The value and limitations of ChatGPT for businesses

This is a guest blogpost by Paul Maker, Chief Technology Officer, Aiimi

For years, AI capabilities have been used across many sectors, but often the promises and potential of the technology has not been realised, leading to reduced excitement about its potential and a lack of understanding for many about what AI is. For many, especially when it comes to business, AI has been a nebulous form of magic that has had negligible impact on their daily working lives.

The arrival of ChatGPT has changed this and reignited the conversation on the benefits that artificial intelligence and machine learning can bring at a time when IT infrastructure hasn’t changed significantly for decades. Take search engines, for example; for years, they have been the search box in the middle of the screen, presenting the user with hundreds of relevant articles to scroll through. ChatGPT has turned  the traditional approach to information search and discovery on its head. Now, rather than providing a mass of answers, ChatGPT’s search function provides a distillation of all relevant information pertinent to the search, and within parameters defined by the user.

This is of course useful in several ways, but it’s important that we remain grounded when it comes to the limitations and potential of the model.

Business use-cases for ChatGPT

Further to its capabilities around surfacing information, ChatGPT can also be used for more technical purposes. Developers, for example, can ask the model to write code —although depending on the instructions provided, results for this advanced use case vary wildly, from the creation of simple and effective applications to totally unusable rubbish.

Regardless, businesses have bought into the renewed excitement, fuelled by the apparent usability of ChatGPT, as high-level use cases, such as the creation of marketing materials and other content, seem to deliver on the promise of AI—which is the automation of mundane and time-consuming tasks. This is where the majority of firms can safely achieve quick wins with tools like ChatGPT. Summarising large internal documents or corporate communications, writing emails or simplifying complex technical material for non-technical staff are all possible with generative AI. As long as sensitive data is not uploaded to open source models, there is a wealth of content-related use cases.  These use cases are a boon for smaller firms, which have been gradually adopting open source AI-tools to deliver advanced capabilities, such as those around clustering, classification and search. Thanks to the efforts of the open source community, such capabilities are no longer the preserve of larger corporations, which have vast amounts of data on which models can be trained.

More widely, generative AI has shown its potential to increase productivity, reduce operational costs and create growth opportunities. Companies that can leverage the technology effectively are likely to gain a competitive advantage, but this will require technical and subject matter experts to work together and validate use cases. Ultimately, there is no substitute for real-world expertise and creativity when it comes to innovation, which means that we must continue to draw upon human ingenuity when developing ground-breaking solutions. When it comes to ensuring that applications of the technology are appropriate and do not leave a business exposed to security and compliance risks, subject matter expertise will be crucial. This is where I foresee a growth in opportunities for those who can clearly identify the risks and rewards that come with the adoption of generative AI.

The current limitations of ChatGPT

ChatGPT has shown the power of large language models and generative AI, but there are a number of problems inherent with the model as it stands. One key limitation of ChatGPT is that it is a giant ‘black box’ that has been trained only on public data, which is full of biased data, questionable content, and that which is factually inaccurate. This has resulted in the model being unable, in certain instances, to distinguish between opinions stated with authority and objective facts, leading to ‘AI hallucinations’ that cause the model to write “plausible-sounding but incorrect or nonsensical answers.

This presents a clear problem for organisations that want to harness the capabilities of large language models that have been trained on open source information. Further to this, if a business or employee intends to harness open source generative AI tools to create content or capabilities that draw on owned information assets, they will need to surrender total control over this information, which potentially exposes the organisation to data compliance and security risks.

To make effective models, a great deal of time and expense must be put into data engineering, ensuring data is clean, secure and usable, which is what has historically put such technology out of reach for smaller businesses. Data models are only as proficient as the data that’s put into them, which rules ChatGPT and other generative AI tools out for many use cases that would require specific datasets and parameters to be set.

I have no doubt that generative AI will one day be harnessed effectively across industries, enhancing human productivity and experiences. If the technology can produce insights at lightning-fast speed, millions of times, and organisations are confident their models are delivering accurate and fair results, the corporate benefits will be unparalleled. But we must be conscious that there is still some way to go before this is possible for most.

The future of ChatGPT

ChatGPT itself may go down the route of becoming a chatbot or the de facto modern search engine, but its arrival has encouraged others to investigate similar AI-powered tools that can deliver business value, reinvigorating the whole space and allowing other technology vendors to ride the new wave of excitement. Microsoft has already announced that Teams Premium will integrate ChatGPT, enabling advanced automation capabilities such as emailing attendees with a 50-word summary, key points and next steps discussed in the call. This will mean a Teams meeting is superior to using another video call function, or in-person meetings—delivering an experience akin to the metaverse, mixing virtual and in-person activity, and enabling humans to interact in a more meaningful way with data.

In this respect, ChatGPT has been a watershed for our collective expectations of AI, and more specifically generative AI. Now, it’s not only advanced automation that excites the imagination, it’s also the advancement of user experience in ways we might never have imagined.

Data Center
Data Management