by Jason McCullagh, Enterprise Search Evangelist, Upland BA Insight
I’m showing my age somewhat, but when I was at school, Google search was only just getting going and these types of paradoxical ideas like ChatGPT and generative AI were akin to something that you only see in a movie, but difficult to believe could happen—like the hoverboard concept in Back to the Future II. The idea that I could get a computer to author an essay and do my homework seemed outside the realm of possibilities! What would Doc Brown say?
What exactly are generative AI and ChatGPT?
I won’t spend endless paragraphs going into what ChatGPT is, as there has already been a monolithic amount written on the topic. But in short, it is an online query-answer and text-generating system (a chatbot) that delivers human-like responses in a conversational manner, amongst other things. It is built on large language models (LLM) and GPT (Generative Pre-training Transformer). The concept was developed by OpenAI. The idea is that it learns from feedback; information is fed into the models to deliver these human speech-like text responses. All of this has been made possible by dramatic advances in machine learning.
ChatGPT can be umbrellaed under generative AI systems whereby it can be used for content generation. It can also be applied to a range of Natural Language Processing (NLP) applications, such as:
- Dialogue Generation
- Language Translation
- Text Summarization
- Text Classification
- Question Answering
- Text Completion
These are just a few examples of the practical uses of generative AI, and we are only just beginning to understand it. Generative AI is not just limited to text, as it can now be applied to images and languages, which is made possible through DALL-E —another flavour of the OpenAI phenomenon.
ChatGPT is not the only player in this market. It feels like there is a new AI tool or platform being released every week; it is hard to keep up. Many larger companies have released similar technologies and capabilities such as Bard AI (Google), Bedrock (Amazon), and Microsoft’s Bing AI, Azure OpenAI Service, and recently launched Copilot.
Does the rise of generative AI mean the end for enterprise search?
So, are these AI technologies reality that are here to stay, or just another “10-minute wonder” with marketing hype that has sent the internet into overdrive? If it is just marketing, then OpenAI have done something correct as over 100 million users signed up to use ChatGPT in under two months – YIKES!
Being honest, I love the buzz, excitement, and energy in the air when there is a shiny new technology to learn, investigate, and utilise. But (and this is a big but), the flip side is that there are still many unknowns around AI and whether they can truly add value in the business world rather than some fun in a keyboard playground.
People are asking whether ChatGPT and AI powered search spells the end for enterprise search. That’s a fair question, but I am of the strong belief that ChatGPT and other similar technologies are not replacement search engines, at least not anytime soon. They are facilitators and enhancers (summarisations, concise/specific answers, and text generation), but not replacements.
The Implications of Generative AI and Enterprise Search
Answers versus documents
Typically, when users search the web, they are looking for answers to specific questions such as, “where is my local supermarket?” or “What is ChatGPT?” But when it comes to organisational content, search is very different.
Enterprise search is normally used to find unstructured information stored within documents, wikis, blogs, and intranets that are spread across a wide variety of business applications. Traditionally, it is about making this content findable by indexing the body of the content or the metadata.
ChatGPT is not designed to point you at something that already exists such as a document or documents. It is more of a content generation/creation function. ChatGPT can create a document about X or an essay about Y, or it answers questions such as “Who is Marty McFly?” or “What is a flux capacitor?” Whereas AI powered enterprise search searches and returns results.
And how does item-level security fit into all of this? Documents and organisational content often live under tight security policies. It would be more than frowned upon if users could use a ChatGPT-like experience to ask, “How much does the Head of A or the Director of B earn?”
ChatGPT and enterprise search software deliver different experiences, but this is not to say that the two cannot live in the same space—both can help deliver organisational information.
Scalability and cost
It is likely that an AI powered search engine would involve the creation of connected LLMs. At a high level, it would involve generating a lot of text and storing it as a vector file that has a massive amount of dimensions for every sentence and paragraph in the document. These would be hit every time a search query is run. Therefore, performance could be a challenge.
All companies must ensure their business is growing and progressing towards more profitability. The market will see a lot of unpredictability around these new services costs as they test new models and trial different methods.
