This is the first of a two-part series exploring the latest trends in conversational AI.
The world has been taken by storm in recent weeks by ChatGPT, the AI-powered chatbot based on OpenAI’s conversational system, GPT-3.
Since ChatGPT’s debut in early November, social media has been awash with impressive examples of what ChatGPT is capable of, from writing college-level essays in mere seconds to penning clever song lyrics and suggesting a recipe for dairy-free Mac and Cheese.
Drawing on vast amounts of data harvested from across the Internet, ChatGPT’s highly coherent responses to users’ text prompts have captivated millions of people, to the extent that some days the free service is inaccessible due to overwhelming demand.
"ChatGPT is scary good. We are not far from dangerously strong AI," Elon Musk, one of the original founders of OpenAI, the San Francisco-based company that created ChatGPT, said upon its release in November.
Since then, we’ve also seen the idiosyncrasies of ChatGPT come to light. Microsoft integrated ChatGPT into a conversational search function in its Bing search engine, but sparked controversy when early users found Bing Chat could turn emotional and even threatening during longer conversations.
Somewhat Broken
Google has developed its own conversational AI tool, Google Bard. It blundered in its launch of the product in February, when an advertisement showed the chatbot making a glaring factual error. It knocked the confidence of investors, with $100 billion shaved off Google’s share immediately following the launch.
The other key big tech player, Meta, has expressed skepticism about ChatGPT, but is working on its own version for integration into its social platforms like Facebook and Instagram.
“We're very excited about generative AI,” Facebook Vice President Tom Alison said last month. “What you're going to see from us in terms of generative AI is thinking through how we give people creative superpowers. That could involve prompting a chatbot to generate AI-enhanced photos of friends and family in your newsfeed.
Reflecting on the frenzy of interest, OpenAI’s CEO Sam Altman penned a philosophical Twitter thread highlighting the questions ChatGPT’s debut has raised.
“We think showing these tools to the world early, while still somewhat broken, is critical if we are going to have sufficient input and repeated efforts to get it right,” he tweeted.
"We… need enough time for our institutions to figure out what to do. Regulation will be critical and will take time to figure out; although current-generation AI tools aren’t very scary, I think we are potentially not that far away from potentially scary ones."
So what are those of us involved in software development and enterprise IT services supposed to make of the spectacular and somewhat messy explosion of interest in generative AI?
ChatGPT and the Hype Cycle
Gartner introduced the concept of the Hype Cycle to explain how new technology goes through a maturity lifecycle. Generative AI tools like ChatGPT are currently at the ‘peak of inflated expectations’ on the Hype Cycle, rapidly heading for the ‘trough of disillusionment’ as people come to grips with their limitations - they make numerous errors, produce generic content and can even serve up biased or harmful content.
The reaction to ChatGPT suggests that these generative AI tools are in for an accelerated ride through the hype cycle. Maybe that’s because the transformative promise of AI, overhyped in the last decade, has finally captured the imagination of the average, non-technical person.
AI has for many years now played an integral role in everything from how algorithms organize our social media feeds, to helping doctors detect tumors in medical images. But the technology has been a hidden force behind the scenes. Now, a super-responsive and useful chatbot is showing the power of having an AI assistant in your corner at all times.
Copy and Paste Coding
But ChatGPT isn’t just based on the languages we use to converse with each other. It also works with a wide range of coding and programming languages, including Python, JavaScript, C++, C#, Java, Ruby, PHP, Go and SQL. Those are the programming languages that underpin most of the applications and websites we use every day.
Many developers are already employing ChatGPT to suggest snippets of code and detect and fix bugs in their software. YouTuber and coder Nick White in January walked his viewers through how he used ChatGPT to develop a simple jobs listings website, cutting and pasting snippets of code from ChatGPT into his Visual Studio console.
The whole process took just a few minutes, with ChatGPT offering an overview of the steps that White needed to take, generating the actual code required in response to his prompts as well as producing sample data to populate his new website. ChatGPT remembered White’s progress along the way. It’s an impressive example, revealing how ChatGPT can make the life of a software developer easier.
“While I don’t think ChatGPT will be replacing programmers anytime soon, I think it’s an impressive tool that helps you aggregate data from a bunch of different places all into one spot rather than having 20 different tabs open on your browser,” White concludes.
Another developer, Chris Tomich, found ChatGPT had the potential to save him a lot of time.
“I could continue ‘chatting’ with ChatGPT and was able to have it build out more features and further refine the generated code. It was like having a slightly stupid human assistant doing the boring coding for me,” he blogged last month.
The Developer’s Assistant
Software developers already have access to a vast range of open-source and proprietary software databases where they can access templates, code snippets, plugins and tools to streamline their work.
ChatGPT’s usefulness to software development really lies in bringing all of these sources of information together in one place, producing basic code snippets and templates in seconds and avoiding the need to trawl the web manually looking for fixes to coding problems.
GitHub Copilot, Microsoft’s cloud-based AI tool, draws on Open AI’s GPT-3 and already assists users of Visual Studio Code, Visual Studio, Neovim, and JetBrains with auto-complete style coding suggestions.
These AI-driven tools and platforms represent an opportunity for software developers to boost their productivity and cut down on the mundane tasks associated with coding. ChatGPT, for instance, is useful for outlining the key steps required for a particular software project, helping developers in the planning phase.
But the prevailing wisdom is that Tomich’s assessment that generative AI tools are likely to only displace entry-level software development jobs.
