Bern Elliot at Gartner presents a comprehensive analysis of the usage, adoption, limitations and security challenges around the integration of ChatGPT inside the enterprise.
While there is a lot of excitement about ChatGPT, as well as conversational artificial intelligence in general, many questions remain about what generative AI is and what it can do for people as well as enterprises.
ChatGPT and other foundation models like it is one of many hyperautomation and AI innovations. It will form a part of architected solutions that automate, augment humans or machines, and autonomously execute business and IT processes. It will also likely be used to replace, recalibrate and redefine some of the activities and tasks in various jobs.
ChatGPT will emerge from its beta phase into an early trial and pilot phase. During that time, adoption is expected to increase, best practices for use will mature, and increased adoption into business workflows and applications is expected. However, it is also possible there will be a negative response to a range of issues, including privacy concerns, misuse of information and bias.
This is common as a technology moves from the peak of inflated expectations to the trough of disillusionment.
Capability
ChatGPT is capable of:
- Generating and helping to improve prose and code development.
- Summarising text.
- Classifying content.
- Answering questions.
- Translating and converting language, including programming languages.
There are four main ways to deploy the ChatGPT technology:
As-is
Input text prompts and receive results via the web-based interface. This is by far the most popular approach when starting out.
Prompt engineering without APIs
Prompt engineering refers to the use of a service like ChatGPT in conjunction with other technologies, as part of a workflow. You can create this workflow manually or by using screen scrape and robotic process automation, RPA technologies.
Prompt engineering using APIs
This approach allows you to set and evaluate prompts programmatically and directly integrate ChatGPT with a broad range of applications.
Interaction with model
It is possible to leverage your own version of GPT2, GPT3 or other foundations model for a bespoke implementation. However, you would not be using the customised version of GPT3 or GPT4, which users cannot change.
Usage
Proceed but do not over-pivot
Recognise that this is early stage and much of what you are hearing is hype. That said, the potential is significant. Explore other emerging generative AI use cases. Go beyond GPT language-focused ones.
Encourage careful experimentation
Encourage out-of-the-box thinking about work processes, but not before you define usage guidelines, ensure understanding of the risks, issues and best practices, and have all generated text reviewed by humans.
Create a task force reporting to the CIO and CEO
Explore existential threats and posed and major opportunities, plan a roadmap for discovery, and scope the skills, services and investments needed.
Limitations
It is only trained on data through September 2021, so it has limited knowledge of events that have occurred since then. It cannot cite its sources, and it is only as dependable as these sources, which may be wrong and inconsistent, either in themselves or in how they are combined by ChatGPT. It cannot yet accept image input or generate images, though in the future, it could be used in combination with visual generative AI models.
You cannot train ChatGPT on your own knowledge bases. Although it gives the illusion of performing complex tasks, it has no knowledge of the underlying concepts; it simply makes predictions. Its data privacy assurances have not yet been subject to rigorous audit. Despite some recent improvements, it cannot be relied on to do math.
Security
Maintain caution when using ChatGPT. While OpenAI and Microsoft, the companies behind the product, have stated that all information shared is confidential and private, they have not yet clarified details of their data usage in certain areas, such as what they do with context-sensitive prompt information.
Until there is further clarity, enterprises should instruct all employees who use ChatGPT to treat the information they share as if they were posting it on a public site or social platform.
Microsoft, which has a lot of experience with these enterprise issues, has been clearer and more initiative-taking in creating security, confidentiality and privacy policies related to ChatGPT than OpenAI has.
With all of this in mind, it is best to create a company policy around rather than block ChatGPT. Your knowledge workers are already using it, and an outright ban may lead to shadow ChatGPT usage, while only providing the organisation with a false sense of compliance.
A sensible approach is to monitor usage and encourage innovation but ensure that the technology is only used to augment internal work and with properly qualified data, rather than in an unfiltered way with customers and partners.
Future impact
There will be new jobs created, while others will be redefined. The net change in the size of the workforce will vary dramatically depending on the industry, location, enterprise size and offerings.
However, it is clear that the use of tools such as ChatGPT, hyperautomation and other AI innovations will focus on tasks that are repetitive and high-volume, with an emphasis on efficiency, increasing productivity and improving quality control.
ChatGPT will also be integrated into business applications. This will make adoption easier, and relevant contextual information will be available in the applications.
Key takeaways
- ChatGPT is one of many hyper automation and AI innovations.
- It will automate, augment humans or machines, and autonomously execute business and IT processes.
- It will be used to replace, recalibrate and redefine some of the activities and tasks in various jobs.
- Prompt engineering refers to the use of ChatGPT in conjunction with other technologies, as part of a workflow.
- ChatGPT and hyperautomation will focus on tasks that are repetitive and high-volume, with an emphasis on increasing productivity and quality.
- It is only trained on data through September 2021, so it has limited knowledge of events that have occurred since then.
- It cannot cite its sources, and it is only as dependable as these sources.
- It cannot yet accept image input or generate images, though in the future, it could be used in combination with visual generative AI models.
- You cannot train ChatGPT on your own knowledge bases.
- It has no knowledge of the underlying concepts and simply makes predictions.
- Despite some recent improvements, it cannot be relied on to do mathematics.
