents a fundamental inefficiency in the technology. Many discussions regarding advancements in AI that have been made public over the past few years may lead some people to doubt whether the technology is even remotely "intelligent" in comparison to humans. For example, you wouldn't have to show a child 10, 000 images of a dog for them to learn it's a dog. Humans have adopted heuristics that allow us to take mental shortcuts and process information much faster. For AI to deliver on its promise of replicating human intelligence, it will need to adopt similar shortcuts for learning. Improvements to the technology need to be made that allow AI to learn with less data before it is a practical solution for marketers. Developments have been underway for years to improve this flaw of AI.

The Bayesian program learning framework (BPL), developed by Brenden M. Lake, Ruslan Salakhutdinov, and Joshua B. Tenenbaum (2015), allow for an AI system to replicate human-like behavior after being exposed to just one dataset. While not yet perfected, the BPL framework is an exponential improvement over the massive data requirements of deep learning algorithms. Gary Marcus is also working to improve the efficiency deep learning algorithms with his company Geometric Intelligence. His XProp software was able to recognize numbers at an error rate of 0. 2% after being exposed to just 150 examples, as opposed to the 700 examples needed for a deep learning algorithm to accomplish the same task.

Despite its performance in recognizing handwritten numbers with an error rate of 0. 2% not even being an improvement on current deep learning algorithms, Marcus' software puts AI technology a step closer to being less reliant on massive amounts of data to be effective. More recently, an increasing number of AI developers have recognized the inefficiency of how much data is needed to power AI and machine learning. As Charles Bergan, vice president of engineering at Qualcomm, said at an MIT Technology Review conference in January of 2018, teaching algorithms using "one-shot learning" would represent a massive shift in the direction of AI. Traditional AI developments have taken a white-knuckling approach by trying to make stronger computers. One-shot learning would shift the focus to creating more efficient algorithms.

On this same note, work is being done to reduce the size of neural networks without reducing accuracy. Efforts to improve AI algorithms tackle critical issues that the technology faces. For AI to be a useful tool for marketers, it will need to interact seamlessly with the rest of an organization to provide value. Firms will be fighting an uphill battle to actualize a positive ROI on an AI system that costs an astronomical amount to implement. For AI to be a practical solution to automating menial tasks, it will need to become more efficient in how it learns. These forms of "lean" AI systems should serve as a benchmark for progression of AI technology. AI systems that can learn using minimal amounts of data, then adjust on the fly using minimal amounts of potentially low quality, unstructured data would represent an advancement in the technology that could greatly improve its efficacy for marketers in firms of all sizes.