ChatGPT is being hailed as a potential game-changer in the world of AI. The tool has impressed experts with its writing ability, proficiency at complex tasks and ease of use. It has generated much excitement and opinions everywhere, on the media and the internet. Some have suggested it could signal a future in which AI has dominion over human content while some believe it might be the end of Conversational AI.
In the heatwave of ChatGPT, Google on 7 February 2023 finally launched its AI technology, Bard with the intention of competing with Open AI’s popular language model ChatGPT-3. Google CEO Sundar Pichai has described Bard as a Conversational AI service that can do things like simplify difficult things apart from providing high-quality responses.
What has ChatGPT done to herald such attention and competition? How might it (and its future iterations) become indispensable in our daily lives? And how would it affect businesses? In this article, Joshua Kaiser – CEO of Tovie AI, an experienced Conversational AI technologies vendor for the Enterprise and Alec Bore – Associate Partner, AI & Automation, Europe of Infosys Consulting, one of the original investors in Open AI, explore the gains and potential losses and what ChatGPT could mean for businesses in pursuing the quality of personalisation using AI technology.
ChatGPT is a prototype dialogue-based AI chatbot capable of understanding natural human language and generating impressively detailed human-like written text. It is the latest evolution of the GPT – or Generative Pre-Trained Transformer – a family of text-generating AIs.
ChatGPT builds on OpenAI’s previous text generator, GPT-3, which is reported to have received a $10 billion investment from Microsoft. The tech giant previously gave OpenAI $1 billion in 2019, as part of a deal to help build Microsoft Azure’s AI supercomputing platform. OpenAI would then exclusively use Azure to run its services.
OpenAI builds its text-generating models by using machine-learning algorithms to process vast amounts of text data, including books, news articles, Wikipedia pages and millions of websites. By ingesting such large volumes of data, the models learn the complex patterns and structure of language. They have acquired the ability to interpret the desired outcome of a user’s request.
ChatGPT builds a sophisticated and abstract representation of the knowledge in the training data, which it draws on to produce outputs. This is why it writes relevant content, and does not spout grammatically correct nonsense.
ChatGPT – the gains and downsides
ChatGPT is an efficient language model with the ability to understand natural language input and use context to generate meaningful answers, without the need to be pre-programmed by bot developers. The model is composed of trillions of parameters and requires supercomputer infrastructure to compose its output and train. Therefore, training and hosting such a model poses a significant financial investment outside of the capabilities of most organisations.
When using ChatGPT’s natural language generation capability it is important to recognise its unpredictability when considering compliance and regulations across different sectors, so reliance on natural language processing generation without important foolproof checkpoints, utilising the application of deeper human cognitive input is vital, otherwise, they run the risk of reputational damage, together with the associated financial and legal penalties. It is crucial to get this right. ChatGPT needs to be released with the appropriate tools to test, tweak, update and constrain the output in various ways. Therefore it is not an off-the-shelf, plug-and-play piece of code. It may help leapfrog and automate some of the heavy lifting involved in the code development process before the final voice and text-based customer service t applications and implementations are signed off.
Although ChatGPT currently understands its function as well as the data and context in which it operates there are gaps in terms of predictability and personalisation. For example, if you are the NHS and you have a bot doing symptom triage, you want to predict how you might respond should a user say they might want to commit suicide. Bots that are narrowly trained on a smaller domain and have predictable answering patterns would be better at dealing with this as it’s something you would like to pre-empt as part of your design. Open-ended NLG models like ChatGPT pose a risk here because their output is not entirely predictable and is a function of what it has been trained on.
ChatGPT has proven the high interest and demand for personalised experiences in the market, as well as the importance for businesses to obtain personalisation through AI-powered chatbot development.
AI-powered chatbot development is a combination of technologies where NLG is a key piece of the puzzle along with natural language understanding, named entity extraction, context extrapolation and low code finite state machines. In today’s market, ChatGPT in particular is highly useful in supporting existing chatbot platforms with small talks or generating training phrases that can be used to train more narrowly focused NLP models like BERT.
Investing in using chatbot technology for automation is different from investing in the research and training of new cutting-edge models. Organisations that are contemplating using chatbot technologies for automation should consider investing in platforms that will allow them to grow with the market and adopt new models like ChatGPT when they are made available.
Since ChatGPT’s official release, Tovie AI has been exploring the different ways the tech phenomenon can enhance its products and solutions: making its banking and insurance offering easier to scale to global markets, adding more innovative features to its telco solution and powering its tools for building conversational interfaces with new capabilities.
At Tovie AI, NLP abstraction is a crucial facet of our platform. Tovie NLP allows organisations to host their own proprietary models along with our own internal ones and even open-source models from the likes of HugginFace for example. This means clients can adopt the right model for their use case and continue to leverage the workflows they have built in our low-code bot configuration engine with the latest NLP models when something new hits the market.
These advancements in NLP will only drive further improvements in the effectiveness, scope and scale that Conversational AI can support in driving value in the Enterprise. Whilst this progress will primarily increase performance and domain coverage, the platform is the key factor in ensuring sustainable success, scaling within the enterprise and as a result, driving operational efficiencies in delivery.
Conversational AI is automating the last frontier of human interaction, namely human conversation. Humans are expensive and have limited capacity, whereas software can scale service concurrency into millions of users. As for the longer term, there will be new value-creation opportunities where chatbots do not just replace a job function but create an entirely new one. One such idea could be a new market for digital friends for people who struggle to make meaningful human relationships in their daily life.