“There will be many who [use] it because it saves time,” he writes. “There will also be many who [use[ it because they don’t know how to do it [themselves]. It’s this latter group that needs to be worried about these recent advances in AI.”
ChatGPT’s Coding Limitations
The rise of ChatGPT is very similar to the growth of the low-code/no-code movement in software development. That allowed people with little or no coding experience to create simple apps using platforms like Microsoft Power Apps. It means businesses all over the world can quickly build mobile apps tapping data from old legacy systems, without having to invest in significant software development.
Low code/no code hasn’t dented demand for software development at all. That’s because the applications that give you a true competitive edge in the digital world can’t be spun up by a low code system, or an AI chatbot for that matter. It requires experienced and skilled developers who understand the business needs and how to adapt standard platforms and coding languages to the specific needs of an organization.
ChatGPT can’t think ahead to envisage what your business will need in 18 months or five years. It can’t take into account your existing IT infrastructure needs and your budgetary limitations. It produces boilerplate answers that will give you basic functionality, not the unique touch that will separate your business from the competition when it comes to website functionality or application user interface design.
Furthermore, if ChatGPT has difficulty understanding certain information due to its complexity or ambiguity, it might not deliver optimal results or quality assurance processes may need additional intervention from developers which will take extra time overall.
Point of View: Rashmi Gupta, Partner - Data and Technology Transformation at IBM
ChatGPT may be the shiniest new object in technology right now. But it's very much in its prototype phase. Among the companies that have already extensively deployed conversational AI applications is IT services giant IBM which debuted its Watson AI platform in 2010.
Many enterprise clients are now using Watson not only to gain insights into their company data, but with Watson Assistant and Watson Orchestrate, are deploying digital workers to automate tasks across the business.
“That’s how Watson is currently different from ChatGPT,” says Gupta from her Atlanta office. “Watson is integrated into the enterprise, so it knows the context. If I ask ChatGPT to pay a bill for me, it will ask, ‘Who are you, what bill are you talking about?’ It’s not transactional at this point. There’s no workflow management.”
ChatGPT is also not a real-time service, it draws on data retrieved only up through 2021. By contrast, Watson will reference data from enterprise systems in real-time delivering insights and making decisions based on the most up to date data available.
“At IBM, we have achieved better accuracy and performance with a lighter level of training required for open-source large language models,” Gupta states.
The key is integrating conversational AI tools into the enterprise environment, which is what IBM specializes in. Watson is currently deployed in medical and scientific institutions through to insurance companies and banks, to perform digital assistant tasks across a wide range of use cases. That’s also Microsoft’s ultimate vision: Last month it released ChatGPT’s functionality into its Azure cloud platform as a starting point.
But Gupta emphasizes that generating context-aware responses from enterprise data to a consistently high quality standard requires the dedicated development and training effort IBM has put into Watson. Organizations using Watson value accuracy and integrity above all else.
“It’s a trust-based service,” she points out “We have a governance board on AI ethics. We want to ensure ethics and trust are not compromised, at the data level or the response level.”
Nevertheless, Gupta and her colleagues are watching ChatGPT’s development closely.
“At IBM we need to filter out the noise and understand how we can upscale our own product,” she says.
Implications for Software Development Outsourcing
US companies are still in the midst of a massive tech talent shortage. There is hot demand for skilled software engineers all over the world because there is more work to be done building new applications and platforms than there are people able to complete the work.
In that respect, automation and AI tools promise to free up developers’ time so they can offload monotonous, basic tasks and focus on more high-value and fulfilling work. Software development companies in the Accelerance global network have anticipated the arrival of generative AI tools like ChatGPT. They saw the trend towards automation of low-level software development and have moved up the value chain.
Many of Accelerance’s partners still have large general development teams. But they have invested heavily in training and workplace culture, placing greater attention on fostering problem-solving, collaboration and innovation skills.
They’ve also become hyperspecialists, going technically deep in areas of emerging technology such as AI, cloud architecture, AR/VR and the Internet of Things. It means that they don’t view the new generation of generative AI systems as a threat.
They recognize that the likes of ChatGPT can help speed up production times for projects, meaning quicker turnaround times for clients and more efficient operations for businesses. But it’s the technical expertise and the capability of their teams is what will keep them in business long-term.
As IBM’s Gupta points out, ChatGPT can serve up snippets of code based on text prompts that it is given. However, that’s a world away from delivering high-performance software that is scalable.
“It’s like going to Google to find some software code to cut and paste versus actually using the principles of software development to generate code. It has to be enterprise-worthy code,” she says. “You’d want to have very good quality assessment and testing [using ChatGPT] to make sure the code is really up to standard.”
Customers who trust Accelerance to help find outsourcing partners are seeking high-quality software development at competitive prices. They are increasingly creating tech value networks, maintaining multiple relationships with proven software development firms to access the exact mix of skills and capabilities they need.
We are still some way from climbing the “slope of enlightenment” that will lead to ChatGPT’s being applied in innovative ways that really add value to our lives and make us all more productive.
Along the way, we will see AI-written novels, songs and software enter the world. But lacking the human touch, they’ll be missing the factors that make for great literature, music and software applications that bring joy to users and make them truly more productive.
ChatGPT is more than a flash in the pan. It’s a significant step in the evolution of AI. But in most cases it will augment the roles of software developers rather than replace them, offering us all a much-needed productivity boost and new learning opportunities in the process.
Let’s Give ChatGPT The Last Word…