OpenAI are continually releasing APIs to perform various tasks like providing deeper control of the models. There is still nothing official about business plans for ChatGPT. OpenAI are actively exploring options and data packs. Given the scalability and cost questions, one might wonder whether organisations will take on the creation and training of LLMs themselves, or will they utilise products and solutions enabled by technology vendors?
Training the models
ChatGPT and others are great when asking them to author an essay or write a song. This uses the LLMs and massive datasets that are primarily based on content from the internet, but they are less helpful when organisations want to use AI-powered search technologies and chatbots for business purposes. This is because the results need to be organisational-specific, which means companies need to train the LLMs based on their data and content. AI–powered search is only as good as the model and the data. Capturing the data is key because this enables AI to make sense of the data more rapidly and of course, as each day passes, there are more ways and tools to help ingest data into the models.
There are questions about information validity. In truth, it can be a little open-ended. AI chatbots, such as ChatGPT and Google’s Bard, have been known to deliver inaccurate misinformation or nonsensical text, referred to as hallucinations. This creates an erosion of trust and once lost, it is hard to reverse. Of course, you can throw more training at the LLMs as mentioned above, but is it that simple? In time, more transparency and explanation about how the AI models work may be available, but it is not today. This said, ChatGPT have recognised that hallucinations are an issue so an enhancement around justifying results/providing context has recently been introduced.
Other ChatGPT hurdles are starting to become apparent, such as including intellectual property (who owns the data after upload). OpenAI say that everything is anonymised but is anything really anonymised when it is in the cloud? And what about plagiarism? This goes back to the opening paragraph about a computer authoring my essay—is it ethical?
We are beginning to see lawsuits around whether technology platforms should be protected from legal responsibility for content posted online by their users. Companies and individuals are threatening defamation lawsuits. So, there are questions around the use of these technologies, who regulates the data, and the validity and/or accuracy of the data. As a result, we are seeing some organisations already blocking access to ChatGPT and other similar websites as they do not want employees adding sensitive or personal information to the LLMs.
Obviously, everything comes down to data. Who owns it and where does it reside?
As generative AI technology advances, people’s opinions and the way that we use it will undoubtedly swing this way and that, but there has been huge investment from many large businesses, and many people are already onboard. So, it does feel like these technologies are here to stay. ChatGPT and other AI powered search systems are very impressive and add real value, however I do not think that enterprise search will be consumed by the ChatGPT AI powered search revolution just yet.
In the end, ChatGPT is just one approach to AI and LLMs. Upland BA Insight currently integrates with a host of machine learning, natural language processing, and generative AI systems. Technology has evolved quickly to go beyond Doc Brown’s 88 mph and into hyper speed. As a result, we are constantly adding new features and functionality to our platforms, taking different approaches, and reviewing the market to see how we can best incorporate ML, NLP, and AI powered search systems but only when it makes sense and adds value.
So, unfortunately, if Marty and Doc did arrive in the not-to-distant future, Marty probably wouldn’t see people whizzing around on hoverboards, but Doc could ask ChatGPT, “Is it really possible to build a time machine out of a DeLorean?” However, it remains to be seen what type of response he would get.
Ready to get started with AI-powered search? Book a strategic session with one of our AI experts below.
About Jason McCullagh
Jason is a highly skilled enterprise search specialist with nearly two decades of experience. He is passionate about helping businesses worldwide harness the power of enterprise search to improve efficiency and drive business transformation. Jason’s personalized approach allows him to identify areas for improvement, including utilizing analytics to enhance user engagement, streamlining and centralizing search by connecting systems, implementing meta-tagging for improved search capabilities, and deploying productivity-enhancing applications to increase productivity and drive overall success for organizations of all sizes. Connect on LinkedIn
Want to learn more about AI enterprise search? BA Insight supports five search engines (Azure Search, Elasticsearch, Microsoft 365, Microsoft Search, and Solr); four AI platforms (Rasa, Microsoft Cognitive Services, and Amazon AI); and it can run in the environment of your choice (on-premise, AWS, Azure Cognitive Search, etc). Read on for more.