There are more companies than ever before claiming to work on some type of AI. But all AI is not the same. As of this writing, there are two different problems AI is trying to tackle. Narrow and General Intelligence.
Narrow Intelligence: In over-simplified terms, is the ability to perform one specific subject matter with defined tasks from a point A to a point B. Ex: Autonomous Vehicles, Natural Language Processing, Image Recognition, GO player.
General Intelligence: Is the ability to perform discretionary tasks as similar to a human as possible. Ex: The failing DARPA robots in the animation.
Although the DARPA robot's failure is weirdly satisfying to watch, general intelligence has made little to no progress over the last several years, requiring exponentially more data than narrow intelligence, to help with adequate training. Over 90% of advancements, subsequent investment and adoption of AI has focused solely around narrow intelligence.
What are chatbots?
Hand-in-hand with the ubiquitous adoption of smart devices, integrated search engines, social media, chat and messaging platforms: rose the volume of data collected to train narrow intelligence based AI systems to map and understand some basic human conversation structures. For example, the majority of questions asked in English, end with a question mark and are requesting a form of answer. Most questions, have a clearly defined context, subject and intent.
Chatbots, when properly thought through and designed well (see Farooq Khalid's Blog Post: Conversation vs. Design), provide an ability to quickly answer pre-organized and programmed questions, collect and present data from a pre-cleaned and correlated data source and perform basic tasks via API. They're generally a little rigid, and prone to break easily, but that's usually more of a design than capability issue.
How are they different than AI Assistants?
AI Assistants build on top of chatbot technology, connect to multiple forms of dirty and uncorrelated data sources and are, by nature, designed to be multi-modal. Allowing them more flexibility to answer a wider variety of questions and perform complex, multi-system actions based on natural language queries.
What criteria should you use to decide between the two?
Here are a few basic criteria that can be used to determine the right fit for your need:
- Answer structured sales questions about your products on your website
- Raise, update and close a service tickets
- Log, alert and approve expenses
- Initiate an alert, reminder or task
- Check calendars and book meetings with your team
- Correlate between multiple points of data to provide analysis
- Prescriptive and predictive actions across application stacks
- Read, scan, file and approve from unstructured documents (images, pdfs, etc.)
- Switch between multiple contexts and subject matter
- Take action by controlling, authorizing and executing on other systems, platforms and tools