While it seems to be obvious for all that AI chatbots like OpenAI’s ChatGPT and Google’s Bard will create countless new ways to gain answers and generate content for your business, there are some limitations you must know, but first:
What is a chatbot?
While it is a much larger philosophical question whether chatbots are alive or not, chatbots like ChatGPT and Bard are what’s called large language models. These large language models, or LLM’s, take in large swaths of information (such as the entire internet) and use a combination of machine learning algorithms and something called natural language processing, or NLP’s, to generate answers to whatever you type into them. These two techniques work in part to create a “neural network” that functions much like an actual brain. Neural networks can understand and interpret the meaning of text, process that data in a way very similar to the way your brain is taking in this information right now. They then respond to the questions, or prompts, they are given with cogent and useful responses. What’s even more crazy is that we don’t even know how these neural networks function, this is where those philosophical questions on whether AI is alive or not come into play, but that’s a topic best left to the intellectuals for now.
What you need to know is that chatbots use logic and inference to answer your questions based on the data it has on hand and that can lead to some problems that you need to know about:
- Chatbots only aggregate from existing content it has been trained on.
This means that chatbots formulate answers and context based on data it can access. For example, while chatbots are great at parsing code and diagnosing why your python script doesn’t work, it cannot use that same logic to provide accurate answers in specialized fields it has not been trained on like advanced medical procedures or the most current theories of quantum physics. Chatbots like Bard and Chat GPT also are limited to public information which means you cannot access private databases like university libraries or scholarly journals.
- Chatbots are only as good as the questions you ask them.
Using different variations of text to ask a question in different ways can lead to much different responses, and thus not give you the proper information right away. You may assume the chatbot knows more than it does, or perhaps the chatbot could do more if you gave it more information. For example, if you asked a chatbot to write a script in python that makes a list of all the hotels within 5 miles of Disneyland, it will spit out code that will take a long and circuitous route of obtaining that information with mixed results, however, if you also provided the API to a site that indexes hotels (like Yelp), it will use that API to create a easy-to-run script that will give you and concise and usable set of data. In the very-near future, every organization will have a “prompt linguist” on staff that can utilize an LLM more efficiently. This role will soon become one of the most important roles within an organization. Click here for our article on what your business needs to know about prompt linguists and how it will be the most important hire you make this decade.
- Chatbots can use copyrighted information.
Information gathered by chatbots (and other neural network-based generative models, like OpenAI’s Dall-E image generator or text-to-speech (TTS) models like Tacotron) which may use copyrighted material that can open you or your business up to lawsuits from the original IP holder. This means that if you publish or sell anything prompted by a chatbot, you could be infringing upon someone’s trademarks, copyrights, or just downright plagiarizing and being sued. Using a chatbot does not mean you own the information it spits out, you should obtain the proper licensing and permission to use another entity’s content.
- Chatbots only carry data up to a certain date (and it’s not today).
Chatbots only gather data up to a certain date, asking questions about current subjects, events or products may lead you to a brick wall. As of publishing this article, ChatGPT and Google Bard have only been trained on information up to September of 2021. If you are looking to find information on the newest updates to coding languages or the stats from last night’s game, you are going to have to rely on the good old tried-and-true Google search (or Bing if you happen to work for Microsoft).
Although chatbots are an absolute gamechanger, they cannot do everything. Minding these limitations will keep your prompts more accurate and help you become more aware of how these wonders of technology actually work.